The Data Centre ClockEngineering the Irish Retrofit

The Method of Procedure was eleven pages long and had been reviewed by three people

before it was printed. It sat on the desk in the control room with a pen clipped to the

front page, and every step that had been completed was initialled by two people, and

every step that was yet to be completed was not initialled by anyone.

This was not bureaucracy. This was the data centre equivalent of measuring twice before

cutting — except in a data centre, the thing you were cutting was live, and if you cut

wrong, three hundred and sixty-seven tenants lost the ability to process transactions

until you cut it back.

Declan was reading page four. He had read it twice already this morning and once the

night before. He knew what it said. He was reading it again because that was what you

did with a Method of Procedure before you executed it. You read it until you were bored

of reading it, and then you read it one more time.

Ann arrived at half eight. She had her notebook and a question she hadn't asked yet.

"What are we testing today?" she said.

Declan looked up. "UPS transfer. We're going to trip Module C and see what happens."

"And what should happen?"

"The load transfers to Modules A and B. The servers keep running. Nobody notices."

She wrote that down. "And if that's not what happens?"

Declan picked up his coffee. "Then people notice."

Mark was in the UPS room when they got there. He had arrived before either of them,

which Ann was beginning to understand was simply how Mark operated — he was always

already there, already looking at something, already writing in his clipboard.

The UPS room at Clonshaugh held three modular UPS units, designated A, B, and C,

each rated at 800 kVA. Total installed capacity: 2,400 kVA. Design load: 1,600 kVA —

which was approximately the facility load at the current PUE of 1.50 divided by the

power factor. The design intent was N+1: any one module could fail, and the remaining

two would carry the load without interruption.

That was the theory. Today was the test.

Mark had the Method of Procedure under his arm and was looking at the bus connections

between Module C and the main distribution board. He looked up when they came in.

"Who signed off the scope?" he said.

Declan said, "What do you mean?"

"The MoP scope. It says single-module trip, load transfer observed, module restored."

He held up the document. "That's N+1 verification. That's not full redundancy testing."

Declan set his coffee down. "It's what we test every year."

Mark nodded slowly, in the way that meant he was acknowledging a fact, not agreeing

with a conclusion. "I know," he said. "I'm just flagging the scope."

The test itself was unremarkable, in the way that successful engineering is always

unremarkable. Declan called the control room. The control room confirmed all systems

nominal. The shift engineer initialled step seven on page four. Declan nodded.

Mark pressed the trip button on Module C.

The UPS system transferred the load to Modules A and B in less than four milliseconds.

The servers in Hall A and Hall B continued to process transactions. The lights in the

UPS room did not flicker. The BMS alarm panel showed a single amber warning: *UPS

Module C — Offline (Planned).* In the control room, nobody got up from their desk.

Three minutes later, Declan restored Module C. The load rebalanced across all three

modules. The amber warning cleared.

Declan initialled step fourteen on page eight. The shift engineer countersigned.

"That's it?" Ann said.

"That's it," Declan said. There was quiet satisfaction in his voice — the satisfaction of

a thing going as planned, which in a data centre was not nothing.

Ann was writing. She looked up. "So the redundancy works."

"The redundancy works," Declan said.

She turned to Mark, who had been standing to one side with his clipboard. He was

writing something. He didn't look up immediately.

"Mark?" she said.

He finished his sentence, then looked up. "Yes," he said. "The redundancy works."

She waited. She had learned to wait.

"For a single-module failure," he said.

The question he asked next was not dramatic. He asked it in the same tone he had used

to note the delta-T readings in the CRAH units, the same tone he had used to count the

condenser circuits on the roof. Flat, precise, directed.

"When did you last test two modules failing simultaneously?"

Declan said, "That's not a standard test."

"I know," Mark said. "When did you last do it?"

The silence was not long. Three seconds, perhaps four. But it was a specific kind of

silence — the kind that follows a question to which the honest answer is never, and

the person being asked is deciding whether the honest answer is the right answer to give

in front of a fund manager with a notebook.

"It's N+1," Declan said. "The design is N+1. You only need to test for what the

design covers."

Mark said, "That's correct."

He wrote something else in his clipboard.

Ann had started, in her notebook, a column she labelled *Things that are true but

incomplete.* She had been adding to it since Monday. It now had six entries. She

added a seventh: N+1 = one failure tolerated. Not two. Not concurrent.

She looked up from the notebook. "Can I ask a stupid question?"

"There are no stupid questions," Declan said, with the automatic reflex of a man who

had said this many times to many people and mostly meant it.

"What's the belt and braces?" she said.

Declan looked briefly pleased — this was a question he could answer fully and without

reservation. "Irish expression," he said. "If you're wearing a belt, your trousers won't

fall down. If you're wearing braces as well, they definitely won't fall down. Belt and

braces means you've covered the same risk two different ways."

"So N+1 is belt and braces."

"N+1 is belt and braces."

She looked at Mark. Mark was looking at the UPS modules, lined up in their grey

cabinets, humming with the sound of power being processed and delivered.

"Except," she said slowly, "both the belt and the braces are attached to the same pair

of trousers."

Mark turned around. He looked at her with an expression she hadn't seen from him

before — something approaching respect.

"That's exactly right," he said. "And the trousers, in this facility, are the transformer."

The transformer room was on the south side of the building, accessed through a

door marked Authorised Personnel Only in the kind of font that meant it. Mark had

the key from Declan's key safe. Ann stayed in the doorway.

The transformer was a cast resin unit, rated 2,500 kVA, stepping 10 kV MV supply

down to 400 V LV for the facility. It was the only transformer. It had been there since

2013 — she knew this because she had read the MV connection agreement on the wall

in Chapter One, and the transformer was part of the same installation.

Mark opened his clipboard to a fresh page. He noted the nameplate: manufacturer,

rating, year. He looked at the cooling fins along the sides. He looked at the condition

of the LV busbars connecting to the main switchboard.

Then he said, to nobody in particular: "Fifty-two weeks."

Ann said, "What?"

"Lead time on a replacement unit of this size. Fifty-two weeks minimum, possibly

seventy-eight. That's from the manufacturer's order to delivery in Dublin." He closed

his clipboard. "If this transformer fails — a winding fault, insulation breakdown,

sustained overcurrent — you cannot buy a replacement next week. You cannot buy

one next month. You can, at best, hire a mobile transformer while you wait for

the permanent replacement."

She looked at the transformer. It was, she realised, unremarkable looking — a grey

rectangular object that happened to be the single component standing between

5 MVA of grid supply and everything in the building. No redundancy. No standby.

One unit.

"And the UPS?" she said.

"The UPS sees the 400V output side of this transformer. If the transformer fails,

the UPS input goes dead. The UPS runs on battery. For" — he looked at her — "do

you know the battery runtime here?"

She didn't. She wrote find out battery runtime in her notebook.

"Ten to fifteen minutes at full load," he said. "That's industry standard for an

installation of this age without extended runtime battery strings." He paused.

"Enough time for the generators to start and pick up the load. But not if the

generators are also on the LV side of this transformer."

She looked at him.

"They're not," he said. "The generators connect directly to the LV main switchboard

via automatic transfer switching, bypassing the transformer. So that's correctly

designed." He said it the way he said most positive observations — without

celebration. A thing being correct was the baseline, not an achievement. "But

the transformer is still the single point of failure for grid supply to this building.

The UPS covers the gap between grid loss and generator pickup. It does not cover

the transformer."

Back in the control room, Declan had the asset register open on his laptop. Mark

had asked to see it.

The register listed 847 components across the facility: UPS modules, CRAH units,

generators, switchgear, PDUs, the transformer, the chillers, the condensers. For each

component: installation date, last service date, next service date, MTBF estimate

where known. It was, as asset registers went, well maintained — Declan's team

updated it quarterly.

Mark ran his eye down the list and stopped at three entries.

"These three," he said, pointing. "Static transfer switches. When were they

last exercised?"

Declan looked. "The STS units. They're — they do automatic testing. Self-diagnostic."

"Self-diagnostic confirms the logic works. Exercising confirms the transfer works

under load." Mark looked at the dates. The last load transfer test on each STS

was more than two years ago. "If an STS fails to transfer, the load on that bus

loses protection. It doesn't lose power — yet. But if the primary source then has

a disturbance, you're transferring to bypass on raw mains." He looked at Declan.

"Have any of them ever failed to transfer in a test?"

Declan said, "Not that I know of."

Mark wrote: STS load transfer test — schedule within 90 days.

Ann had one more question before she left for the afternoon. She found Mark in

the car park, writing up his notes before the drive back.

"The Tier classification," she said. "Where does Clonshaugh sit?"

He thought about how to answer this in a way that was both accurate and useful.

"Uptime Institute Tier II," he said. "N+1 redundancy, no requirement for

concurrent maintainability. You can maintain most systems, but not all of them,

without taking the load down." He paused. "Tier III would require concurrent

maintainability across all distribution paths — meaning you could take any single

component offline for service without a planned downtime window. That requires

2N distribution topology. Dual power paths from utility to rack, independent

all the way."

"And Clonshaugh has one path."

"Clonshaugh has one transformer, one MV connection, one set of generators, and

N+1 UPS modules. You can service the UPS modules concurrently. You cannot

service the transformer concurrently. You cannot service the MV connection concurrently."

She wrote: Tier II. Single path to transformer. Concurrent maintainability: partial.

"Is that a problem?" she asked.

"It depends on what your tenants are paying for," he said. "Tier II availability

is 99.741 percent per year. That's approximately twenty-two hours of potential

downtime annually, spread across all planned and unplanned events. For most

tenants, that's acceptable. For a tenant running payment processing on GPUs at

thirty kilowatts a rack—" He stopped. "That's a conversation for when you have

the tenant brief."

Declan was the last one in the building. He stood in the UPS room after Mark had

left, looking at the three modules — A, B, C — all green, all nominal, all carrying

roughly equal shares of the load now that the test was done and the system had

rebalanced.

He had run this room for eight years. He knew the sound of it — the background

hum that changed pitch slightly when the load shifted, the soft click of the

ventilation dampers, the particular whine of Module B's fan that had been there

for two years and that every service engineer had told him was within tolerance.

The test had gone perfectly. The redundancy had caught the trip. The MoP was

complete, signed, and filed.

And yet.

When did you last test two modules failing simultaneously?

He hadn't answered. He hadn't answered because the answer was never, and because

never was true, and because true wasn't the same as wrong. N+1 wasn't designed

for simultaneous failure. N+1 was designed for single failure. He had been testing

for what the design covered.

He was now thinking about what the design didn't cover.

He turned off the light in the UPS room — the modules kept running in the dark,

as they were designed to do — and walked back to the control room to write up

the test record.

He wrote: *UPS transfer test — Module C. Single module trip. Load transferred

successfully to A and B. Transfer time <4 ms. All systems nominal post-restoration.*

He signed it. He dated it.

Then he sat for a moment with the pen in his hand.

He added one line: *Action: Schedule simultaneous dual-module scenario assessment

with MEP engineer. Date TBD.*

It was the smallest possible thing. It was also, he thought, the right thing.

He capped the pen and went home.

The conference room at Clonshaugh was not designed for five people with opinions. It was designed for two people with a speakerphone. A narrow table, four chairs that didn't match, a whiteboard that had been cleaned so many times the surface was ghosted with the ghosts of old marker — numbers, dates, someone's mobile number, a diagram of something that might have been a cooling loop or might have been a doodle.

Mark had arrived first. He had pulled the table away from the wall so all five chairs could face the whiteboard, and he had written two things on it in black marker, in handwriting that was precise without being small:

PUE = Total Facility Power ÷ IT Equipment Power

And underneath:

Clonshaugh: ?

Ann arrived next, with Sarah.

Sarah was younger than Ann expected. Early thirties, dark jacket, laptop bag, no hard hat. She carried herself with the quiet precision of someone who had spent three years reading EU regulations and was now accustomed to being the only person in the room who had actually read them. She did not shake hands. She opened her laptop.

Tom came in behind them. Late fifties, SCSI lanyard, a notebook with a calculator clipped to the cover. He looked at the conference room the way a quantity surveyor looks at any room — he was pricing it. He sat down, uncapped a pen, and said nothing.

Declan was last. He brought coffee for everyone, in mismatched mugs. He set them on the table and sat at the end nearest the door, the way a man sits when he's not entirely sure whether he's the host or the subject.

Mark uncapped the marker. "I've walked the power chain and the cooling chain. Now we put numbers on it." He pointed at the formula. "PUE. Power Usage Effectiveness. The ratio of everything the building draws from the grid to what the IT equipment actually consumes. A perfect building — which doesn't exist — would be 1.0. Every watt from ESB goes to computing. No overhead. No cooling. No lights."

Tom looked up from his notebook. "What's the world average?"

"About 1.58," Mark said. "Uptime Institute, 2024 survey. Hyperscale targets are below 1.20. Google reported 1.10."

Tom wrote that down.

Mark turned to Declan. "What's ours?"

Declan set down his coffee. "1.50."

He said it the way he'd said "the cooling works" on the roof — with a note of ownership that wasn't quite pride, but was close. He had brought it down from 1.65 over seven years. New CRAH coils. Blanking panels in every rack. A chilled water optimisation that he'd argued for over two budget cycles. 1.50 was his number.

Mark wrote it on the whiteboard.

Clonshaugh PUE: 1.50

"What that means," Mark said, "is that for every watt of IT, we use half a watt of overhead. Cooling, power conversion, lighting, everything that isn't the servers."

He wrote the arithmetic below it.

IT Load: 2,400 kW

Facility Load: 2,400 × 1.50 = 3,600 kW

Overhead: 1,200 kW

Ann looked at the numbers. "1,200 kilowatts of overhead. Continuously."

"Continuously," Mark said. "Twenty-four hours a day, three hundred and sixty-five days a year."

Sarah opened her laptop. She had been waiting for this moment with the particular stillness of a person who knows the next number in the sequence before anyone else in the room.

"The EU Taxonomy Delegated Act 2021/2139," she said. "Section 8.1. For a data centre to qualify as a Taxonomy-aligned sustainable economic activity, one of the criteria is PUE less than or equal to 1.3. Annualised. Metered."

She turned the laptop so the table could see the screen. The number was there in black and white.

PUE ≤ 1.3

Declan looked at it. Then he looked at the whiteboard.

1.50 and 1.3. The gap was 0.2.

Tom leaned forward. He had been watching the whiteboard the way he watched a cost plan — looking for the number nobody had calculated yet.

"What does that 0.2 cost us per year?" he said.

It was the first time he had spoken since he sat down. The room turned to him. Mark handed him the marker.

"You do it," Mark said.

Tom stood up. He didn't need the calculator — the arithmetic was multiplication, and he was a man who had multiplied numbers for a living for thirty years. But he used the calculator anyway, because in his experience, people believed numbers that came from a machine more than numbers that came from a head.

He wrote as he spoke.

PUE gap: 1.50 − 1.30 = 0.20

Excess overhead: 2,400 kW × 0.20 = 480 kW

Annual excess energy: 480 × 8,760 hrs = 4,204,800 kWh

At €0.12/kWh: €504,576 per year

He underlined the number. Then he capped the marker and sat down.

Nobody spoke for a moment.

Half a million euro a year. For 0.2 of a ratio that most people outside this room had never heard of.

Sarah added the next line. She didn't stand up — she read it from her laptop, and Mark wrote it on the whiteboard.

"Grid emission factor for Ireland: 0.2241 kilograms CO₂ per kilowatt-hour. SEAI, 2026."

Mark wrote:

Carbon saving: 4,205 MWh × 0.2241 = 942 tCO₂/year

At €71/tCO₂: €66,882/year carbon cost reduction

Tom looked at the carbon number. "That's on top of the electricity saving?"

"That's on top," Sarah said. "The carbon tax is legislated to reach €100 per tonne by 2030. At that rate, the carbon cost of the PUE gap rises to €94,200 per year."

Tom added both numbers in his head. Over half a million in electricity. Sixty-seven thousand in carbon today, rising to ninety-four thousand in four years. He wrote in his notebook: €571K/yr current. €599K/yr by 2030. For 0.2 PUE points.

He looked at Declan. "Is that number right?"

Declan looked at Mark.

"Yes," Mark said.

Declan stood up and walked to the whiteboard. He didn't pick up the marker. He stood in front of the numbers Tom had written and looked at them the way you look at something that is true and unwelcome at the same time.

"I brought it down from 1.65," he said.

It was not a defence. It was a fact. He had spent seven years improving this building, and the number on the whiteboard was the result of that work. But the number on the laptop — 1.3 — didn't care about the starting point. It cared about the destination.

"You did," Mark said. "0.15 improvement over seven years is real work. The question now is the next 0.2."

"And what does the next 0.2 involve?"

Mark put his hands in his pockets. "That's the next six months of work."

Sarah said, quietly: "We don't have six months."

The room turned to her. She looked at her laptop. "The EED — Energy Efficiency Directive — Article 12. Data centres above 500 kilowatts IT load must report PUE, WUE, CUE, and renewable energy factor to SEAI annually. The first reporting cycle opened in 2024. We're at 2,400 kilowatts. We're in scope."

Declan sat back down. "What's WUE?"

"Water Usage Effectiveness," Sarah said. "Annual water consumed for cooling divided by IT energy. If you have cooling towers, it's material. If you have dry coolers, it can be zero."

"We have cooling towers?"

"You have nine condensing units on the roof," Mark said. "R-410A. They're air-cooled, not water-cooled, so your WUE for direct cooling water is effectively zero. But the question will come up in reporting, and the answer needs to be documented."

