Ireland’s grid hit its limit earlier than anyone expected. In the scramble that followed, data centers learned something policymakers already knew: nothing sparks innovation like running out of options.
In January 2022, EirGrid stopped issuing new data center connections in Dublin. No timeline. No exceptions. Just a freeze.
Over the previous decade, data center electricity use in Ireland had climbed 531%, according to the Central Statistics Office. By 2024, these sites were consuming 22% of the country’s electricity, more than every home in Ireland combined (Silicon Republic, 2024).
Dublin carried most of that load. More than 80 data centers were active nationwide, and many of the biggest campuses, Clonshaugh, Blanchardstown, and Tallaght, belonged to Amazon (Data Center Dynamics, 2025). Apparently Bezos didn’t get the memo about putting all your eggs in one basket.
The sector had added €7.3 billion to the economy. But it had also consumed the infrastructure that made that growth possible. That was the tradeoff.
Then the grid blinked. Everything paused. Crews walked away from half-finished buildings. Interconnection studies failed. Power requests that once took weeks now stalled for months, or came back marked “denied.”
Google’s 780,000-square-foot build in Grange Castle Business Park became the high-profile case. Capital was lined up. Design approved. Site ready. The deal collapsed over one missing input: electricity. The rejection letter from South Dublin County Council cited “insufficient capacity in the electricity network” (Irish Times, 2024). That one sentence signaled the end of business as usual.
There was no appeal. No workaround. The supply didn’t exist. So the hyperscalers changed course.
When the grid said no
When the grid says no, hyperscalers don’t lobby. They shop for reactors.
Microsoft secured 835 megawatts through a 20-year agreement with Constellation Energy, investing $1.6 billion to help restart Three Mile Island by 2028 (Constellation Energy, 2024). Amazon lined up more than 2,200 megawatts of nuclear-backed supply, including a 1,920 MW deal with Talen Energy through 2042 (Reuters, 2025). Meta followed with 1,121 MW from Constellation’s Clinton plant, set to begin in 2027 (Constellation Energy, 2025). Google committed to 500 MW from Kairos Power’s small modular reactors, with deployment expected between 2030 and 2035 (Google Blog, 2023).
You’d almost think they were planning a TED Talk. In a normal context, these would have been impressive climate gestures. But they were survival moves driven by immediate necessity. Procurement timelines were tight. Alternatives didn’t exist. What once lived in ESG decks now sat on the desks of grid engineers.
In Dublin, the operators who stayed began retrofitting what they already had. Microsoft reprogrammed UPS units to return electricity to the grid during peak demand (Microsoft News, 2023). Amazon turned waste heat from its Tallaght site into local heating, with the system projected to save 1,500 tonnes of carbon annually (About Amazon, 2024).
The moves weren’t coordinated, but the pattern was clear: secure energy upstream, optimize local infrastructure, and squeeze more value out of what was already built.
That shift was regulatory too. The EU’s 2023 Energy Efficiency Directive introduced enforceable targets for energy reuse and mandatory reporting for data centers over 500kW. Compliance costs ran high: €200,000 to €500,000 to start for small facilities, and well over €1 million for large sites, with ongoing annual costs in the hundreds of thousands (EU Directive Analysis, 2023).
Meanwhile, hardware lifecycles kept getting shorter. Equipment that used to run five years was now cycling out in three. The ITAD sector hit $25.31 billion globally in 2024, growing 8–12% annually (Grand View Research, 2024).
Recovery wasn’t a sustainability initiative anymore. It wasn’t optics. It was uptime. That’s what made it stick. It was the only way to keep building when new power wasn’t available.
The new rules of reuse
Power constraints changed how data centers behaved. Equipment that would’ve been decommissioned stayed in service. Heat was rerouted. Retired hardware found new roles inside constrained networks.
Lifecycle extension turned into a financial lever. In 2023, Microsoft stretched server lifespans from four to six years, saving $3 billion (Data Center Knowledge, 2024). Google reported the same figure. Amazon’s six-year refresh cycle is expected to save nearly $1 billion per quarter (Horizon Technology, 2024).
The economics worked at the component level, too. Third-party maintenance typically costs 40–60% less than OEM extended warranties. Refurbished servers retain 60–80% of their original performance at 30–50% of the cost of new equipment. And with Moore’s Law slowing, newer machines only offer 2–5% performance gains for larger workloads (Industry Research Analysis, 2024). Not exactly worth a forklift upgrade. So instead of buying new, they started working smarter with what they had.
A mid-sized enterprise with 1,400 servers might spend $7–14 million replacing them every five years, assuming $5,000–$10,000 per server (ITS ASAP, 2025). Extending those lifecycles by a year using recovered components and strategic maintenance could cut that to $6–12 million. The savings often cover grid resilience upgrades or mandatory EU compliance costs.