Sarah made a note. She was building a compliance map in her head — the kind of map that had dates instead of roads and regulations instead of towns. PUE. WUE. CUE. REF. Each one a metric. Each metric a reporting obligation. Each obligation a deadline.

Ann had been quiet. She had her notebook open — the same notebook she'd been writing in since the first walk through the building — and she had added a new column. She labelled it What 0.2 means and underneath she had written four entries:

Electricity: €504,576/yr

Carbon: €66,882/yr (rising to €94,200)

Taxonomy: fail

EED: in scope, not yet reported

She looked at the column. Each line was a different language — financial, environmental, regulatory, compliance — but they were all saying the same thing. The building was 0.2 above a threshold, and every year that gap remained open, it cost money, generated avoidable carbon, and failed a classification that mattered to the kind of investor she was trying to attract.

"What gets us from 1.50 to 1.30?" she said.

Mark took his time. He wasn't being dramatic — he was sequencing, because the answer was not one thing but several things in a specific order, and the order mattered.

"Metering first," he said. "Revenue-grade meters at the MSB for total facility power and at the PDU level for IT load. Right now, PUE 1.50 is estimated from the ESB bill. That's indicative but not verifiable. An EED auditor needs metered data. Category 2 minimum under EN 50600-4-2. Cost: sixteen to thirty thousand euro."

He held up a second finger. "Setpoint optimisation. The supply air temperature has been 18 degrees since 2013. Your servers are almost all ASHRAE A2-class — they tolerate 35 degrees inlet. Running at 18 when you could run at 23 or 24 is overcooling by five to six degrees. Each degree saves two to four percent of cooling energy. Zero capital cost — it's a BMS configuration change."

A third finger. "Containment. Hot aisle, cold aisle — physical barriers to stop hot exhaust mixing with cold supply. Three to five hundred euro per rack. PUE improvement of 0.1 to 0.3."

A fourth. "Free cooling. Dublin is below 18 degrees for 7,200 hours a year — 82 percent of annual hours. For most of the year, the outdoor air can do the cooling without running a compressor. Dry coolers on the roof, adiabatic assist for the warm months, BMS automation to switch between modes. Capital cost: five hundred to seven hundred thousand. Annual saving: three hundred thousand plus."

He stopped. Four actions. Metering, setpoints, containment, free cooling. Sequenced so that each one enabled the next, and each one paid for itself before the next began.

Tom had been writing. He had a grid on his notebook page now — four rows, columns for capex, annual saving, and payback. He was filling it in as Mark spoke, using his calculator for confirmation, checking each number against what he'd heard.

He looked up. "Phase 1 — metering and setpoints — costs under fifty thousand and saves a hundred and fifty thousand a year. That doesn't need board approval. I can put that through as operational expenditure."

Ann nodded. She had been thinking the same thing.

"Phase 2 — containment — is a hundred and fifty to two hundred thousand, payback under twelve months. Phase 3 — free cooling — is five to seven hundred thousand, payback under two years." He tapped his calculator. "The full programme is under a million euros with a combined payback of twelve to eighteen months."

He wrote one more number at the bottom of his grid, underlined it twice:

10-year NPV at 8%: ~€4.9 million.

Sarah leaned over to look at his notebook. She didn't say anything, but she smiled — not a wide smile, just the recognition of a number that would survive scrutiny.

The meeting was ending. Mark wiped nothing from the whiteboard. The numbers stayed.

Ann looked at the five of them — Declan with his coffee gone cold, Mark with his hands in his pockets, Sarah with her laptop open to a regulation, Tom with a grid of numbers that made the case clearer than any report she'd read.

"We're all looking at the same building," she said. "We've been looking at it for two weeks. And none of us saw the same thing until today."

Declan looked at the whiteboard. PUE 1.50. Overhead 1,200 kW. Half a million a year.

"PUE 1.50 was the result of real work," he said. He wasn't arguing. He was placing a marker — for the record, and for himself. "I'm not saying it's enough. I'm saying it was real."

Mark nodded. "It was real. Now the target is real too."

They left the conference room in ones and twos. Sarah packed her laptop, checked her phone, and walked out without small talk — she had a compliance timeline to build and the EED reporting deadline was not negotiable. Tom stayed behind with his calculator, running the numbers one more time, because a QS who presented an unverified number to a board was a QS who didn't get invited back.

Declan stood by the whiteboard. He took a photo with his phone and sent it to himself. Subject line: Overhead.

Ann was last. She looked at the whiteboard one more time — the formula, the gap, Tom's arithmetic, Mark's sequence.

Five people, five reactions. None of them were wrong.

She turned off the light and closed the door. The whiteboard stayed.

The email from ESB Networks arrived at 14:17 on a Wednesday. Planned outage. 10 kV feeder maintenance. Duration: four hours, 02:00 to 06:00, Tuesday next. Declan read it twice, printed it, pinned it to the notice board in the control room, and added the date to the whiteboard calendar in red marker.

He had managed thirty-one planned outages in ten years. Every one of them had gone the same way: UPS holds the load, generators start, load transfers, ESB does their work, grid comes back, load retransfers, generators cool down, everyone goes home. The building didn't know the difference. The tenants didn't know the difference. That was the point.

He began the preparation the way he always did — methodically, without rushing, because preparation was where mistakes were caught and execution was where they weren't.

Generator pre-start checks: oil levels, coolant, belt tension, battery condition. All four gensets — three duty, one standby. Each one rated at 1,250 kVA, configured N+1. Total installed capacity: 5,000 kVA for a building that drew 3,600 kVA at current PUE. Margin of 1,400 kVA. Comfortable.

Fuel: the day tank on each genset held enough for four hours at full load. The bulk tank in the yard held 15,000 litres — enough for approximately eight hours at full load across the fleet. He had topped it off on Monday. The fuel gauge read 98%.

He briefed the night shift at 22:00. Method of Procedure signed. The shift engineer initialled every step. Declan went home, set his alarm for 01:30, and lay in the dark thinking about nothing, which was how he slept before planned outages — not because he was worried, but because his mind knew the sequence and was running it without permission.

He was back at 01:45. Mark was already there.

Declan had stopped being surprised by this. Mark arrived before things happened. It was a professional habit that Declan recognised because he shared it — the difference between them was that Declan arrived early to run things, and Mark arrived early to watch things. Both were valid. Neither was comfortable with the other.

"You here to observe?" Declan said.

"If that's alright."

Declan handed him a hard hat and a radio. "Stay behind the yellow line in the generator room. Don't touch anything. If I say move, move."

Mark nodded. He had his clipboard and a stopwatch.

Ann arrived at 01:55. She had a takeaway coffee and a notebook. She stood in the control room doorway, out of the operational path, and said nothing. She was learning when not to speak, which was harder than learning when to ask questions. Tom was in London for a client presentation — he'd asked Declan to send the BMS screenshots in the morning. Sarah had seen the outage in the maintenance schedule and made a note: generator endurance — ask Mark.

At 02:00, the grid dropped.

The building changed. Not violently — not the way a house goes dark when the power cuts. The UPS modules engaged, the hum shifted pitch, and the emergency lighting came on in the corridors. The servers in Hall A and Hall B didn't pause. They kept processing transactions, streaming data, running queries, because the UPS batteries were holding the entire IT load, silently, on stored energy.

The BMS panel showed the countdown. Battery runtime: estimated 14 minutes at current load.

Fourteen minutes. The industry quoted ten to fifteen for VRLA batteries at this age and load. Declan knew this battery set was seven years old — original equipment, last capacity test eighteen months ago. He watched the countdown.

Ann watched it too. Fourteen minutes. She had never timed anything in her professional life by the second. She did not write in her notebook. There was nothing to write yet — just a number counting down, and the question of whether it would reach zero.

The generator start sequence initiated at 02:00:04 — four seconds after grid loss. The automatic transfer switch sent the start signal. In the generator room, the first engine turned over, caught, and stabilised. Then the second. Then the third. The fourth stayed on standby — N+1, one spare.

At 02:00:31, the transfer switch closed. Generator power reached the main switchboard. The UPS modules synchronised and began recharging. The battery countdown stopped. The building was running on diesel.

Declan exhaled. The shift engineer initialled step nine.

Everything had worked. The sequence was textbook. Thirty-one seconds from grid loss to generator load acceptance. The BMS showed all systems nominal. Hall A nominal. Hall B nominal. Cooling plant running on generator power. The building didn't know.

Mark wrote something on his clipboard. He didn't look up.

The generators ran for three hours and forty-seven minutes. At 05:47, ESB restored the 10 kV feed. The automatic transfer switch detected stable grid power, waited the programmed sixty-second verification delay, and transferred the load back. The generators ran unloaded for five minutes — the cooldown cycle — then shut down.

Declan signed the MoP. The shift engineer countersigned. Test complete. No interruption. No alarm. No incident.

Ann closed her notebook. "That was impressive," she said.

Declan allowed himself a small satisfaction. "That's the building working the way it's supposed to."

Mark was still writing. He finished his note, looked up, and said: "How much fuel did that use?"

Declan checked the gauge. "About 1,200 litres. Four hours at roughly seventy percent load."

"And the bulk tank holds fifteen thousand."

"Yes."

"So at that burn rate — call it 300 litres per hour at seventy percent — fifteen thousand litres gives you about fifty hours." Mark paused. "But that's at seventy percent. At full load, what's the burn rate?"

Declan thought about it. "Closer to 400 litres per hour."

"So at full load, the bulk tank gives you about thirty-seven hours."

"Roughly."

Mark wrote the number down. "When was the last time you tested beyond eight hours?"

The control room was quiet. The night shift was standing down. The generators were cooling in the yard, ticking softly the way engines do when they've been working and have stopped. Ann had gone to make a phone call. It was just Declan and Mark.

"The fuel contract," Mark said. "What does it guarantee?"

Declan pulled it up on his phone. He'd signed it four years ago, renewed automatically. The supplier was a Dublin firm — reliable, local. He scrolled to the service level.

"Best endeavours to deliver within four hours of request," he read.

"Best endeavours," Mark said.

"Yes."

"Not guaranteed."

Declan looked at the screen again. The words were clear. Best endeavours. Not guaranteed within four hours. Not guaranteed at all, if you read it strictly. Best endeavours meant they would try. It didn't mean they would succeed. It didn't mean they had a tanker on standby at two in the morning during a winter storm when every generator in Dublin was calling the same number.

Mark sat down on the bench by the control room window. He looked at his clipboard. The numbers were simple: thirty-seven hours of fuel at full load. A supplier contract that promised to try. No 48-hour island test in the facility's history.

"Resilience isn't starting," Mark said. "The generators started in thirty-one seconds. That's excellent. But starting is the first five minutes of a story that might last five days."

He flipped to a clean page in his clipboard. "I'd recommend a 48-hour island simulation. Planned. Fuelled. Monitored. You disconnect from the grid deliberately — or simulate it — and you run the building on generators for two full days. You test the fuel consumption rate, the coolant temperatures over sustained load, the automatic fuel transfer from bulk to day tank, the alarm thresholds, the handover between shift teams."

Declan said nothing for a moment. He was thinking about ten years of generator maintenance logs. Perfect records. Monthly start tests. Annual load bank tests — four hours, full rated load, every parameter within tolerance. Every test had been a sprint. Thirty seconds on. Four hours sustained. Then back to grid.

Nobody had tested the marathon.

"When was the last time anyone in Ireland did a 48-hour island test?" Declan asked.

Mark considered this. "The hyperscalers do it routinely. Purpose-built facilities with on-site fuel storage for 72 hours or more, contractual fuel delivery with SLA-backed response times, and dedicated tanker allocation. At this scale—" He looked around the control room. "I'd say it's rare. Not because it's difficult. Because nobody asks for it."

Declan walked to the generator room after Mark left. The four engines sat in a row — each one the size of a small car, grey and green, bolted to concrete plinths, exhaust pipes running up through the roof. They were warm. He could smell diesel and hot metal, the particular scent of engines that had been working and had earned their rest.

He put his hand on the nearest radiator grille. Still warm. The engine had started in under eight seconds. It had carried its share of the load for nearly four hours without a fault, without an alarm, without a temperature excursion. The oil pressure had held. The coolant temperature had held. The voltage and frequency had held.

And yet.

The bulk tank was at 92% — down from 98%. Four hours had cost 1,200 litres. A real outage — a storm that took out the ESB transformer, a cable fault that took days to repair, a grid event that nobody had planned for — would burn through that tank in a day and a half. After that, the building's survival depended on a phone call to a supplier whose contract said "best endeavours."

He thought about Storm Éowyn. Storm Debi. The Beast from the East. Planned outages came with a start time and an end time. Real emergencies came with neither.

He stood in the generator room and looked at the fuel lines running from the bulk tank to the day tanks. Automatic transfer — when the day tank dropped below 60%, the solenoid valve opened and fuel flowed from the bulk. He'd watched it happen during every test. It worked.

But at sustained full load, the transfer rate mattered. If the generators consumed fuel faster than the bulk-to-day transfer could supply, you had a problem that no amount of starting reliability could fix. It was a flow rate question, not a starting question.

He made a note: Verify bulk-to-day tank transfer rate at full load. Confirm flow exceeds consumption.

Ann found him in the loading bay at seven o'clock. The sun was coming up over the industrial estate, catching the generator exhaust stacks and the ESB cable route and the diesel tank behind the security fence. She had her notebook.

"How did the generators perform?" she asked.

"Perfectly," he said. And it was true. The generators had performed perfectly.

"But?"

He looked at her. She was getting better at hearing the word that came after the full stop.

"The tank holds thirty-seven hours at full load. The fuel contract says 'best endeavours.' Nobody has ever tested what happens when we need more than a day and a half of fuel and the supplier can't get here."

Ann wrote in her notebook. She wrote slowly, the way she did when the information was important enough to get the words exactly right.

"What does a 48-hour test cost?" she said.

"In fuel — maybe eight thousand euro. In planning and supervision — Mark would have to spec it. Call it twenty to thirty thousand all in, with the consultant and the monitoring."

"And what does it prove?"

"It proves the building can survive without ESB for two days. Or it proves it can't. Either way, we know."

She wrote the number. Then she looked at the diesel tank behind the fence. 15,000 litres. Thirty-seven hours. A contract that said "best endeavours."

"Is there an alternative to diesel?" she said.

Declan had been reading about this. "HVO. Hydrotreated Vegetable Oil. Drop-in replacement — same engines, same tanks, same injectors. Lower carbon, no sulphur. The hyperscalers are switching. It's about thirty percent more expensive per litre, but it reduces the Scope 1 emissions from generator operation to near-zero on a lifecycle basis."

"Near-zero?"

"Depends on the feedstock certification. But for EED and SFDR reporting, it's a material improvement. Sarah would know the numbers."

Ann looked at her notebook. Three entries from this morning: 48-hour test. Fuel contract. HVO.

None of them had been in her acquisition report. None of them had been in any acquisition report she'd ever read. The building's resilience had been described in one line: Backup power: N+1 diesel generators. As though the generators were the whole story, and the fuel was an assumption.

Declan stayed after Ann left. He went back to the control room and pulled up the maintenance log on the computer. Every generator start for the last five years. He scrolled through them.

Monthly: 168 start tests. Zero failures. Average start time: 8.2 seconds.

Annual: 10 load bank tests. Zero failures. Duration: 4 hours each. All parameters within tolerance.

A perfect record. Sixty tests, zero failures, five years. But every test had been the same test — the short test. The thirty-second test. The four-hour test. Start the engine, run it at load, bring the grid back, shut it down.

He had been testing the engine. He had not been testing the endurance.

He opened a new entry in the maintenance log. He typed carefully, because what went into the log stayed in the log, and what was in the log had to be followed up.

Action: Commission 48-hour island simulation test. Scope to include sustained generator operation, fuel consumption monitoring, bulk-to-day tank transfer rate verification, coolant temperature trending, shift handover protocol under generator power, and fuel delivery logistics. MEP consultant to specify. Budget estimate: €20–30K. Target date: TBD — within 90 days.

He saved the entry. He printed it. He pinned it to the notice board next to the ESB outage notification.

Then he picked up his phone and called the fuel supplier.

Mark had brought the connection agreement.

Not a copy — the original, in a clear plastic sleeve, the same document Ann had seen framed on the wall of the electrical intake room on her first visit. She hadn't asked him to bring it. He'd taken it down because it was, in his assessment, the single most important piece of paper in the building, and nobody had read it properly since 2013.

He laid it on the conference room table. The whiteboard still had Tom's arithmetic from the PUE meeting — nobody had wiped it, and it had become a kind of permanent installation, like the calendar on the wall that Sarah was building. Sarah was at Clonshaugh that day, but in the BMS room with Declan — she'd asked for a walkthrough of the monitoring architecture before she started mapping the EED requirements. Neither of them was needed for a conversation about the connection agreement. That was Ann's territory now.

Ann looked at the document. Maximum Import Capacity: 5 MVA. Voltage: 10 kV. Connection type: ESB Networks medium voltage distribution. Issued: 2013.

"Five MVA," she said. "We talked about this on the first day."

"We did," Mark said. "And at the time, it was context. Today it's a constraint."

Tom was at the end of the table with his calculator and a notepad. He had been quiet since the PUE meeting, working through the numbers that session had produced, turning them into the kind of phased investment structure that boards understood. But today's meeting wasn't about cooling. It was about power. And power, in a data centre, started with the wire.

Mark stood at the whiteboard. He drew a single line across the top — the ESB 10 kV cable — and wrote 5 MVA above it.

"This is the pipe," he said. "Everything in this building, every watt, comes through this line. 5 MVA is the maximum the connection allows. Our current demand—" He wrote the number below.

Current: 3,600 kW at PUE 1.50 ≈ 4.0 MVA

"We're at eighty percent of MIC," he said. "That sounds like headroom. It isn't. ESB Networks will typically flag any connection that consistently draws above seventy-five percent of MIC. We're already in that zone."

Ann looked at the numbers. "And if we densify?"

Mark had done this calculation before the meeting. He wrote it out.

Target: 400 racks × 20 kW = 8,000 kW IT

At PUE 1.30: 8,000 × 1.30 = 10,400 kW ≈ 11.6 MVA

He underlined 11.6 MVA and drew an arrow back to the 5 MVA at the top.

Tom put his calculator down. Not because he'd finished the calculation — because the calculation had finished itself. 11.6 MVA against a 5 MVA pipe. The building's ambition was more than twice its connection.

"That's not a cooling problem," Tom said. "That's a grid problem."

"It's both," Mark said. "But the grid problem comes first. You can design the most efficient cooling system in Ireland and it doesn't matter if you can't get the power into the building."

Ann had a question she'd been carrying since the first walk. She'd written it in her notebook on day one and circled it: What happens when we need more power?

"Walk me through it," she said. "What does a connection upgrade look like?"

Mark took his time. This was not a simple answer, and he was careful with answers that had timelines attached to them, because timelines in Irish grid connections were the kind of thing that turned investment plans into investment problems.

"Step one: pre-application enquiry to ESB Networks. You tell them what you want — how much additional capacity, at what voltage, at what location. They tell you whether it's feasible on the existing network." He wrote on the whiteboard as he spoke.

1. Pre-application → feasibility (8–12 weeks)

2. Formal application → technical assessment (12–16 weeks)

3. ESB Networks design → connection offer (6–12 months)

4. Customer acceptance → build programme (12–24 months)

"Total timeline from enquiry to energisation: eighteen months to three years. Depending on grid zone capacity, transformer availability, and whether the network needs reinforcement upstream of your connection point."

Ann counted the stages. Four stages. Eighteen months minimum. Three years possible.

"And that's for staying on 10 kV?"

Mark shook his head. "At 11.6 MVA, you're above the 10 kV threshold. ESB Networks medium voltage distribution goes up to about 5 MVA on a standard connection. Above that, you're looking at 38 kV — a dedicated high voltage substation on site, with ESB Networks still as the distribution system operator. Your own HV switchgear, your own transformer, your own protection equipment."

Tom wrote a number on his notepad and circled it. Ann leaned over. She couldn't read Tom's handwriting, but she could read the number of digits.

"How much?" she said.

"For a 38 kV substation with dual transformer, protection equipment, civils, and the ESB connection works?" Mark paused. "One and a half to three million euro, depending on scope. Plus the ESB connection charge, which is separate and can be three hundred thousand to over a million depending on network reinforcement."

Tom had stopped writing. He was looking at the whiteboard — the timeline, the costs, the gap between 5 MVA and 11.6 MVA — and doing what quantity surveyors do when the numbers don't fit the programme: recalculating the programme.

"The cooling retrofit," he said. "We priced that at under a million, payback in twelve to eighteen months. This" — he pointed at the grid numbers — "is three to four million with a three-year timeline. Is that before or after the cooling?"

Mark turned to face him. "Before. Or more precisely, in parallel. The cooling retrofit reduces overhead — it gets PUE from 1.50 toward 1.30. That saves money and carbon, but it doesn't change the IT load. The density increase changes the IT load. If you're going from 6 kilowatts per rack to 20 kilowatts per rack, the cooling retrofit and the grid upgrade have to run concurrently. The cooling retrofit starts now and delivers in twelve to eighteen months. The grid application starts now and delivers in eighteen to thirty-six months. The grid is the longer programme."

"So the grid application starts now," Ann said.

"It starts now," Mark said. "Not after the cooling is done. Not after the board approves Phase 3. Now. Because the grid timeline is the longest single item in the entire programme, and it is on the critical path for densification."

Ann looked at the connection agreement on the table. 2013. The year the building was commissioned. Thirteen years ago. The building had changed — new tenants, new load profiles, new ambitions. The connection hadn't changed at all.

"Has anyone reviewed this since it was issued?" she asked.

Declan, who had been sitting at the back of the room with his coffee, said: "No."

It was the shortest answer he'd given in five chapters of meetings, and it was the most honest. Nobody had reviewed the connection agreement since 2013. Nobody had checked whether the MIC was still appropriate, whether the demand profile had shifted, whether the network upstream had changed. The connection had been taken for granted — like the transformer, like the fuel contract, like the belt and the braces.

Mark capped the marker. He looked at the whiteboard — two meetings' worth of numbers, from PUE to grid capacity, from Tom's NPV to the 38 kV cost estimate. The whiteboard was becoming a map of the building's future, drawn in dry-erase marker by five people who each saw a different piece of it.

He turned to the room. "I'm going to write all of this up. The cooling survey, the PUE gap analysis, the generator endurance question, the grid connection recommendation. Formal report. You'll have it in two weeks."

Ann nodded. "And then?"

"And then you read it. You decide what you want to do — the phased cooling retrofit, the grid application, the 48-hour island test, the metering. You build a programme. And when you have a scope of works, you call me and I'll price it."

The room was quiet for a moment. Mark was leaving. Not dramatically — he wasn't walking out. He was finishing a phase of work and moving to the next. He had another project starting next week. A different facility, a different client, a different set of problems that required the same precision and the same questions. This was how consulting worked. You arrived, you observed, you diagnosed, you reported, and then you left — because the client needed time to read, to decide, and to commit. An engineer who stayed too long became a dependency. An engineer who left at the right time became a resource.

Tom shook his hand. "Good report. I'll check your cost estimates against SCSI."

Mark almost smiled. "I'd expect nothing less."

Declan walked Mark to the car park. It was the same walk they'd done every evening for two weeks — past the loading bay, past the generator compound, past the diesel tank and the ESB cable route. Declan knew this path better than anyone. He'd walked it thousands of times, in daylight and in darkness, in storms and in sunshine. The path hadn't changed. What he saw when he walked it had.

"Will you come back?" Declan said.

"When there's a scope." Mark opened his car door. "You've got a good building, Declan. It works. The question isn't whether it works today — it's whether it works in 2030. And that question has an answer, but the answer isn't free."

Declan nodded. He'd been hearing variations of this from Mark for two weeks — not criticism, not complaint, just the persistent diagnostic pressure of a man who asked questions that nobody else had thought to ask.

"The fuel supplier called back," Declan said. "They can do a guaranteed four-hour delivery window if we upgrade the contract. Costs twelve thousand a year more."

Mark considered this. "That's a start. Put it in the programme."

He got in the car.

Ann watched Mark drive out of the Clonshaugh car park. She stood by the security barrier with her notebook and a coffee she'd forgotten to drink.

She had, in two weeks, learned more about this building than she had in the previous six months of ownership. She had walked the power chain, felt the transformer hum, watched the generators start in the dark, seen Tom write half a million euros on a whiteboard, listened to Sarah name regulations she'd never heard of, and watched Declan discover that testing an engine wasn't the same as testing endurance.

She had a building full of problems. A team of four — Declan, Sarah, Tom, and herself. A report coming in two weeks. And a connection agreement from 2013 that couldn't carry the future she was planning.

She opened her notebook to the page she'd started on day one. She crossed out the heading — Site Visit Notes — and wrote a new one:

Programme.

Declan appeared beside her. He had his hard hat in his hand, which meant he was going back to the data hall.

"What do we do now?" he said.

"We wait for the report. Then we build a programme."

"And the grid application?"

She looked at the ESB cable route — the single line from the street to the building, the same cable that had carried every watt since 2013. Five MVA. One path. One connection. Four hundred tenants and a future that needed twice the power.

"That starts Monday," she said.

Sarah arrived at Clonshaugh on a Monday morning with a roll of brown paper, a box of markers, a laptop, and Mark's report.

The report had arrived on Friday — forty-seven pages, precisely formatted, with an appendix of photographs, a cost summary, and a recommended action list that ran to two pages. Sarah had read it twice over the weekend. She had underlined eleven items. She had cross-referenced each one against her own regulatory database — a spreadsheet she had been maintaining for eighteen months, covering every EU, Irish, and CRU obligation that applied to data centres. The eleven items from Mark's report mapped to seven different regulations. Three of the regulations had deadlines within the next twelve months. Two of them interacted with each other in ways that neither regulator had intended.

She asked Declan for the meeting room wall.

"All of it?" he said.

"All of it."

He helped her clear the whiteboard — Tom's PUE arithmetic and Mark's grid numbers, photographed before they were erased — and she taped the brown paper across the full width of the wall. Three metres. She drew a horizontal line across the centre with a black marker and wrote a scale along it: months, from now to December 2030. Five years. Sixty months. She used a ruler.

Then she began.

The first deadline was the nearest. She wrote it in red.

EED Article 12 — Annual reporting to SEAI. PUE, WUE, CUE, REF. Threshold: 500 kW IT. Deadline: 15 May annually.

"We're at 2,400 kilowatts," she said. "We've been in scope since 2024. We haven't submitted."

Declan, standing at the back of the room, said nothing. He had received the SEAI letter. He had mentioned it to Ann. He had not yet responded.

Sarah wrote the second deadline below the first, in the same red.

EU Taxonomy — PUE ≤1.3 (annualised, metered). Delegated Act 2021/2139 Annex I Section 8.1. Required for Taxonomy-aligned classification.

"We're at 1.50. Tom calculated the cost of the gap in the last meeting — over half a million a year. But the Taxonomy isn't about cost. It's about classification. If the fund reports this asset under SFDR Article 8, the Taxonomy alignment is a disclosure requirement. We can't disclose what we don't meet."

She looked at Ann. Ann was writing.

The third deadline.

CRU/2025236 — 80% renewable energy procurement obligation for data centres connected to the Irish grid.

"CRU Decision Paper, published 2025. All grid-connected data centres must procure at least 80 percent of their electricity from renewable sources. Corporate Power Purchase Agreements, Guarantees of Origin, or verified green tariffs. Clonshaugh's current renewable share is approximately 40 percent — the grid average."

Tom, sitting at the end of the table, raised his pen. "What does a corporate PPA cost?"

"It depends on the counterparty, the duration, and the volume. For 2,400 kilowatts — call it 21,000 megawatt-hours per year — a 10-year PPA at current rates would be in the range of €0.08 to €0.10 per kilowatt-hour. That's below the current CRU blended rate of €0.12. The PPA may actually reduce the electricity cost."

Tom wrote the number. He underlined it.

She kept going. The brown paper filled.

F-Gas Regulation EU 2024/573 — Phase-down of high-GWP refrigerants. R-410A (GWP 2,088) on the roof. Service restrictions tightening. Replacement horizon: 2027–2030.

Mark had flagged this in Chapter 2. Nine condensing units, all R-410A. The phase-down didn't ban the refrigerant overnight — it restricted supply through a quota system that ratcheted tighter each year. The practical effect: service costs rising, availability declining, and a point somewhere in the next three to five years where replacing the unit became cheaper than recharging it.

"Mark's report recommends budgeting for condenser replacement as part of the cooling retrofit," Sarah said. "The F-Gas timeline and the PUE timeline converge on the same equipment."

She drew a line on the brown paper connecting the F-Gas deadline to the PUE deadline. Two different regulations, two different regulators, one set of condensers on the roof.

Carbon tax — €71/tCO₂ (Budget 2025). Legislated escalation to €100/tCO₂ by 2030. Finance Act schedule.

"At PUE 1.50, Clonshaugh's Scope 2 emissions are approximately 7,060 tonnes CO₂ per year." She wrote the arithmetic on the paper:

31,536 MWh × 0.2241 kgCO₂/kWh = 7,065 tCO₂/yr

"At €71 per tonne, the carbon tax exposure is approximately €502,000 per year. At €100 per tonne in 2030, it rises to €707,000 — and that's before any PUE improvement. Every 0.1 PUE reduction cuts roughly 470 tonnes per year."

Tom was writing steadily now. He had opened a new page in his notebook, labelled it Compliance Costs — Annual, and was building a column.

EPA IE Licence — Industrial Emissions Directive. Threshold: 50 MW thermal input for combustion plant.

Sarah paused. "This one is conditional. Clonshaugh at 2.4 megawatts IT isn't near the EPA threshold. But if we densify to 8 megawatts and add on-site generation — CHP, BESS, the energy centre that Mark's report mentions as a future option — the thermal input from gas engines could push past 50 megawatts. At that point, you need an EPA Industrial Emissions licence."

Ann looked up from her notebook. "How long does that take?"

"Twelve to eighteen months for a straightforward application. Longer if there are environmental objections. And the licence conditions include continuous emissions monitoring, annual environmental reporting, and a Financial Provision — a bond that covers the cost of site remediation."

"How much is the bond?"

"Depends on the site assessment. For a data centre with diesel generators and a gas CHP plant? Two hundred thousand to five hundred thousand euro."

Tom added a line to his column.

SFDR — Sustainable Finance Disclosure Regulation. Article 8 fund classification.

Sarah turned to Ann for this one. "Your fund reports under SFDR Article 8. That means you disclose the environmental characteristics of your portfolio assets. Clonshaugh's PUE, WUE, REF, CUE, and Taxonomy alignment all feed into that disclosure. If the numbers are poor — or worse, if they're missing — the fund's Article 8 classification is at risk."

Ann put her pen down. "What do you mean, at risk?"

"I mean the supervisory authorities — in Ireland, the Central Bank — can challenge a fund's Article 8 classification if the underlying disclosures don't support it. An Article 8 fund that holds a data centre with PUE 1.50, no metered EED submission, and 40 percent renewable share is presenting a sustainability profile that the data doesn't support."

The room was quiet.

CRREM — Carbon Risk Real Estate Monitor. Misalignment Year.

Sarah wrote this one carefully. "CRREM models the carbon intensity pathway that a property asset must follow to remain aligned with the Paris Agreement targets. When an asset's actual carbon intensity exceeds the CRREM pathway, it is said to be misaligned. The year that happens is the Misalignment Year."

She looked at the room. "CRREM does not publish a data centre pathway. The standard CRREM pathways are for offices, retail, logistics. For data centres, the carbon intensity metric is different — it's measured per megawatt-hour of IT energy, not per square metre. The bands we use — 200, 300, 400 kilograms CO₂ per megawatt-hour of IT — are derived, not published. They're screening-level estimates, not CRREM-certified thresholds."

She wrote on the brown paper in blue marker — the colour she used for caveats:

CRREM DC pathway: LBE-derived. T3/T4. Not published by CRREM Foundation. Methodology disclosed in all reports.

"At PUE 1.50 and grid factor 0.2241, Clonshaugh's current carbon intensity is approximately 336 kilograms CO₂ per megawatt-hour of IT. Against the derived 300 band, the Misalignment Year is approximately 2031. Against the 200 band, it's already misaligned."

Tom said: "What does misalignment mean in money?"

"In money, it means institutional investors start asking whether the asset is on the wrong side of the pathway. Green finance becomes harder to access. Valuations are affected — not by the building's income, but by its carbon trajectory. The asset can be profitable and misaligned at the same time."

Sarah stepped back from the wall. The brown paper was covered. Red deadlines, black descriptions, blue caveats, and — this was the part that made the room go quiet — green lines connecting the ones that interacted.

EED reporting required PUE metering. PUE metering was Phase 1 of the cooling retrofit. The cooling retrofit addressed the Taxonomy threshold. The Taxonomy threshold fed into SFDR disclosure. SFDR disclosure supported the fund's Article 8 classification. The F-Gas phase-down affected the same condenser units that the cooling retrofit would replace. The carbon tax escalation made every PUE improvement worth more each year. The CRU renewable obligation required a PPA that might reduce electricity cost. The grid connection upgrade enabled the density increase that triggered the EPA threshold.

Lines between lines between lines.

Ann stood up and walked to the wall. She studied it the way she had studied the power chain on the first day — tracing connections, following paths, seeing the system.

"We're not behind on any single one of these," she said. "We're behind on all of them at once."

Sarah nodded. "That's the point. They were designed by different regulators in different years for different purposes. Nobody coordinated them. But they all land on this building in the same decade."

Declan, from the back of the room, said: "We're going to need Mark back."

Sarah turned to him. "We're going to need a programme."

Tom closed his notebook. He had a column of numbers — annual compliance costs, one-off investment costs, cost of non-compliance, cost of delay. He had added them up three times, because the total surprised him.

"The cost of doing nothing," he said. "I can price it now."

He stood up and wrote on the brown paper, below Sarah's timeline, in his particular handwriting — compact, precise, the handwriting of a man who had been signing cost plans for thirty years.

Annual cost of current gaps:

PUE overhead: €504K

Carbon tax (current): €502K

Carbon tax premium (2030 vs today): +€205K

Non-compliance risk (EED, SFDR, CRU): unquantified but material

Total quantified annual exposure: ~€1.2M/yr, rising

He underlined the number. Then he wrote below it:

Programme cost (from Mark's report): ~€3–4M over 3 years

Annual saving once complete: ~€1.2–1.5M/yr

Payback: 2.5–3 years on quantified items alone

He capped the pen. "The programme pays for itself. The question isn't whether to do it. It's the sequence."

The meeting ended. People left. The calendar stayed.

Nobody took it down. Nobody suggested it should be moved, or summarised into a slide, or filed in a folder. It stayed on the wall of the conference room at Clonshaugh, and for the rest of the programme — for every meeting, every decision, every argument about priority and budget and timing — the calendar was there.