Operators began to improvise. It was the kind of ad hoc brilliance that gets you promoted… or audited.
Some pulled servers from retirement, stripped the fans, and modified them for immersion cooling. Instead of scrapping them, they redeployed them in smaller, unstable sites. These machines handled power fluctuations, ran cooler, and lasted longer, without needing replacements that couldn’t be sourced.
But scaling those workarounds revealed a deeper problem: for recovery to work, it had to be fast, precise, and secure.
In traditional data center decommissioning workflows, recycling often destroys the most valuable parts. In SSDs, the NAND memory chips carry both the cost and the carbon. Enterprise SSDs cost 5–8 times more than HDDs (Western Digital, 2024). Shredding removes not just risk, but the entire ROI. It’s like setting your wallet on fire to make sure no one steals it.
Targeted recovery flips that equation. It extracts NAND intact, rebuilds devices, and preserves performance. Companies like Reconext have created single-chain recovery systems with accredited erasure and no performance loss.
This approach is already operational. In constrained regions, access to verified reusable components keeps work moving. It closes procurement gaps when the supply chain can’t.
As rack densities increase and cooling requirements tighten, these recovery systems are the only reason some sites are still expanding. Disposal is now a compliance risk. Recovery is the infrastructure that lets you keep going.
The pattern is already spreading
Ireland crossed the 10% threshold in 2022, with data centers drawing more than a tenth of the country’s electricity. The moratorium followed soon after.
Other regions are moving along the same curve. Northern Virginia now leads the world. Data centers there consume 25.59% of the state’s electricity. North Dakota hit 15.42%, Iowa 11.73%, Nebraska 11.58%, and Oregon 10.24% (Quartz, 2024). Denmark projects 15% by 2030. Singapore expects to reach 12% by the same year, up from 7% today.
The timeline is accelerating. Virginia’s build-up took 3–4 years. Ireland followed within two. The next wave, North Dakota, Iowa, Nebraska, and Oregon, did it in less than 24 months.
Policy reactions are speeding up, too. In Amsterdam, restrictions kicked in December 2023, capping new data center growth at 670 MVA until 2030. Operators must now meet a PUE below 1.2 and capture waste heat (DLA Piper, 2024). Singapore, after a 2019–2022 moratorium, shifted to a competitive model – 80 MW was awarded in 2023, 300 MW in 2024, all with PUE ≤1.3 requirements (Lexology, 2024).
Each new constraint forces a change in strategy. Dense urban markets fall out of favor. Secondary regions with spare capacity rise in importance. Location choices follow power, not just market demand. You go where the grid lets you.
Thermal factors are now in the mix. Cooler climates boost efficiency. Cities with district heating become attractive. Some operators are modeling heat integration into their budgets either to offset carbon, reduce load, or meet local compliance rules.
And it’s not just data centers. Battery plants are being redesigned for material recovery. Construction teams are switching to modular methods that reduce dependency on global supply chains. Semiconductor factories are clustering near reliable water and power.
Meanwhile, the pressure on data centers keeps rising. Global data center capacity is on track to hit 84 gigawatts by 2027 (Goldman Sachs Research). AI workloads push infrastructure harder. Racks get denser. Cooling gets tighter. Every system is running hotter and faster. And no, they can’t just open a window.
That pressure is reshaping recovery. Systems need to shorten lead times, reduce supply risks, and return components to use faster.
Power is the first constraint. Everything else follows.
How readiness turns into growth
After the moratorium, waiting wasn’t an option. Operators had to move, and move fast.
The nuclear procurement wave that followed were direct responses to grid limits. Each move followed a physical constraint. No electricity, no expansion. Procurement moved from the margins to the center.
And that move changed the outcomes. When programs are optional, they stall. When infrastructure stalls without them, budgets get approved. The pressure focused the response. Nobody was optimizing. They were building systems that could survive the next wave.
Early movers kept building. They secured power, unlocked capital, and stayed ahead of the freeze. Others paused. They hit tighter constraints, with fewer paths forward. That gap is now defining who keeps moving.
Companies that adapt early get more options. They secure resources, stabilize operations, and extract value from systems already in place. Recovery becomes part of the infrastructure – just another way to stay in motion.
Ireland didn’t set out to become a test case for constraint-led innovation. But the grid made that decision for them. The steps taken across energy, hardware, and recovery now form a playbook others can use.
This pattern isn’t confined to Ireland. Texas, don’t look so smug. You’re next. Every sector that depends on physical infrastructure will face its version of this pressure. It might begin with energy. It might come from supply shortages, permitting delays, or compliance costs.
The source will vary. The sequence won’t.
The companies still building are the ones that adapted before they had to.
If recovery is starting to feel less like a sustainability play and more like core infrastructure, we should talk.
Not when it’s easy. Not when there’s time. But now, while the grid still likes you.