Declan walked past it on his way to the data hall every morning. He noticed when Sarah updated it — a date moved, a line added, a green tick on a deadline met. The ticks were rare at first. They became less rare.

Ann walked past it when she arrived for site visits. She had started coming weekly instead of monthly. The calendar was the reason. Not because it scared her — although it had, at first — but because it showed her the shape of the problem. And a problem with a shape was a problem you could programme.

Tom walked past it and checked his numbers. He adjusted the cost of non-compliance upward after each new regulatory development, because the cost of non-compliance only moved in one direction.

And Sarah — Sarah walked past it every time she came to Clonshaugh, and she straightened it when the tape came loose, and she replaced the markers when they dried out, and she added new deadlines when they appeared, because they always appeared. The regulatory landscape didn't stop. It didn't pause. It didn't wait for the programme to catch up.

The calendar was the title of the book, made visible. The clocks were on the wall, and they were ticking.

The fire suppression inspection was scheduled for a Tuesday, as it was every year. A specialist contractor arrived at nine, signed in at the desk, and spent four hours opening every clean agent cylinder panel, checking every VESDA sample pipe connection, and testing every abort switch in both data halls. It was routine. Declan walked the inspection with him, as he always did, because a fire system you didn't understand was a fire system you couldn't trust.

Hall A passed without a comment. The system in Hall A used Novec 1230 — a fluoroketone suppressant, FK-5-1-12, installed during a retrofit in 2019. GWP of 1. Atmospheric lifetime measured in days. The inspector checked the cylinder pressure, confirmed the room integrity hold time of ten minutes, tested the VESDA aspiration points, and signed the certificate.

Hall B was different.

The system in Hall B was original equipment — installed in 2013, same year as the building. FM-200. HFC-227ea. The inspector checked the cylinders, checked the hold time, tested the detection, and signed the certificate. Everything passed. The system worked.

Sarah was in the corridor when the inspector left. She was on her way to the BMS terminal — she had been spending two days a week at Clonshaugh since the calendar went up on the wall, working through the EED reporting requirements that her timeline had made urgent. Ann was in Frankfurt for a fund LP meeting. Tom had the F-Gas figures in an email from Declan — he'd look at the liability side when he had the replacement cost confirmed. She glanced at the inspection certificate that Declan had pinned to the notice board outside Hall B.

She stopped.

"FM-200," she said. "What's the GWP?"

Declan looked at the certificate. "I don't know off the top of my head."

"3,220," Sarah said.

They stood in the corridor between the two data halls. Hall A on the left: Novec 1230, GWP 1. Hall B on the right: FM-200, GWP 3,220. One building, two suppression agents, and the higher-GWP agent was subject to the same regulation that was phasing down the refrigerant on the roof.

"EU 2024/573," Sarah said. "The F-Gas Regulation. It applies to any fluorinated greenhouse gas with a GWP above 150. FM-200 is 3,220. R-410A on the roof is 2,088. Both are in scope. Both are on the phase-down schedule."

Declan looked at the door to Hall B. He had walked through it ten thousand times. He had never thought of the fire suppression system as a regulatory problem. It was a fire system. It worked. The inspector had signed the certificate twenty minutes ago.

"When Mark flagged the R-410A on the roof," he said, "he was talking about the cooling system."

"He was. But the regulation doesn't distinguish between cooling and fire. It regulates the gas. If the gas has a high GWP, it's in scope — regardless of which system it's in."

Declan went to his office and did something he had never done before. He pulled the datasheets.

Not the operating manuals — he knew those. Not the maintenance schedules — he kept those current. The datasheets. The technical specifications that listed every chemical property of every gas in the building: molecular weight, boiling point, ozone depletion potential, global warming potential, atmospheric lifetime.

He spread them on his desk. Four documents.

Cooling — R-410A: GWP 2,088. Nine condensing units on the roof. Total charge: approximately 135 kg across the fleet. CO₂-equivalent: 135 × 2,088 = 281,880 kgCO₂-eq = 282 tCO₂-eq.

Fire suppression — FM-200 (Hall B): GWP 3,220. Total charge: approximately 480 kg (two banks of cylinders, sized for the room volume). CO₂-equivalent: 480 × 3,220 = 1,545,600 kgCO₂-eq = 1,546 tCO₂-eq.

He stared at the number. The fire suppression system in Hall B carried five times the CO₂-equivalent of the entire cooling refrigerant inventory. Five times. And he had never calculated it, because nobody had ever asked him to, because fire and cooling were different systems managed by different contractors on different service schedules.

Novec 1230 (Hall A): GWP 1. Total charge: approximately 520 kg. CO₂-equivalent: 520 kgCO₂-eq. Negligible.

Total building F-Gas exposure: 282 + 1,546 = 1,828 tCO₂-eq.

Of which 85% sat in one system in one hall that he had walked past every day for ten years without thinking about its regulatory classification.

He took the numbers to Sarah. She was at the desk she had claimed in the ops room — a corner space with her laptop, her regulatory spreadsheet, and a view of the car park that she never looked at because she was always reading.

"I have a number," he said.

He laid the datasheets on the desk. He'd highlighted the GWP figures and written the CO₂-equivalent calculations in the margins in his own handwriting — not neat, but clear. Declan's handwriting was the handwriting of a man who wrote things down so they wouldn't be forgotten, not so they would be admired.

Sarah looked at the calculation. She checked it — not because she didn't trust him, but because that was what she did with numbers. She nodded.

"1,828 tonnes CO₂-equivalent," she said. "Total building F-Gas inventory."

"The FM-200 in Hall B is the big one," Declan said. "Five times the cooling. I didn't know that."

"Most people don't. Fire suppression F-Gas exposure is the number that surprises everyone." She opened her regulatory spreadsheet. "Under EU 2024/573, any facility with more than 500 tonnes CO₂-equivalent of F-Gas must register in the EU F-Gas registry and comply with the record-keeping and leak-checking obligations. We're at 1,828 tonnes. We're well above the threshold."

"Are we registered?"

Sarah looked at him. The silence answered the question.

Tom arrived that afternoon. He had been working on the programme budget — the master spreadsheet that turned Sarah's calendar and Mark's report into euros and months. He found Declan and Sarah in the ops room with datasheets spread across the desk and a conversation that had moved past the problem and into the solution.

"What's the replacement?" Tom asked.

Declan had researched this. He didn't enjoy research — he was an operator, not a consultant — but Mark wasn't here, and the building needed answers that the maintenance log couldn't provide. He had spent the morning reading manufacturer specifications, and he had learned something that he would not have learned if Mark had been in the room: the learning itself had value.

"Novec 1230," he said. "Same agent that's already in Hall A. GWP of 1. Atmospheric lifetime of five days. Not caught by the F-Gas Regulation. Zero phase-down risk. The system in Hall A has been running since 2019 without a single issue."

"The cylinders are different," he continued. "FM-200 operates at 25 bar. Novec 1230 at 25 or 42 bar depending on the system — the higher-pressure version gives a more compact cylinder bank. The pipework may need modification. The room integrity test would need to be repeated for the new agent, because the hold time depends on the agent density, and Novec has a different density to FM-200."

Tom had his calculator out. "Cost?"

"For Hall B — approximately 550 square metres of data hall — I'd estimate €150,000 to €250,000 for a full agent swap. That includes new cylinders, pipework modifications, revised room integrity testing, VESDA recommissioning, and the disposal of the FM-200 stock."

Sarah added: "The FM-200 stock has residual value. There's a reclaim market — specialist firms that recover and recertify the gas for facilities that still need it. At current prices, the 480 kilograms might be worth €30,000 to €50,000 as reclaimed stock."

Tom wrote both numbers. Net cost: €100,000 to €220,000.

Tom leaned back. He was looking at the number in the context of the programme budget — the master spreadsheet that now covered cooling retrofit, grid application, PUE metering, free cooling, generator endurance testing, fuel contract upgrade, and a growing column of regulatory compliance costs.

"What's the cost of keeping it?" he said.

Declan had thought about this too. "The FM-200 service costs are rising. The gas itself is getting harder to source — the phase-down restricts supply, so every refill costs more. If we had a discharge — a system activation, even a false one — the cost of recharging the FM-200 at 2026 prices would be approximately €80,000 to €120,000. That's just the gas. Add the contractor call-out, the room integrity retest, and the downtime while the system is recharged — you're looking at €120,000 to €180,000 for a single discharge event."

"And the Novec equivalent?"

"A recharge at the same scale using FK-5-1-12: €40,000 to €60,000. The gas isn't subject to F-Gas quotas, and there's no reporting obligation on the refill."

Sarah looked up from her phone. She had been searching while Declan was talking — her habit of verifying everything in real time, the same habit that had built the calendar on the wall.

"There's a complication," she said. "3M. They've exited all PFAS manufacturing. Completed the exit at the end of 2025. Novec 1230 was their brand name for FK-5-1-12. The brand is discontinued."

Declan stopped. He had just spent the morning researching an agent whose manufacturer had left the market.

"The chemical is the same," Sarah continued, reading from her phone. "FK-5-1-12. The 3M patent expired in July 2020. There are at least thirteen manufacturers with UL component recognition for the same chemical, and nine with FM Global approval. Supply hasn't disappeared — it's diversified. But the Novec brand is gone, and there's something else." She scrolled. "EU ECHA has proposed broad restrictions on all PFAS compounds. FK-5-1-12 is a fluorinated substance. It may face additional regulatory pressure in the future — separate from F-Gas."

Tom put his pen down. "So we're replacing one regulated agent with another one that might get regulated?"

"Not the same regulation," Sarah said. "F-Gas targets high-GWP gases. FK-5-1-12 with GWP 1 is exempt from that. The PFAS proposal is different — it targets the chemical class, not the warming potential. But it hasn't been enacted. FK-5-1-12 is still the best available replacement for FM-200 today. The alternative is IG-541 — an inert gas blend, nitrogen, argon, CO₂. Zero GWP, zero fluorinated compounds, no PFAS exposure at all. But it needs bigger cylinders and more storage space."

Declan looked at Hall A. The system installed in 2019 used the same chemical. It worked. It had always worked. The manufacturer had changed, but the chemistry hadn't.

"We go FK-5-1-12," he said. "Same as Hall A. If the PFAS regulation catches up, we deal with it then — but at least we're not sitting on 1,546 tonnes of CO₂-equivalent in the meantime."

Tom added a line to his spreadsheet: F-Gas fire suppression swap — Hall B. CapEx €150–250K. Agent: FK-5-1-12 (multi-source, expired 3M patent). Avoided risk: €80–120K per discharge event. Ongoing saving: reduced F-Gas compliance burden. Note: monitor EU ECHA PFAS proposal — IG-541 as future contingency.

Declan did something that surprised Sarah. He asked for a marker.

He walked to the conference room, to the brown paper calendar on the wall, and he added a new line. He wrote it in red — the colour Sarah used for deadlines — in his own handwriting.

F-Gas suppression replacement — Hall B. FM-200 → FK-5-1-12. Budget: €150–250K. Target: before next annual inspection.

He taped it to the bottom of the calendar, below Sarah's regulatory timeline, below Tom's cost-of-doing-nothing numbers, below the grid application that Ann had started on Monday. His line was the newest, and the most personal. It was the first item on the calendar that Declan had put there himself.

Sarah watched from the doorway. She didn't say anything. She didn't need to. The calendar was a programme, and Declan was contributing to it — not maintaining the building, not defending it, but improving it.

That evening, Ann called from Dublin. She had been reading the programme budget that Tom had emailed — the latest version, with the F-Gas fire suppression line item added.

"Fire and cooling," she said. "The same regulation."

"The same regulation," Sarah confirmed. "Two systems, one clock. And the clock on the fire side is actually tighter, because the FM-200 GWP is higher than the R-410A. If the phase-down bites the fire system before the cooling system, we could be in a position where the building can't be reinsured with an F-Gas-non-compliant suppression agent."

Ann paused. "Say that again."

"Insurance. The fire suppression system is part of the property insurance underwriting. If the underwriter determines that the suppression agent is subject to phase-down and the building hasn't planned for replacement, they may add an exclusion or increase the premium. It's not a regulatory penalty — it's a commercial consequence of regulatory non-compliance."

Ann had never connected fire suppression to insurance underwriting to F-Gas regulation to the same calendar that tracked PUE and carbon tax. But it was the same calendar. It was always the same calendar.

"How many more of these are there?" she said. "Connections we haven't found yet."

"I don't know," Sarah said. "That's why we need the programme. A checklist catches the items. A programme catches the connections."

Declan drove home that night past the industrial estate entrance. The Clonshaugh sign was still faded. The security barrier was still the same height. The building looked the way it always looked — grey cladding, grey sky.

But inside the building, in a conference room that used to hold four chairs and a speakerphone, there was a wall covered in brown paper, covered in timelines, covered in numbers. And one of those numbers, written in his handwriting, was a fire suppression line item that he had researched, costed, and added to the programme without being asked.

He had been an operator for ten years. He was becoming something else. He didn't have a word for it yet. The closest he could get was: the person who sees the whole system, not just the part that hums.

Mark would have done the calculation in half the time. Mark would have known the GWP from memory, would have cited the regulation by article number, would have drawn the connection between fire and cooling in a single sentence. But Mark wasn't here. And the building still needed the answer.

The answer was the same regardless of who found it.

Sarah sat at the BMS terminal in the ops room at Clonshaugh and looked at the screen. Tom had been here last week — she'd passed him in the car park, heading out with a cost model on his laptop screen. He was building the lifecycle numbers for the monitoring spec; he'd said he'd have a figure by Friday. She had the regulatory side. He had the cost. For now, she worked alone. It showed her one number.

Total building load: 3,614 kW.

She needed fifty.

Not fifty guesses. Not fifty estimates derived from nameplate ratings and assumptions. Fifty metered, logged, exportable data points that would satisfy an EED Article 12 submission: power consumption per IT function, cooling energy by zone, UPS losses, lighting, PUE calculated from continuous metering at EN 50600-4-2 Category 2 minimum, WUE from a water sub-meter that she already knew didn't exist, and carbon usage effectiveness derived from all of the above.

The BMS gave her the building. She needed the building broken into its parts.

She clicked through the screens. Total MSB-A power. Total MSB-B power. Chiller status: running. CRAH units: all six showing green. Generator status: standby. UPS status: online. Temperature: a single return-air average from Hall A. No Hall B temperature displayed — the sensor had been offline since February.

She closed the screen and opened her laptop. She had brought the EED Delegated Regulation 2024/1364 — the document that specified exactly what data a facility above 500 kW IT load must submit, in what format, at what frequency. She had read it four times. She had built a checklist of every data field the regulation required. The checklist had forty-three items.

She started ticking the ones the BMS could provide. She got to seven.

Declan found her an hour later, still at the terminal, with her checklist and an expression he recognised. It was the expression of someone who had come to the building expecting a filing cabinet and found an empty room.

"What are you looking for?" he said.

"Everything," she said. Then, more precisely: "IT load metering at the PDU level. Cooling energy separated by component — CRAH fan power, chiller compressor power, pump power, condenser fan power. UPS efficiency — input versus output, real-time. Temperature logging per rack row, not per hall. Humidity by zone. Power factor at the MSB. Water consumption — although you don't have a cooling tower, so WUE should be zero, but I need a meter to prove it."

Declan sat down. He didn't interrupt. He was learning to listen to Sarah the way he had learned to listen to Mark — not because she was always right, but because she was always specific.

"The BMS was installed in 2013," he said. "It manages the cooling plant, the fire panel interface, the lighting, and the generator start sequence. It's a building management system. It was not designed to be a data centre monitoring platform."

"I know."

"The DCIM was added in 2016. Different system, different contractor, different protocol. It was supposed to do the rack-level monitoring — power per rack, temperature per row, capacity planning."

"What happened?"

Declan paused. "It was partially commissioned. The rack power meters were installed on thirty positions. Twelve of them went offline within the first year — communication faults, protocol mismatches, a firmware update that bricked a batch of sensors. The contractor came back twice. Then the contract expired and nobody renewed it."

Sarah asked to see it. Declan walked her through the building's monitoring infrastructure the way he had walked Ann through the power chain in Chapter 1 — methodically, without rushing, because the infrastructure was physical and you had to stand in front of each piece to understand what it did and didn't do.

The BMS head-end was in the ops room. A Trend controller — BACnet protocol, building-standard, perfectly adequate for HVAC management, lighting schedules, and fire panel integration. It had 340 points — a point being a single data connection to a sensor, a valve, a motor, a damper. Of those 340 points, approximately 280 were operational. The remaining 60 had been disabled over the years — sensors that had failed and not been replaced, valves that had been manually overridden and never returned to automatic, points that had been added for equipment that was subsequently removed.

"When a sensor fails," Declan said, "I can see it on the alarm screen. But if nobody's going to fix it this week, I disable the alarm so it doesn't clutter the overnight log. That's been the practice since 2015."

"How many disabled alarms?"

"Sixty-three."

Sarah wrote the number down. Sixty-three disabled alarms. Sixty-three points where the building had stopped talking to the management system, and the management system had been told to stop listening.

The DCIM was in a separate rack in the IT comms room. A Nlyte instance, version 8, installed on a virtual machine. Declan logged in. The dashboard showed a floor plan of Hall A with rack positions marked in green, amber, and grey. Green: monitored and reporting. Amber: monitored but stale data. Grey: no data.

Of the thirty rack power meters originally installed, eighteen showed green. Eight showed grey. Four showed amber — they were reporting, but the last data point was from October 2024. Eighteen months of stale data displayed as current, because nobody had configured the staleness threshold.

"The DCIM doesn't talk to the BMS," Declan said. "Different protocol. BACnet on the BMS. SNMP on the DCIM. There was supposed to be a gateway — a protocol translator — installed during the DCIM commissioning. It was on the scope of works. I have the quotation. It was value-engineered out during the budget review."

Sarah looked at him. "So the building management system and the data centre monitoring system are completely separate."

"Completely."

They stood in Hall A, between Row 4 and Row 5. Sarah pointed at the ceiling — cable trays, busway, smoke detectors, VESDA sample pipes, and every three metres a small white sensor box mounted to the rack rail.

"Temperature sensors," Declan said. "Inlet side, every third rack. They read. They display on the BMS if you navigate to the right screen. They don't log."

"They don't log?"

"They display in real-time. If you're looking at the screen when the temperature spikes, you see it. If you're not looking at the screen, the data is gone. There's no historian. The BMS logs alarms — threshold exceedances. But if the temperature rises from 22 to 28 degrees and back down again in ten minutes, and the alarm threshold is 32 degrees, nobody knows it happened."

Sarah thought about what that meant for PUE measurement. EN 50600-4-2 Category 2 required continuous metering with historical data retention. Continuous meant logged, timestamped, exportable. Real-time display without logging was not continuous measurement — it was observation, and observation didn't survive a shift change.

"Mark mentioned this," she said. "In his report. He recommended metering at PDU level for IT load and sub-metering at the MSB for facility load. He priced it at sixteen to thirty thousand."

"He did. That's the power metering. This is the environmental monitoring — temperature, humidity, airflow. Different budget, different sensors, different scope. The metering gives you PUE. The environmental monitoring gives you ASHRAE compliance evidence, thermal trend analysis, and the data to support setpoint optimisation."

She was beginning to see the full picture. The building had a BMS that managed systems. It had a DCIM that monitored racks — partially, intermittently, without history. It had temperature sensors that displayed but didn't log. It had power meters on eighteen of four hundred racks. It had no water sub-meter, no cooling component sub-metering, no protocol gateway between the two platforms, and sixty-three disabled alarms.

The building could run. It could not report.

They went back to the ops room. Sarah closed her laptop and opened her notebook — a physical notebook, lined, with a pen. She had started using paper at Clonshaugh because the digital tools she used in the office felt too distant from the building. Paper was something Declan understood. Paper stayed on the desk when the laptop closed.

"I came here thinking compliance was a reporting problem," she said. "Fill in the forms, submit the data, meet the deadline. What I'm looking at is not a reporting problem. It's a measurement problem. And a measurement problem is an infrastructure problem."

Declan nodded. He had been waiting for someone to say this — not because he wanted to hear it, but because it was true, and truths that were spoken aloud could be budgeted for.

"I can't report what you can't measure," Sarah said. "And you can't measure what you haven't wired."

Declan looked at her. "She's not wrong," he said — to himself more than to her, but she heard it, and it was the closest Declan had come to a compliment since she'd arrived.

They sat together for two hours and built the spec. Not the full DCIM upgrade — that was a €100,000-to-€300,000 project that required Mark's input on scope and architecture. The minimum viable monitoring infrastructure — the smallest investment that would allow Clonshaugh to produce a defensible EED Article 12 submission.

Sarah wrote the requirements. Declan added the site constraints.

Power metering (Mark's spec, confirmed):

- Revenue-grade meter at MSB-A and MSB-B: total facility power. €8–12K.

- IT load metering at PDU level (8 PDUs): IT power. €5–10K.

- UPS input/output metering: conversion efficiency. €3–5K.

Environmental monitoring (new scope):

- Temperature/humidity logging at every fourth rack position (100 sensors), with BMS historian integration. €15–25K installed.

- Cooling component sub-metering: CRAH fan power, chiller compressor, CHW pumps, condenser fans — 12 sub-meters. €8–12K.

- BACnet-SNMP gateway to connect BMS and DCIM. €5–8K.

- DCIM sensor recovery: replace 12 offline rack power meters, update firmware on 4 stale units. €6–10K.

- Data historian: 12-month rolling log with export to CSV for EED submission. €3–5K (software licence + configuration).

Total minimum monitoring investment: €53–87K.

Declan looked at the total. "That's not nothing."

"No," Sarah said. "But without it, every number on that calendar" — she pointed toward the conference room, toward the brown paper, toward the timeline she had built — "is an estimate. With it, every number is metered. The difference between an estimate and a measurement is the difference between a provisional EED submission and a compliant one."

Sarah sent the spec to Ann that evening. Subject line: We need to spend money to save money.

The email was three paragraphs. Sarah had learned that Ann read short emails in full and long emails in summary. Three paragraphs. The problem. The spec. The cost.

The building cannot produce a defensible EED Article 12 submission without metering infrastructure. The current BMS provides building-level data only. Rack-level, component-level, and environmental monitoring either don't exist or are partially offline. The minimum investment to produce compliant data is €53–87K. Without this, Mark's Phase 1 (PUE metering) and Phase 2 (setpoint optimisation) cannot be measured — we'll be optimising blind and reporting by estimation.

Ann read it on the train home. She read it again at the kitchen table. She thought about Mark's four-step sequence from the PUE meeting — metering, setpoints, containment, free cooling. Metering was first. She had understood that as power metering — the revenue-grade meters at the MSB and PDU. What Sarah was showing her was that metering was bigger than power. Metering was the entire nervous system of the building — every sensor, every logger, every protocol connection, every alarm threshold.

Measurement is infrastructure. Mark had said that. She hadn't understood it until Sarah showed her the sixty-three disabled alarms and the twelve offline sensors and the gateway that was value-engineered out.

She picked up her phone and approved the spec. No meeting. No board paper. Operational budget. The same decision pathway Tom had identified for Phase 1 — under the delegation threshold, no committee required, money spent to enable measurement that would justify every subsequent investment.

She typed one line back to Sarah: Approved. Start Monday.

Sarah read the reply at her desk in Dublin. Two words and a date. She had spent the weekend building a forty-three-item compliance checklist and the week discovering that thirty-six of those items depended on infrastructure the building didn't have.

She had arrived at Clonshaugh believing compliance was her job — read the regulation, map the requirement, fill in the form. She was leaving with a different understanding. Compliance was a programme, and the programme started with wiring.

She looked at her checklist. Seven items the BMS could provide. Thirty-six it couldn't. And a spec — her spec, co-authored with Declan, approved by Ann — that would close the gap.

It was the first time she had written an engineering specification. She suspected it wouldn't be the last.

She was about to close the laptop when her phone buzzed. Ann's email. Subject line: Portfolio review — Q3 infrastructure assets — call Thursday. The body was two lines: a dial-in number and a note that the fund's investment committee wanted a progress update on Clonshaugh before quarter-end.

Sarah stared at the subject line. She did not know what portfolio review meant in practice. She did not know what the investment committee would ask. She knew only that Thursday was in four days, that the monitoring spec had been approved but not yet installed, and that the form was still empty.

She forwarded Ann's email to Declan with one line: We should talk before Thursday.

The meeting was in Dublin. Not at Clonshaugh — at the fund's office, in a boardroom on the third floor with a view of the Liffey that nobody looked at, because the screen at the end of the table had numbers on it, and numbers were what this room was for.

Ann had called it. Not a site visit, not a corridor conversation, not a standing meeting in a conference room with brown paper on the wall. A proper boardroom presentation to the fund's investment committee — two portfolio directors, a risk officer, and Ann. Four people who made decisions in euros and decades, not kilowatts and months.

Sarah presented. She had built a slide deck — twelve slides, no animation, no stock photography, no platitudes about sustainability. Numbers. Pathways. Deadlines. The same information that lived on the brown paper wall at Clonshaugh, translated into the language that investment committees understood: risk-adjusted return, compliance trajectory, and asset classification.

Tom sat beside her with the cost model open on his laptop, ready to answer the question that always came from people who controlled capital: how much, and when does it come back.

Slide one was the carbon baseline.

Sarah had built this from the monitoring data that was now, finally, beginning to flow from Clonshaugh. The new meters had been live for six weeks — not long enough for annual figures, but long enough for a verified monthly baseline that she could annualise with stated uncertainty.

"Clonshaugh emits approximately 7,060 tonnes of CO₂ per year," she said. "Scope 2 — grid electricity. That's 31,536 megawatt-hours at the SEAI 2026 grid emission factor of 0.2241 kilograms CO₂ per kilowatt-hour."

She broke it down.

Scope 1: Diesel generators — approximately 45 tonnes CO₂ per year from testing and the annual planned outage. Reducible to near-zero via HVO conversion. FM-200 fire suppression in Hall B — 1,546 tCO₂-eq inventory, not an annual emission but a reportable F-Gas holding. Replacement programme underway.

Scope 2: Grid electricity — 7,060 tCO₂/yr. The dominant emission source. Reducible through two levers: PUE improvement (less electricity consumed) and renewable procurement (lower grid emission factor applied to the electricity consumed).

Scope 3: Tenant operations — not currently measured. DCIM monitoring (now partially restored) will enable tenant-level energy attribution. Required for CSRD reporting from 2026 onward.

"The total footprint is overwhelmingly Scope 2," Sarah said. "Grid electricity is both the problem and the lever. Everything we do on cooling, PUE, and renewable procurement addresses the same source."

Slide three was the pathway.

Sarah had modelled five scenarios over ten years, each building on the previous. She displayed them as lines on a single chart — current trajectory and four intervention levels.

Scenario A — Do nothing: PUE stays at 1.50. Grid emission factor declines slowly as Ireland adds renewable capacity. Emissions decline from 7,060 to approximately 5,200 tCO₂/yr by 2035 — purely from grid decarbonisation, no facility action. CRREM Misalignment Year: approximately 2031.

Scenario B — PUE improvement only: Cooling retrofit delivers PUE 1.30 by 2028. Emissions drop to approximately 4,600 tCO₂/yr by 2028, declining further with grid decarbonisation. Misalignment Year pushed to approximately 2035.

Scenario C — PUE + 80% renewable PPA: PUE 1.30 plus corporate PPA covering 80% of consumption (meeting CRU/2025236 obligation). Effective emission factor drops from 0.2241 to approximately 0.045 kgCO₂/kWh for the covered portion. Total emissions: approximately 1,400 tCO₂/yr by 2029. Misalignment Year pushed past 2040.

Scenario D — Full programme: PUE 1.20 (target), 80% renewable, HVO generators, Novec fire suppression, monitoring and continuous optimisation. Total emissions below 1,000 tCO₂/yr by 2030. CRREM-aligned beyond 2045.

Scenario E — Dispose: Sell Clonshaugh before the compliance gap becomes a valuation haircut. Tom had modelled this as the alternative — because every investment case needs a counterfactual.

The risk officer leaned forward. "What's the disposal discount?"

Tom answered. "At PUE 1.50 with no programme, a buyer's due diligence would identify the full compliance gap — EED, Taxonomy, CRU, F-Gas, grid connection. The typical discount for a non-compliant DC asset in the Irish market is fifteen to twenty-five percent on the net asset value. For Clonshaugh, that's a haircut of three to five million euros."

The room was quiet.

Slide five was the investment case.

Tom took this one. He stood up — the first time in the presentation — because numbers delivered standing carried more weight in a boardroom than numbers delivered sitting. He had learned this over thirty years of QS presentations, and the habit was as much a part of his practice as the calculator.

"Total programme cost across all phases: approximately €3.8 million over thirty-six months." He listed the major items:

Cooling retrofit (containment, free cooling, controls): €875K

Grid connection upgrade (38 kV substation, ESB works): €2.1M

Monitoring infrastructure (BMS/DCIM): €75K

Fire suppression swap (Hall B): €200K

Generator 48-hour test + fuel contract upgrade: €45K

PPA procurement and legal: €150K

Professional fees (MEP consultant, commissioning): €350K

"Against that, the annual benefit once complete:"

PUE energy saving: €850K/yr

Carbon tax reduction: €430K/yr (at €100/tCO₂ 2030 rate)

PPA electricity saving vs blended rate: €420K/yr

Density revenue uplift (20 kW/rack from 6 kW): €1.8M/yr additional tenant revenue

Taxonomy alignment: qualifies for green finance instruments

CRREM alignment: protects asset value for institutional hold

"Programme payback on energy and carbon savings alone: three years. Including density revenue: eighteen months. Ten-year NPV at eight percent discount rate: €7.6 million on energy and carbon savings alone. Density revenue is additional — subject to tenant occupancy and lease structure. I haven't included it in the NPV because a QS doesn't put gross revenue in a capital appraisal. That's a commercial decision, not an engineering one."

He sat down. He had delivered the number that mattered — €7.6 million NPV, defensible, auditable — and he had drawn the line that separated his number from the number the room would want to hear.

The portfolio director on the left — the one who ran infrastructure assets — asked the question Ann had been waiting for.

"Is this repeatable?"

Ann had been sitting quietly through the presentation. She had heard every number before — in the conference room at Clonshaugh, on the calendar wall, in Tom's spreadsheets, in Sarah's regulatory map. What she hadn't done, until this room, was place Clonshaugh in the context of the portfolio.

The portfolio held two data centres. Clonshaugh was the small one. The other was larger. Much larger.

She looked out the boardroom window. The Liffey was grey. The cranes on the quays were motionless. She turned back to the room.

"We've got a bigger one in Tallaght," she said. "Twelve hundred racks. Ten megawatts. Thirty-eight kV."

The room shifted. The risk officer put down his pen. The portfolio directors looked at each other.

"If we can't solve it here," Ann said, "we can't solve it there."

The sentence changed the meeting. It changed the programme. It changed, in a way that Ann was only beginning to understand, the way she thought about every asset in the portfolio.

Clonshaugh at 400 racks and 2.4 megawatts was manageable. The compliance gaps were significant but bounded. The retrofit programme was €3.8 million — material but not transformative. The engineering was proven. The team was functional. The calendar was on the wall and the green ticks were beginning to appear.

Tallaght at 1,200 racks and 10 megawatts was a different scale. Everything they had learned at Clonshaugh — the PUE gap, the F-Gas exposure, the monitoring infrastructure, the grid connection, the commissioning methodology, the regulatory convergence — would apply at Tallaght. But at four times the power, the numbers multiplied. The cooling retrofit would be larger. The grid connection would be 38 kV, not 10 kV. The generator fleet would be bigger. The carbon footprint would be four times heavier. The compliance programme would be four times more complex.

And the regulatory deadlines would be the same. The same EED. The same Taxonomy. The same CRU. The same F-Gas. The same calendar, at four times the scale.

Sarah understood immediately. "Tallaght is in scope for everything Clonshaugh is in scope for. EED, Taxonomy, CRU, F-Gas. But at 10 megawatts, it's also in scope for the CRU dispatch threshold and potentially the EPA IE Licence. We're not just repeating the programme. We're extending it."

Tom was already doing the arithmetic. He didn't share it yet — the numbers were rough and he never shared rough numbers — but the scale was clear. If Clonshaugh was €3.8 million, Tallaght was €12 to €18 million. If Clonshaugh's NPV on savings alone was €7.6 million — before density revenue — Tallaght's was €30 million or more before you touched the commercial upside. The investment case didn't just scale linearly. It compounded — because the revenue per megawatt at higher density was higher, the carbon reduction was larger, and the regulatory benefit applied to a bigger base.

Ann's crack was not dramatic. It was not a moment of failure or surprise. It was a shift in perspective — the kind that happens when you've been solving a problem and suddenly realise the problem was smaller than you thought, and the solution was bigger.

She had arrived at Clonshaugh six months ago thinking it was a property asset. Lease management. Cap rate. WAULT. She had learned it was a machine — power in, heat out, every system connected to every other system. She had learned it had regulatory clocks converging on a single decade. She had built a team that could identify the gaps, quantify the costs, and build a programme.

And now she understood: Clonshaugh wasn't the project. Clonshaugh was the methodology. The project was the portfolio.

She looked at Sarah. "The net zero pathway you've built for Clonshaugh — the carbon baseline, the scenarios, the CRREM alignment, the compliance map — can you rebuild it for Tallaght?"

Sarah nodded. "Same framework. Different inputs. The methodology is the same. The numbers are bigger."

"How long?"

"If Mark comes back and does the engineering survey, I can have the compliance map in two weeks. Tom can have the cost model in three. The pathway would be ready for this committee in six weeks."

The meeting ended. The portfolio directors approved Scenario D for Clonshaugh — full programme, €3.8 million, thirty-six months, begin immediately. They requested a preliminary assessment of Tallaght within sixty days.

Tom packed his laptop. The cost model had survived scrutiny — every number sourced, every assumption stated, every payback verified against SCSI benchmarks and RICS methodology. He had presented in DC economics, not property economics. Revenue per rack. Cost per kilowatt cooled. CRM capacity revenue at €149,960 per megawatt per year. He was thinking in a language he hadn't spoken six months ago.

Sarah gathered her slides. Twelve slides, forty-five minutes, one approval. The net zero pathway had survived the boardroom because it was built on metered data, disclosed methodologies, and stated uncertainties. The CRREM bands were marked Tier 3/4. The PPA rates were ranges, not points. The carbon tax was legislated, not assumed. Everything that could be verified had been. Everything that couldn't had been flagged.

Ann stayed in the boardroom after everyone left. She looked at the screen — still showing the pathway chart, five scenarios, ten years, one building. She thought about the other building. Twelve hundred racks. Ten megawatts. Thirty-eight kilovolts. A facility she had visited twice and understood less than she now understood Clonshaugh.

She picked up her phone and called Mark's office. She got voicemail.

"Mark, it's Ann. The committee approved the full programme for Clonshaugh. We're also looking at Tallaght — twelve hundred racks, ten megawatts, thirty-eight kV. I need you to come back. And bring a bigger scope."

She hung up. She looked out the window at the Liffey. Same grey water, same cranes, same city. But the portfolio had just changed shape, and the clock that had been ticking for one building was now ticking for two.

The tenant audit was scheduled for a Thursday. A financial services company — one of Clonshaugh's larger tenants, occupying sixty racks in Hall A — had sent their compliance officer with a checklist and a mandate. PCI-DSS. Payment Card Industry Data Security Standard. The company processed credit card transactions, and PCI-DSS required them to verify the physical security of every facility where their data was stored or processed.

The auditor arrived at nine. She was polite, thorough, and systematic. She signed in at the front desk, accepted a visitor badge, and asked three questions before she reached the data hall.

"Can I see your zone map?"

"Your what?" Declan said.

"The zone map. A diagram showing the security zones — perimeter, building entry, corridor, data hall, rack level — and how they align with your fire compartmentation and cooling zones."

Declan looked at her. He knew the answer. He knew every zone in the building, every access point, every badge reader, every camera, every fire compartment wall. He had managed access at Clonshaugh for ten years. He knew which tenants had access to which rows, which contractors needed escort, which doors required two-factor, and which ones were propped open on Tuesdays for the cleaning crew.

He knew all of it. And none of it was written down.

"I don't have a map," he said. "I have a system."

"Is the system documented?"

The silence was familiar. It was the same shape as the silence after Mark had asked about simultaneous UPS failures. The same shape as the silence after Sarah had asked about EED registration. The answer was no, and no was becoming a pattern.

The auditor walked the building. She was professional about it — she didn't challenge Declan, didn't embarrass him, didn't point out deficiencies in front of the tenant's IT staff. She walked, she photographed, she took notes, and she asked questions that Declan could answer but couldn't evidence.

"This door between the corridor and the data hall — is it on a separate access control zone?"

"Yes. Badge reader, PIN, audit trail."

"Show me the audit trail."

He pulled it up on the access control system. The log showed every badge swipe for the last ninety days — name, time, door, granted or denied. The system worked. The data was there.

"Good. Now show me how this zone boundary aligns with the fire compartmentation."

Declan paused. The fire compartment wall between the corridor and Hall A was on his left. The access control door was in front of him. They were not in the same position. The fire wall ran three metres to the south of the access door — meaning there was a three-metre section of corridor that was inside the fire compartment of Hall A but outside the access control zone of Hall A. Anyone with corridor access could stand in that three metres and be technically inside the fire compartment of the data hall without having swiped into the data hall.

It didn't matter in practice. The three metres was a dead-end corridor with a fire hose reel and nothing else. Nobody went there. Nobody had a reason to go there.

But the auditor noted it. And the note said: Security zone boundary does not align with fire compartment wall at Hall A corridor junction.

After the auditor left — professionally, with a handshake and a promise to send the report within ten days — Declan called Ann, Sarah, and Tom.

They met in the conference room. Sarah's calendar was on the wall. Declan's F-Gas line was at the bottom. Tom's cost-of-doing-nothing numbers were still visible in the corner. The room had become the programme office, whether anyone had named it that or not.

Declan described the audit. He was honest about it — not defensive, not dismissive. He had passed that stage somewhere between the PUE meeting and the fire suppression datasheet. He described the zone misalignment. He described the missing map. He described the three metres of corridor that didn't match.

"She's going to flag it," he said. "Not as a critical finding — PCI-DSS grades findings, and a zone misalignment with no data exposure is probably a medium. But it's on the report. And it stays on the report until we close it."

Ann nodded. "What do we need?"

"A zone map. A real one — not what's in my head. A drawing that shows every security zone, every fire compartment, and every cooling zone, and where they align and where they don't."

They built it together.

Declan pulled up the building floor plan on the large screen — a CAD drawing from 2013, last updated in 2018 when the Hall B expansion was completed. It showed walls, doors, columns, cable routes, and plant rooms. It did not show security zones, fire compartments, or cooling zones, because in 2013 nobody had asked a floor plan to show all three at once.

Sarah opened the fire safety file — the document that every building was required to maintain under Irish fire regulations. It contained the fire compartmentation drawings: walls rated at 60 or 120 minutes, fire doors with closers, fire dampers in the ductwork, and the boundaries between compartments. She traced them onto a printed copy of the floor plan in red.

Tom opened the lease plans — the demarcation drawings that showed which tenant occupied which rack rows, where the boundaries were, and which areas were shared. He traced them in blue.

Declan described the security zones from memory. Perimeter — the fence line and vehicle gate. Building entry — lobby, visitor desk, man-trap. Corridor — general access, badge-controlled. Hall A — data hall access, badge plus PIN. Hall B — same. Plant rooms — restricted, escorted access only. Roof — restricted, escorted. Generator compound — restricted. He traced them in green.

Four layers on one floor plan. Red for fire. Blue for lease. Green for security. And then — because the floor plan needed it — yellow for cooling zones, which Declan added from memory: CRAH zones 1 through 6, each serving a different section of the data hall, each with a different supply temperature setpoint, each connected to a different chiller circuit.

The floor plan filled up. And the gaps became visible.

Three misalignments.

The first was the corridor junction that the auditor had found — the three-metre gap between the fire compartment wall and the security door at Hall A. Low risk, easy fix: extend the access control zone to match the fire boundary, or move the badge reader to the compartment wall. Cost: €2,000 to €5,000 for a reader relocation.

The second was less obvious and more significant. The cooling zone for CRAHs 4, 5, and 6 served rack rows that crossed the fire compartment boundary between Hall A and Hall B. The fire wall separated the two halls — a 120-minute rated partition. But the chilled water pipework and the CRAH supply ductwork passed through the fire wall via penetrations that were sealed with fire-rated collars. The cooling was continuous. The fire compartmentation was not — because the fire dampers in the ductwork would close in a fire event, cutting cooling to the racks on the far side of the wall.

"If the fire dampers close in Hall B," Declan said, "CRAH 5 and 6 still run, but the air can't reach rows 12 through 16 in Hall B. Those racks lose cooling."

Sarah looked at the floor plan. "And those racks are in the financial services tenant's cage."

Nobody spoke for a moment.

Tom wrote: Fire damper closure = cooling interruption to tenant racks in Hall B rows 12–16. SLA exposure: TBD. Engineering fix: TBD. Mark needed.

The third misalignment was administrative. The access control system logged every badge swipe, but the logs were not cross-referenced with the fire safety evacuation list. In an emergency, the evacuation roll call was taken from a paper sign-in sheet at the lobby desk. The electronic access log showed who was in the building. The paper sheet showed who had signed in. The two lists were not synchronised — a contractor who tailgated through a held-open door would appear on neither.

"That's a life safety gap," Sarah said. "In a fire event, the evacuation controller doesn't know how many people are in the building."

Declan said: "I know how many people are in the building."

Ann, from her chair at the end of the table: "What happens when you're not here?"

They worked for three hours. The floor plan — now covered in four colours of marker, annotations, and sticky notes — was photographed, scanned, and saved to the shared drive. It was not a professional drawing. It was not an engineering document. It was four people's combined knowledge, captured on paper for the first time.

Tom priced the fixes. The badge reader relocation: €2–5K. A proper zone alignment survey by a security consultant: €8–15K. The fire damper / cooling zone engineering review: €5–10K (Mark needed). Electronic evacuation integration — linking the access control system to the fire panel roll call: €15–25K for the software, hardware, and integration.

Total: €30–55K. On the scale of the programme, it was a small number. On the scale of the audit finding, it was the difference between a medium and a closed.

Declan added the security items to Sarah's calendar on the wall. He wrote them in green — the colour he'd chosen for security on the floor plan. The calendar now had red, black, blue, and green. It was becoming a programme in colour.

Before they left, Declan saved the zone map to the shared drive. He labelled the file carefully:

CLONSHAUGH_ZONE_MAP_v0.1_DRAFT — NEEDS ENGINEERING REVIEW

It was the first time he had documented something that had previously existed only in his head. It was imperfect — hand-drawn zones on a 2013 floor plan, with annotations in four colours and sticky notes that wouldn't survive a strong draught. But it was on paper. It was on the server. And it had a version number.

v0.1. Not v1.0. The distinction mattered. He was asking for help, not claiming completeness.

Sarah noticed the version number. She didn't say anything, but she understood what it meant: Declan was no longer the person who kept the building running by knowing everything himself. He was becoming the person who documented what he knew, shared it with the team, and marked it as draft so that someone with more expertise — Mark, when he came back — could review and improve it.

Ann walked past the conference room on her way out. The calendar covered the wall. The zone map was pinned in the corner. The whiteboard still had faint ghosts of Tom's PUE arithmetic. The room looked like a war room — which, she thought, was exactly what a programme office should look like when the programme was real.

She thought about Tallaght. Twelve hundred racks. If the zone misalignment at Clonshaugh — a 400-rack facility — had taken three hours and four people to map, how long would Tallaght take? How many zones? How many fire compartments? How many cooling circuits crossing how many lease boundaries?

The methodology was the same. The scale was the problem. And the methodology would have to scale with it.

Mark walked back into the Clonshaugh conference room on a Wednesday morning in April, carrying drawings.

Not a clipboard. Not a notebook. Drawings — A1 sheets, rolled, in a cardboard tube. He set the tube against the wall and looked at the room.

The room had changed.

Sarah's brown paper calendar covered the entire west wall — three metres of regulatory deadlines, colour-coded, annotated, with green ticks appearing in places where they hadn't been before. Tom's cost-of-doing-nothing numbers were written in the corner, underlined twice. In the opposite corner, pinned at an angle, was a hand-drawn floor plan in four colours — fire in red, lease in blue, security in green, cooling in yellow — labelled in Declan's handwriting: CLONSHAUGH_ZONE_MAP_v0.1_DRAFT — NEEDS ENGINEERING REVIEW.

Mark looked at the zone map for a long moment. He didn't comment. He didn't need to. They had been working.

Ann was at the table with coffee. Tom was beside her with his calculator and the programme spreadsheet — now at version seven, two hundred and forty-three rows, every cost sourced, every payback calculated. Sarah had her laptop open. Declan was standing by the door, as he always did at the start of meetings, because he liked to see who came in and what they brought.

Mark had brought drawings.

He didn't open with the engineering.

He opened with a number. He wrote it on the whiteboard — the same whiteboard where Tom had written the PUE gap in Chapter 4, where the grid numbers from Chapter 6 still ghosted faintly under the erasure — and he wrote it large.

€149,960 / MW / yr

Tom's head came up.

"CRM T-4 clearing price," Mark said. "Capacity Remuneration Mechanism. The price the Irish capacity market pays for dispatchable generation that can deliver when called. SEMO published the result from the PCAR2829 T-4 auction. That's the number."

The room was quiet. Tom was looking at the whiteboard with the expression of a man who had spent thirty years benchmarking buildings and had just been shown a number from a different planet.

"Per megawatt," Tom said. "Per year."

"Per megawatt, per year. For generation capacity that is available when the system operator calls. Not energy sold. Capacity. The right to be available."

Tom's calculator was already in his hand. He didn't use it yet. He was doing the multiplication in his head, because the multiplication was simple — it was the implication that was complex.

Clonshaugh had four diesel generators. Total installed capacity: 5 MVA — approximately 4 MW at 0.8 power factor. Derate for availability and qualification: call it 3 MW dispatchable. At €149,960 per megawatt per year:

3 × €149,960 = €449,880 per year.

Nearly half a million euros a year. For generators that currently sat idle 8,750 hours a year and ran for 10 hours of testing.

Mark unrolled the first drawing. Heavy cartridge paper — A1, the kind that held a pencil line cleanly and didn't curl back at the edges when you held it flat. He weighted the corners with two coffee cups and his clipboard. The sheet was marked up in pencil — not printed, drawn — with annotations in a tight, even hand. In the bottom-right corner, a revision block: Rev 01. MEP schematic — energy centre concept. For discussion only. Not for construction. He had written that last line himself. He always did.

A single-line diagram — not of Clonshaugh as it was, but of Clonshaugh as it could be. An energy centre.

"This is not about backup power," he said. "Backup power keeps the lights on for eight hours when the grid fails. An energy centre keeps the business growing when the grid has nothing left to give."

He walked them through the topology. The existing diesel generators — four units, already qualified for standby — with HVO fuel conversion (from Chapter 5), controls upgrade for grid synchronisation, and CRM qualification testing. A new BESS installation — battery energy storage — sized at 2 MWh, capable of providing grid services, peak shaving, and short-duration UPS support. And potentially — this was the drawing he hadn't yet shared — a reciprocating gas engine for CHP, combined heat and power, providing both electrical generation and useful thermal output.

"The gas engine is a decision point," Mark said. "Not today's decision. But the architecture needs to allow for it, because once you build the energy centre platform, adding a CHP module later is integration, not construction."

Declan had been quiet. He knew generators. He had started them, maintained them, tested them, watched them run in the dark during planned outages. But this was not a generator conversation. This was an energy centre conversation, and the distinction mattered.

"So we've been solving the wrong problem," Declan said.

Mark turned to him. "You've been solving the right problem for 2013. The problem changed."

Sarah had the regulatory dimension. She had been reading the SEM — the Single Electricity Market — documentation since the Tallaght conversation in Chapter 10, because if the fund was going to invest in on-site generation at scale, the regulatory framework for market participation was her domain.

"CRU/2025236 requires dispatchable generation capability for data centres above the threshold," she said. "That's the compliance driver. But the same dispatchable generation, if correctly structured and registered, qualifies for SEM participation — DS3 system services, FASS, and the capacity market itself."

She looked at the room. "The compliance cost and the revenue stream are the same asset."

Tom put down his calculator. He had been running numbers for three minutes — the CRM revenue, the BESS capex, the gas engine option, the DS3 service payments — and the model was not behaving the way his models normally behaved. In his experience, compliance was a cost. Regulation was a burden. You budgeted for it, you absorbed it, you moved on.

This was different. The numbers said that building the asset to comply with CRU/2025236 generated revenue that exceeded the compliance cost. The energy centre didn't just pay for itself. It generated income. Ongoing, contracted, capacity-market income.

"Say that again," Tom said.

Sarah repeated it. "The compliance cost and the revenue stream are the same asset."

Tom was quiet for a long time. Then: "I need to rebuild the model."

Ann asked the question that connected the chapter to the book.

"If ESB can't give us more power, what do we do?"

The grid connection from Chapter 6 was still the longest item on the programme. Eighteen to thirty-six months for an MIC increase. The 5 MVA pipe couldn't carry the 11.6 MVA that full density required. And ESB Networks' capacity constraints in north Dublin were real — not speculative, not theoretical, but reported, discussed, and constraining every DC operator in the grid zone.

Mark pointed at his single-line diagram. "You generate. Not as backup. As primary. The grid becomes one source of power, not the only source. Behind the meter, you have generation capacity that serves the building directly and participates in the capacity market. The MIC stays at 5 MVA — or increases modestly — and the energy centre provides the incremental capacity."

"What does it cost?" Ann said.

Tom had the number. He had been building it for the last ten minutes from the components Mark had described, cross-referenced against published costs he had sourced since the programme started.

"BESS at 2 MWh: €500,000 to €800,000 installed. Generator controls upgrade for CRM qualification: €80,000 to €120,000. HVO conversion — already budgeted at €30,000 from Chapter 5. Civils and integration: €150,000 to €250,000. Total Phase 1 energy centre: €760,000 to €1.2 million."

He looked at his calculation. "Against CRM revenue of approximately €450,000 per year for the existing generator fleet alone. Payback on Phase 1: twenty to thirty months."

He tapped the calculator with his pen. "And that's before the BESS revenue from DS3 services, which Sarah tells me could add another €50,000 to €100,000 per year."

Sarah raised the question she always raised — the one that kept the ESG account honest.

"If we build a gas engine, how does that square with our ESG commitments?"

Mark deferred to her. This was her domain.

"The CRU renewable obligation is 80 percent of electricity consumed," Sarah said. "That's procurement, not generation. A corporate PPA from a wind farm covers the 80 percent. The gas engine provides dispatchable capacity — it runs when the grid needs it or when the building needs it, not continuously. The annual run hours might be 500 to 1,500, depending on market signals and grid events."

She paused. "It's not clean. The gas engine burns natural gas and produces CO₂. The Scope 1 emissions are real and reportable. But the obligation is 80 percent renewable by procurement, not 80 percent by on-site generation. A CPPA alongside the gas engine keeps the carbon account compliant."

The room was quiet. Sarah had delivered the distinction precisely — compliant, not clean. The regulation allowed it. The carbon account supported it. The ESG disclosure would report it accurately. But it was not zero-carbon generation, and she was not going to pretend it was.

"Renewable obligation doesn't mean renewable generation," she said. "It means renewable procurement. The distinction keeps the lawyers busy and the engineers honest."

Mark unrolled the second drawing. The microgrid topology.

"The question isn't whether you can generate," he said. "It's whether you can disconnect from the grid and keep running."

The diagram showed Clonshaugh in three modes. Mode 1: grid-connected, import only — the current state. Mode 2: grid-connected with behind-the-meter generation — the energy centre phase. Mode 3: islanded — disconnected from the grid entirely, running on on-site generation plus BESS.

"Islanding capability is the difference between backup and independence," Mark said. "In Mode 3, the building is its own grid. The generators, the BESS, and — if installed — the gas CHP engine provide all the power. The building operates without ESB."

Declan looked at the diagram. He had spent ten years connected to the grid, dependent on the grid, tested against grid failure, planning for grid restoration. The idea of operating without the grid — deliberately, commercially, as a feature rather than an emergency — was new.

"How long can you island?" he asked.

"Depends on fuel supply and BESS state of charge," Mark said. "With HVO generators and a 48-hour fuel guarantee — which you've already contracted — you can island for two days minimum. With a gas engine and mains gas connection, indefinitely."

"Indefinitely."

"Indefinitely. The gas network is not the electricity network. Different infrastructure, different constraints. Gas supply to north Dublin is not capacity-constrained the way the electrical grid is."

The meeting ended with a scope, not a decision. Tom had enough to build the business case — CRM revenue, BESS economics, phased capex, DS3 services. Ann had enough to take to the fund committee — the energy centre as grid constraint response, revenue generator, and CRU compliance asset. Mark had enough to write the engineering brief — single-line diagrams, protection coordination, islanding controls, BESS sizing.

What none of them had yet was the answer to Sarah's last question, asked as she packed her laptop.

"What does Clonshaugh look like in 2030 if we do nothing?"

Nobody answered. The question sat in the room after everyone left. It would sit there until Chapter 15.

Ann watched Mark and Tom walk to the car park together — the engineer and the quantity surveyor, side by side, talking about revenue per megawatt. She had never seen them walk together before. She had never seen Mark lead with a number before. She had never seen Tom silent before a number before.

The engineer who had learned to speak commercial. The QS who had just discovered that a data centre was not a building with computers. It was an energy asset.

She looked at the whiteboard. €149,960 / MW / yr. The number was still there. She left it.

The new CRAH units arrived on a Tuesday in three crates, each one wrapped in shrinkfilm and strapped to a pallet. Declan watched the crane lift them from the truck and set them on the loading bay, and he thought about the old units they were replacing — six CRAHs that had run since 2013, eleven years of continuous service, and not one of them had ever been formally commissioned.

Installed, yes. Switched on, yes. Tested by the contractor during the fit-out, signed off by the original engineer, and left to run. But commissioned — in the way Mark used the word, which meant tested under load, witnessed by an independent engineer, documented with recorded data, snagged against the design specification, and formally handed over with a signature that said I am satisfied that this equipment performs as specified — no. None of the original equipment had been commissioned to that standard.

Mark was in the plant room, reviewing the pre-functional checklist. He had written it — twelve pages, forty-seven items, each one requiring a signature, a measurement, or a photograph. The checklist covered everything from electrical isolation verification to chilled water valve stroke testing to the VESDA aspiration point confirmation for the new Novec 1230 system in Hall B.

He looked up when Declan came in.

"New CRAHs are on the loading bay," Declan said.

Mark nodded. "They don't go live until every item on this list is signed."

"I know."

"Every item."

"I know, Mark."

The pre-functional checks took two days. Not because they were complex — most of them were visual inspections, torque checks, and continuity tests — but because Mark required every item to be witnessed and recorded, and witnessing took time.

Declan understood the process intellectually. He had been an operator long enough to know that shortcuts during installation created problems during operation — the cable gland not tightened, the earth bond missed, the protection setting changed from the FAT value "to make it fit." He had fixed enough of those problems over the years to know where they came from. They came from the gap between the factory and the site, between the drawing and the building, between the specification and the reality.

What he hadn't understood — until now — was that the commissioning process existed specifically to close that gap. Not to catch mistakes, although it did. Not to blame contractors, although it could. But to create a documented record that said: at this time, on this date, this equipment was tested against this specification, and it performed as required. Or it didn't — and here's the snag list, and here's the retest, and here's the evidence that the snag was closed.

He watched Mark check the chilled water valve on the first new CRAH unit. Mark opened the valve manually, confirmed full stroke, then returned it to automatic and commanded it from the BMS. The valve responded. Mark measured the response time — 43 seconds from closed to full open. The specification said 30 to 60 seconds. Within tolerance.

He photographed the gauge. He recorded the time. He signed the line.

Forty-six items to go.

The functional performance tests came next. This was where the equipment moved from static to dynamic — not just installed and powered, but running under load, responding to setpoints, interacting with the chilled water system, the BMS, the monitoring infrastructure.

Declan drove the testing. He knew what the systems should do because he had run systems like them for a decade — the hum of the fan at speed three, the click of the valve actuator, the way the return air temperature should track the setpoint within two degrees after fifteen minutes of stabilisation. He knew this from experience. He knew it in his hands and his ears and the particular sense that operators develop after thousands of hours in plant rooms.

Mark knew what the specification said the systems should do, because he had written the specification. He had specified the fan speed curves, the valve response times, the chilled water flow rates, the supply air temperature tolerance, and the alarm setpoints. His knowledge was on paper — verified against EN 50600-2-3, cross-referenced with the manufacturer's commissioning manual, and backed by calculations.

The commissioning process was where those two kinds of knowledge met.

They tested the first CRAH at 50% load. Fan running, chilled water flowing, supply air at 18 degrees. Mark recorded inlet temperature, outlet temperature, airflow velocity at the tile face, chilled water flow rate, and fan power draw. Every number went into the commissioning record.

Then 75% load. Then 100%.

At 100%, the return air temperature settled at 32.4 degrees. The specification said 30 to 35 degrees return. Within tolerance. Mark signed.

Declan watched him sign. He had run the old CRAHs for eleven years without a commissioning record. He knew they worked because they worked — because the hall stayed cool, because the alarms didn't fire, because the tenants didn't complain. That was knowledge. It was valid. But it was not evidence.

The integrated systems test was the commissioning event that brought everything together. Not individual equipment — the system. The new CRAHs, the new monitoring sensors, the Novec 1230 fire suppression in Hall B, the BACnet-SNMP gateway that connected the BMS to the DCIM for the first time, and the data historian that logged every reading for EED Article 12 export.

Mark had written a test sequence — a scripted series of events designed to prove that the systems worked together, not just individually. Sensor reads temperature. BMS logs it. DCIM displays it. Historian stores it. Alarm threshold exceeded. BMS triggers alarm. DCIM escalates to dashboard. Shift engineer acknowledges. All recorded. All timestamped. All exportable.

The test took four hours. Three people in the room — Mark with the test script, Declan at the BMS terminal, and Sarah at the DCIM dashboard. Sarah had asked to be present, because the monitoring infrastructure she had specified in Chapter 9 was being commissioned, and she wanted to see — with her own eyes, on her own screen — whether the data she needed for EED compliance would actually flow from the sensor to the report.

It did. Mostly.

At 14:22, sensor R4-T07 — a rack-level power meter on Row 4, position 7 — read zero. The DCIM showed 0.0 kW. The rack was drawing approximately 5.8 kW based on the busway meter upstream. The sensor was installed, powered, communicating — and wrong.

Declan's instinct was immediate: "Note it. Fix it next week. It's one sensor."

Mark stopped. He put down his pen.

"If it isn't in the witness report, it didn't happen," he said. "And if a defect isn't in the snag list, it doesn't get fixed."

The snag was recorded. Defect notice DR-013-007. Rack power meter R4-T07. CT orientation suspected — a current transformer installed backwards would read zero regardless of load. Assigned to the DCIM contractor. Retest required within seven days.

Declan was frustrated. Not angry — he had passed angry somewhere around Chapter 5 and arrived at a quieter place. Frustrated, because the sensor failure was minor. It was one meter out of a hundred. The system worked. Ninety-nine sensors were reading correctly. The BMS-to-DCIM gateway was passing data. The historian was logging. Sarah's EED export format was populating. One meter reading zero did not change the building's ability to operate.

But it changed the building's ability to prove.

Mark gave him a moment. Then: "The DCIM contractor installed that CT. The commissioning process caught it. If we hadn't tested under load, that meter would have read zero for months — and nobody would have noticed until Sarah ran the EED report and found a gap in the data. Then it's not a sensor failure. It's a reporting failure. And the cost of a reporting failure is not a CT — it's an SEAI compliance notice."

Declan looked at the snag list. Seven items. Six closed. One open — DR-013-007. He had seen snag lists before. He had seen them on every project he'd ever worked on — the punch list, the defect log, the things that weren't right yet. He had never seen a snag list treated as evidence. In his experience, the snag list was a to-do list. In Mark's process, the snag list was a legal document.

The final step was the witness report.

Mark printed it — three pages, landscape format, every test recorded with date, time, reading, pass/fail, and witness signature. He spread it on the table in the conference room, under the brown paper calendar, beside Tom's cost numbers, opposite Declan's zone map.

"This is the commissioning record for Phase 1 of the Clonshaugh cooling retrofit," Mark said. "It covers the new CRAH units in Hall A, the monitoring infrastructure across four rack rows, and the Novec 1230 fire suppression in Hall B. Every item has been tested against the design specification. Every test has been witnessed. Every result has been recorded. One defect remains open — DR-013-007 — scheduled for retest next Tuesday."

He turned the report to face Declan.

"Your signature," he said.

Declan looked at the report. Three pages of data — temperatures, flow rates, valve response times, sensor readings, alarm tests, BMS integration checks. Every line a measurement. Every measurement a fact. Every fact signed by the person who witnessed it.

He had been running this building for ten years. He had kept it alive through storms, through planned outages, through tenant complaints, through equipment failures, through a pandemic. He knew every sound the building made — which hum was normal, which click was routine, which alarm was real and which was the sensor in the UPS room that had been drifting since 2019.

He knew this building. But knowing and proving were different things. Knowing lived in his head. Proving lived on paper. And paper was what survived when the person who knew left the building.

He picked up the pen. He signed.

Declan Murphy. Facility Manager. Clonshaugh Data Centre. Witness to commissioning record Phase 1. Date: [today].

Ann arrived after the signing. She had been at the fund's office for the Tallaght preliminary assessment — Sarah's compliance map was two-thirds complete, Tom's cost model was taking shape, and the sixty-day clock the investment committee had set was on track.

She found Declan in the conference room, sitting in front of the signed witness report, looking at it the way he'd looked at the maintenance log after the generator test — not with satisfaction, but with the quieter feeling of a thing done properly.

"How did it go?" she said.

"One defect. A sensor reading zero. CT orientation — the contractor installed it backwards. Mark caught it in the integrated test. Retest Tuesday."

She sat down. "And everything else?"

"Everything else passed. CRAHs within spec. Monitoring logging. BMS-to-DCIM gateway operational. Novec system in Hall B — room integrity confirmed, hold time ten minutes, VESDA commissioned. Sarah's EED export is populating."

Ann looked at the commissioning record. Three pages. She had read acquisition reports of eighty-four pages that told her less about a building than these three pages told her about Clonshaugh.

"This is what we'll need at Tallaght," she said. "This process. This standard."

Declan nodded. "I know."

Mark packed his commissioning kit — the calibrated thermometers, the thermal camera, the airflow hood, the clipboard that had been with him since Chapter 2. He had one more item to close.

He walked to the corner of the conference room and unpinned Declan's zone map — the hand-drawn floor plan in four colours, labelled v0.1, marked NEEDS ENGINEERING REVIEW.

He studied it for two minutes. He made three annotations in pencil — corrections to the fire compartment boundary at the Hall A/B junction, a note about the fire damper that crossed the cooling zone, and a mark where the access control door needed to be relocated.

He pinned it back to the wall and wrote at the bottom, in his particular handwriting:

Reviewed. Three items noted. Suitable as basis for engineering drawing. Issue to draughtsperson as v0.2. — M.

Declan saw it later that evening. He stood in front of the zone map and read Mark's annotations. Three items. Not a red pen through the whole thing. Not a rewrite. Three items — because the other forty items Declan had drawn from memory were correct.

v0.1 had become v0.2. His knowledge, reviewed by Mark, was now a document.

He turned off the conference room light. The calendar stayed on the wall. The commissioning report stayed on the table. The zone map stayed in the corner, with Mark's pencil annotations and the new version number.

The building was the same building. The same grey cladding, the same hum, the same cable trays and busway and smoke detectors. But it was proven now — not all of it, not yet, but the first phase. The first set of equipment that could point to a piece of paper and say: this was tested, this was witnessed, this was signed.

Declan went to write up the maintenance log. He wrote the entry for the day. And at the bottom, he added a line he would not have written six months ago:

Commissioning Phase 1 complete. Witness report signed. One open defect — DR-013-007, retest 7 days. All other items: PASS.

The email arrived at 16:47 on a Friday. Ann forwarded it to Mark, Declan, and Sarah with one word: Thoughts?

A prospective tenant. An AI company — Dublin-based, venture-funded, scaling fast. They wanted to know if Clonshaugh could host a GPU training cluster. Forty racks. Thirty kilowatts per rack. 1.2 megawatts of IT load in a footprint that currently drew 240 kilowatts.

Thirty kilowatts per rack. Five times the current average density. In forty positions that were designed for six kilowatts each.

Mark read the email at his desk at home. He read it again. Then he opened a fresh page in his notebook and wrote the number that mattered:

30 kW/rack. 40 racks. 1,200 kW. Air cooling ceiling: ~25 kW/rack.

The tenant was asking for something the building couldn't do. Not wouldn't — couldn't. The physics wouldn't allow it. And physics, unlike budgets and timelines, did not negotiate.

He had twenty-four hours before the tenant meeting. He started the feasibility assessment. Tom was already in Ann's calendar for Monday — the cost model would follow the engineering. Sarah had flagged the EU Taxonomy implications on high-density cooling in a note to Ann two weeks earlier. The engineering came first.

He began with the physics, because the physics determined whether the conversation was about how or whether.

Air has a specific heat capacity of 1.006 kJ/kg·K. Water: 4.18. For the same temperature differential, water carries four times the heat per kilogram. For the same heat removal at 30 kW per rack, you need four times the air volume compared to water — and at 30 kW, the air volume required through a standard hot-aisle containment exceeds what the raised floor plenum can deliver.

At 6 kW per rack — Clonshaugh's current average — air cooling was comfortable. The perforated tiles in the raised floor delivered cold air at sufficient volume and velocity. The CRAHs pressurised the plenum. The containment kept hot and cold separate. The system worked.

At 20 kW per rack — the original density target — air cooling still worked, with good containment, VSD-driven fans, and optimised supply temperature. This was the world they had been planning for since Chapter 4.

At 30 kW per rack, air cooling hit a wall. The wall was not design — it was physics. The specific heat capacity of air set an absolute limit on how much heat could be removed by blowing air through a rack at any reasonable velocity. Above approximately 25 kW per rack, the air volume required either exceeded the floor tile capacity or created noise levels above 85 dBA — the occupational exposure limit.

The answer was not more air. The answer was a different medium.

Mark's first option was direct-to-chip liquid cooling.

A copper cold plate bonded directly to the GPU die — the hottest component, the primary heat source. Coolant — a water-glycol mixture, typically 18 to 45 degrees Celsius supply temperature — circulated through micro-channels in the cold plate, picked up heat at the source, and returned to a Coolant Distribution Unit on the data hall floor.

The CDU was the interface. Facility chilled water on one side, server-level coolant on the other, separated by a plate heat exchanger. The CDU controlled the supply temperature, the flow rate, and the pressure. It was, in cooling terms, the translator between the building and the chip — the same way the transformer in Chapter 1 translated between ESB's voltage and the building's voltage.

At 30 kW per rack, direct-to-chip captured 60 to 70 percent of the server's heat at the source. The remaining 30 to 40 percent — drives, memory, power supplies, ancillary components — was still rejected as hot air into the data hall. The air cooling system still ran, but at a fraction of the load. The CRAHs handled the residual. The liquid handled the peak.

"The question," Mark wrote in his feasibility note, "is whether Clonshaugh's existing chilled water plant can feed CDUs at the required flow rates and temperatures."

He modelled it. The chiller had capacity — the cooling retrofit from Phase 1 had sized the new plant for PUE 1.20 at 2,400 kW IT, which meant the chiller could handle the thermal load from forty racks at 30 kW if the total IT load stayed within the MIC envelope. The problem was the distribution pipework. The chilled water headers had been sized for CRAH connections — large-bore, low-velocity, long runs. CDU connections required higher flow rates at each rack row, with secondary branch pipework that didn't exist.

Chiller: adequate. Pipework: inadequate. Solution: CDU manifold with secondary plate heat exchanger.

His second option was immersion cooling.

Servers submerged in a dielectric fluid — a non-conductive liquid that absorbed heat directly from every component surface. Single-phase immersion used a circulated fluid that remained liquid throughout. Two-phase used a fluid with a low boiling point that vaporised on the hottest surfaces, carrying enormous heat through phase change — the same principle as a refrigerant, but inside the tank with the servers.

Mark noted the option but didn't recommend it for retrofit. The capital cost was high — €1,500 to €3,000 per kW of IT load for a full immersion deployment. The operational complexity was significant: maintenance procedures changed entirely, because you couldn't slide a server out of a tray when it was submerged in fluid. The rack form factor changed. The floor loading changed. The fire suppression strategy changed. The monitoring approach changed.

He wrote: Immersion cooling offers the highest density potential (50–100+ kW/rack) and the lowest PUE contribution from cooling (<1.03). Not recommended for Clonshaugh retrofit due to capital cost, operational disruption, and the hybrid nature of the existing facility. Recommend as an option for Tallaght greenfield zones only.

Declan was at the door. He had read the tenant email and driven to Clonshaugh at seven on a Saturday morning, which told Mark everything he needed to know about how Declan felt about this opportunity.

"Can we do it?" Declan said.

"We can do it. Not with air."

"What then?"

"Liquid. Direct-to-chip. CDUs on the hall floor, pipework from the existing chiller, cold plates on the GPUs."

Declan processed this. He had spent ten years managing air-cooled systems. Fans, ducts, raised floors, containment curtains. He understood air. Air was visible — you could feel it, hear it, see the curtain sway when the CRAH changed speed. Liquid was different. Liquid was invisible inside pipes, silent, pressurised. A leak in an air system was a draught. A leak in a liquid system was a flood.

"You'd put the servers in a bath?" he said.

Mark almost smiled. "That's immersion. I'm not recommending immersion — not here. I'm recommending cold plates. The liquid stays inside a sealed loop. The servers stay in conventional racks. You add a coolant connection to each rack alongside the power connection. Think of it as plumbing, not swimming."

Then the floor.

Mark called the structural engineer at ten o'clock on a Saturday morning. He apologised for the timing. The structural engineer — a colleague from a previous project, Dublin-based, available because structural engineers who work on data centres are accustomed to weekend calls — pulled up the Clonshaugh structural drawings from the original planning file.

"Raised floor is rated for 12 kN/m² distributed and 3 kN point load per tile," the engineer said. "That was standard for 2013 IT equipment at 6 kW per rack. A loaded 6 kW rack with servers, cabling, and PDUs is approximately 500 to 800 kilograms."

"What does a loaded 30 kW GPU rack weigh?" Mark asked.

"With full GPU population, power supplies, NVLink interconnects, and the server chassis? 1,200 to 1,800 kilograms. Depending on the manufacturer and configuration."

Mark wrote the numbers. Rack weight at 30 kW: up to 1,800 kg. Floor tile rating: 3 kN point load — approximately 300 kg per tile corner. Four corners per rack, but the load concentrated at two front and two rear pedestals.

"The floor won't take it," the engineer said. "Not without reinforcement. You need either supplementary steelwork under the raised floor — cross-bracing between pedestals — or you set the GPU racks on a slab-mounted plinth that bypasses the raised floor entirely."

Mark thanked her and hung up. He looked at Declan, who had been listening to the call with the expression of a man whose jaw was tightening slightly.

"The floor," Declan said.

"The floor."

By noon, Mark had four pages. A feasibility note — not a report, not a design, but a structured assessment of what was possible, what was required, and what it would cost.

Option A: Direct-to-chip liquid cooling (recommended)

- 40 racks × 30 kW, CDUs on hall floor

- Chiller plant: adequate (Phase 1 retrofit)

- Pipework: new CDU manifold + secondary PHX. €120–180K.

- CDUs: 4 units at €40–60K each. €160–240K.

- Cold plates: server-specific, tenant responsibility (GPU vendor supply)

- Leak detection: mandatory under every CDU and rack. €20–30K.

- Floor reinforcement: supplementary steelwork or slab-mounted plinths for 40 positions. €60–100K.

- Power: busway upgrade for 40 positions at 30 kW. €80–120K.

- Total facility-side capex: €440–670K (excluding tenant IT equipment)

- Revenue: 40 racks × 30 kW × tenant rate (significantly above 6 kW rate)

Option B: Immersion cooling (noted, not recommended for retrofit)

- Higher density ceiling (50–100+ kW/rack)

- Capital cost: €1,500–3,000/kW installed

- Operational complexity: high

- Recommended for Tallaght greenfield zones only

Option C: Decline the tenant (counterfactual)

- Revenue foregone: 1.2 MW at high-density rates

- Competitive risk: tenant goes elsewhere

He wrote the summary line: The facility can accommodate 30 kW/rack density in a dedicated hybrid zone. The infrastructure cost is material but bounded. The revenue opportunity is significantly higher per rack than current tenancy. Recommend proceed to detailed design.

Ann arrived at two o'clock. She had read the feasibility note on her phone driving up from Dublin. She had one question, and she asked it standing in the doorway of the conference room, before she sat down.

"If we do this, what happens to our PUE?"

Mark had been waiting for this question. He had a careful answer, because the answer was not simple and Ann deserved precision.

"PUE improves," he said. "At 30 kW per rack, 60 to 70 percent of the heat is captured by the liquid loop and rejected at a higher temperature than air cooling can achieve. That means the chiller runs at a higher COP — less electricity per unit of cooling. The residual air cooling handles less load. Total cooling energy per kW of IT goes down. PUE improves."

"By how much?"

"For the 40-rack zone, the cooling PUE contribution drops from approximately 0.30 to 0.08. For the whole facility — blended — PUE might drop from 1.30 to 1.22 or lower, depending on the mix."

"Good. But?"

Mark nodded. She was learning to hear the but.

"PUE alone stops telling the whole story when your cooling medium changes. The right metric for a liquid-cooled zone is TUE — Thermal Utilisation Effectiveness. It captures the efficiency of the liquid loop, the CDU, and the heat rejection path as a single number. PUE measures the building. TUE measures the cooling technology within the building."

Ann wrote it down. TUE. Ask Sarah how it fits into EED reporting.

Declan had been reading the feasibility note. He was on page three — the floor loading section — and his jaw had returned to its tightened position.

"How much?" he said.

Mark looked at the cost summary. "Facility-side: four hundred and forty to six hundred and seventy thousand. That's pipework, CDUs, leak detection, floor reinforcement, and power distribution. It doesn't include the tenant's IT equipment or the cold plates — those are the GPU vendor's scope."

Declan looked at Ann. Ann looked at the number. She looked at it the way Tom would look at it — per kilowatt, not per rack, not per square metre.

€440–670K for 1,200 kW of new IT capacity. That was €367–558 per kW of IT installed — facility-side only. The revenue per kW at 30 kW density was three to four times the revenue at 6 kW density. The tenant was paying a premium for density that the market was demanding and the building could now provide.

"Tom needs to see this," she said.

"Tom's not here," Declan said.

"I know. I'm calling him now."

She picked up her phone. The feasibility note was four pages. Tom would turn it into a business case by Monday. The business case would go to the fund committee by Wednesday. The tenant meeting was Thursday.

The plan hadn't finished and the target had already moved. The building that started at 6 kW per rack — a 2013 design for a 2013 market — was looking at a future where density was the revenue driver and cooling technology was the enabler.

Declan stood at the door of Hall A. Same raised floor. Same ceiling height. Same cable trays and busway overhead. The hall looked the way it always looked. But underneath the raised floor, the pedestals were rated for 800 kilograms. And in forty positions, the future wanted 1,800.

The building hadn't changed. What it needed to become had.

Ann, phone to her ear, waiting for Tom to answer, looked at Declan. She smiled — the first genuine smile in the book, not relief, not satisfaction, but the particular expression of someone who has just realised that the problem she's been solving has become an opportunity she didn't expect.

"That's not a problem," she said. "That's a market."

Six months.

Six months since Ann had driven to Clonshaugh for the first time and parked in a car park that told her nothing. Six months since Declan had handed her a hard hat and said "You'll want this on the roof later." Six months since she'd touched the transformer casing and asked if it should be warm, and Declan had answered by explaining the first law of thermodynamics without calling it that.

They were in the conference room. All five of them, for the last time — or the last time at Clonshaugh. There would be other rooms, other buildings, other meetings. But this was the room where the programme had been built, and the programme was about to be tested.

Sarah's calendar covered the west wall. It was battered now — six months of marker, tape, sticky notes, and annotations in four colours. Some deadlines had green ticks. Some had amber flags. None were red. The zone map was in the corner, upgraded to v0.3 — Mark's pencil annotations incorporated, Declan's additional cooling zones added, the whole thing redrawn by a draughtsperson and returned as a proper engineering document with a revision block and a title border.

Tom's whiteboard had been cleaned and rewritten so many times that the surface was permanently ghosted. The current version showed the programme budget — actual versus forecast, phase by phase, with a variance column that Tom updated every Friday.

The fund's legal counsel was on the screen. Joining from London. She had a single question, and the entire meeting existed to answer it.

"Is this facility compliant — yes or no?"

Sarah answered. She had been preparing for this moment since the brown paper went up on the wall, and she answered the way she answered everything — precisely, with sources, and without overclaiming.

"Clonshaugh is a Tier 2 facility under CRU/2025236," she said. "Five MVA MIC, 2.4 megawatts IT load, ESB Networks medium voltage distribution connection. Tier 2 obligations are: 80 percent renewable energy procurement, EED Article 12 reporting, and — if the facility exceeds the dispatch threshold — dispatchable generation capability."

She opened her laptop. The compliance dashboard — the dashboard that existed because of the monitoring infrastructure she had specified in Chapter 9, because of the sensors Declan had helped commission in Chapter 13, because of the metering Mark had recommended in Chapter 4 — showed every metric in real time.

PUE: 1.32, trailing twelve-month average. Category 2 metered. Trending toward 1.20 as the free cooling plant delivered its first full summer. The Taxonomy threshold was 1.3. They were close — not compliant yet on the twelve-month annualised figure, but the trajectory was right.

Renewable energy factor: 82%. Corporate PPA with a wind project in Clare, contracted for ten years. 80% obligation met. The remaining 18% was grid default supply at the prevailing renewable share.

EED Article 12: Submitted. First verified report filed with SEAI, on time, with metered data at EN 50600-4-2 Category 2. PUE, WUE, CUE, REF — all four metrics reported. No compliance notices. No enforcement action.

F-Gas: Hall B fire suppression converted from FM-200 to Novec 1230 — GWP reduced from 3,220 to 1. R-410A condenser replacement scheduled as part of Phase 2 cooling retrofit — timeline on the calendar, budget in the programme. F-Gas registry updated.

Carbon tax exposure: Reduced from €502K/yr to approximately €340K/yr through PUE improvement and PPA. At €100/tCO₂ in 2030, the programme saves approximately €230K/yr in carbon tax versus the do-nothing scenario.

CRREM Misalignment Year: Pushed from 2031 (do-nothing) to approximately 2039 (current trajectory). Full programme completion targets 2045+. Disclosed as LBE-derived DC pathway bands, Tier 3/4.

Legal counsel looked at the screen. She had a pen in her hand and a legal pad. She had written three things.

"The PUE," she said. "It's 1.32. The Taxonomy threshold is 1.3. You're above it."

Sarah nodded. "By 0.02, on the trailing twelve-month average. The free cooling plant has been operational for four months. The annualised figure will cross 1.3 within the next reporting period. The trajectory is demonstrated in the metered data."

"Can you guarantee it will be below 1.3?"

"No. PUE is a metered metric that varies with load, ambient temperature, and operational decisions. I can guarantee the infrastructure is in place to achieve it. I can demonstrate the trajectory. I cannot guarantee a number that depends on physics and weather."

Mark, from his chair at the back, said quietly: "Nobody can guarantee PUE. The Delegated Act requires annualised metered data, not a guarantee. The infrastructure to achieve ≤1.3 is installed and commissioned. The data will demonstrate compliance."

Legal counsel wrote something down.

Tom presented the financial summary. He stood up — the boardroom habit — and spoke to the screen.

"Total programme expenditure to date: €2.1 million of the €3.8 million approved. Phases 1 and 2 complete. Phase 3 — grid connection and energy centre — underway, with ESB Networks design stage at week fourteen of an estimated twenty-four-week programme."

He listed the returns.

"PUE energy saving: €620K per year, actual, metered, from twelve months of utility bills compared to the pre-programme baseline. Carbon tax reduction: €162K per year at current rates, rising with the legislated escalation. PPA saving versus blended rate: €380K per year. Total quantified annual benefit: €1.16 million per year against a programme that is sixty percent complete."

He paused. "The programme has already paid back €1.16 million against €2.1 million spent. At this run rate, full payback occurs in twenty-two months from today — which is before Phase 3 completes."

Legal counsel asked: "And the density tenant?"

Ann took this one. "Forty racks at 30 kilowatts. 1.2 megawatts. Direct-to-chip liquid cooling in a dedicated zone. Facility-side capex approved at €550K. Tenant lease signed last Thursday. Revenue per rack: four times the current average. First rack energised in eight weeks."

Mark presented the engineering summary. He did it in one page — a single sheet, landscape, twelve items in a table. Not a report. A programme.

#	Item	Status	Evidence
1	EED Article 12 metering infrastructure	Complete	Cx report Phase 1, SEAI submission
2	PUE baseline established (Category 2)	Complete	12-month trailing data
3	Containment — all racks	Complete	Cx report Phase 1
4	Free cooling plant	Complete	Cx report Phase 2
5	Setpoint optimisation (18°C → 23°C)	Complete	BMS log, 30-day monitoring
6	Fire suppression — Hall B (FM-200 → Novec)	Complete	Cx report, F-Gas registry
7	80% renewable PPA	Active	10-year CPPA, Clare wind
8	Generator fuel contract upgrade	Active	4-hour guaranteed delivery
9	Grid connection upgrade (38 kV)	In progress	ESB Networks design stage, wk 14/24
10	Energy centre — BESS + CRM qualification	In progress	CRM pre-qualification submitted
11	Liquid cooling zone — GPU tenant	In progress	CDU manifold installed, first rack wk 8
12	Condenser replacement (R-410A phase-out)	Scheduled	Phase 3 Q2, budget approved

He laid the page on the table. "Twelve items. The first one's done. That's how every programme starts."

Declan looked at the table. Items 1 through 6 — six green ticks. He had signed the witness reports on four of them. His name was on the commissioning evidence for the CRAHs, the monitoring, the fire suppression, and the containment. His knowledge — the knowledge that had lived in his head for ten years — was now in files, in reports, in a programme that would survive his retirement.

Legal counsel asked her question again. The same question, reframed.

"So is this facility compliant — yes or no?"

Sarah looked at the screen. She looked at Ann. She looked at the calendar on the wall — battered, annotated, six months of deadlines and decisions and green ticks accumulating slowly, like evidence.

"It is on a programme to be fully compliant," she said. "That's a better answer than yes."

Legal counsel paused. Then she said: "Explain."

"A yes means a snapshot. It means today, right now, this minute, every box is ticked. Some of our boxes are ticked. Some are in progress. The PUE is 1.32, not 1.30 — it will cross the threshold, but it hasn't yet. The condenser replacement is scheduled, not complete. The grid connection is at design stage, not energised." She paused. "But we have a programme. Every obligation is identified. Every obligation has a timeline, a budget, and an owner. Every completed item has commissioning evidence. Every in-progress item has a milestone date. And every metric is metered — not estimated, not assumed, metered."

She looked at the screen. "A yes is fragile. It's true on the day you say it and it may not be true next quarter. A programme is resilient. It tells you where you are, where you're going, and what happens if you fall behind."

Legal counsel wrote for thirty seconds. Then she looked up.

"That's sufficient for the position paper. Thank you."

She disconnected.

The room exhaled. Not visibly — nobody sighed, nobody slumped — but the particular tension of a meeting conducted for an audience that holds authority released, and the five of them were alone again.

Ann looked at the team. Declan with his hard hat on the table. Mark with his one-page programme. Sarah with her dashboard. Tom with his calculator.

"This started because I walked into a building I'd owned for six months and didn't understand," she said. "I thought it was a property asset. It's not." She looked at the calendar on the wall. "It's a machine, an energy asset, and a compliance programme. And it took five people to see it."

She paused.

"The clock hasn't stopped. It never stops."

The meeting broke up. Sarah packed her laptop. Tom closed his spreadsheet — version twelve now, three hundred and fourteen rows, every number earned. Mark rolled up his one-page programme and put it in the cardboard tube he'd brought back in Chapter 12 — the same tube, repurposed.

Declan stayed.

He walked the floor alone. Hall A. The rows of racks, the new CRAH units on the north wall, the hum that never stopped. The monitoring sensors — green LEDs every fourth rack position, logging, timestamping, feeding the dashboard that Sarah checked every morning from Dublin.

He stopped at Row 7.

He looked up at the containment curtain. The one that had been sagging since Chapter 2 — twenty centimetres of gap where the Velcro had separated from the rail, hot air leaking into the cold aisle, money escaping as heat.

Someone had fixed it. The Velcro was straight. The curtain hung properly. The hot aisle and the cold aisle were sealed.

He didn't know when it was fixed. He didn't know who did it. It might have been done during the containment installation — Phase 1, the commissioning he'd signed off in Chapter 13. It might have been done by a shift engineer who'd noticed it and fixed it without being asked. It didn't matter who. It mattered that it was done.

He thought about Mark's phrase from Chapter 2: Common and normal aren't the same thing.

Six months ago, a sagging curtain was common. Today, it was fixed. That was the difference between a building that ran and a building that was managed.

He turned off the hall light. The servers kept running, the cooling kept running, the monitoring kept running. The building didn't need the light. It needed what it had always needed — power, cooling, and someone who understood that keeping it alive and proving it worked were not the same thing.

He went to the maintenance log. He wrote:

Containment curtain, Row 7, Hall A. Inspected. Repair confirmed. All rows checked. All clear.

Ann was in the car park. The same car park where she'd arrived six months ago, where Declan had handed her a hard hat, where Mark had driven out after the cooling survey and after the grid meeting and before he came back with drawings.

The building looked the way it always looked. Grey cladding. Faded sign. Security barrier. The same industrial estate in north Dublin that you drove past on the M50 without a second thought.

But she could trace the cable route now. She could see the transformer pad behind the fence. She could see the new dry cooler arrays on the roof — Phase 2, commissioned, running. She could see the BESS enclosure in the generator compound — Phase 3, installed, awaiting CRM qualification. She could see the ESB temporary works compound at the site boundary — the 38 kV substation, under construction.

The building hadn't changed from the outside. Everything had changed inside. And everything that had changed was measured, documented, and proven.

She picked up her phone. She looked at Mark, who was loading the cardboard tube into his car.

"I need to talk to you about Tallaght," she said. "Twelve hundred racks. Ten megawatts. Thirty-eight kV."

Mark opened his notebook.

"When do we start?"