The math doesn't add up. A single AI datacentre demanding up to one gigawatt of power. That is the equivalent of a small nuclear reactor. CoreWeave's £8.2 billion investment in Scotland sounds like a bet on the future. But the future has a present problem: there is not enough juice in the grid to run it.
I have spent years dissecting smart contracts. I trace each function call, every gas cost, every edge case. Now I am looking at a different kind of contract—the physical infrastructure that powers the AI boom. The same principle applies: trust the code, verify the trust. In this case, the code is the power purchase agreement, the grid connection, the transformer lead time. And the trust is breaking.
Context: The Promise vs. The Reality
CoreWeave is a GPU cloud provider. It started as a crypto miner, then pivoted to AI compute. Its selling point: cheap, available Nvidia H100s. The Scottish datacentre is its largest project yet. 82 billion pounds. Located near abundant wind power. The narrative is seductive: green energy, low cost, high performance.
But the details whisper a different story. The article—thin as it is—highlights "power supply concerns." That vague phrase hides a minefield. Grid capacity bottlenecks. Intermittent renewables. Community backlash. Investor jitters. All of it tied to one thing: electricity.
Core: The Technical Underbelly
Let me break it down. A modern AI GPU cluster draws 40 to 100 kilowatts per rack. That is ten times a traditional datacentre rack. For a gigawatt-scale facility, the grid must deliver a steady, high-quality feed. Scotland's northern grid, managed by SSEN, is not built for that. The existing transmission lines are already strained. Upgrading them takes five to nine years of regulatory wrangling. CoreWeave cannot wait that long.
The power consumption of an AI datacentre is not just the GPUs. Cooling adds another 20%. Interconnects, storage, lighting—it adds up. The typical operating cost breakdown shows electricity at 30-40%. If the power price spikes due to grid constraints or a forced switch to diesel backup, the entire business model cracks. The low-price promise evaporates.
And then there is the quality. Wind power is intermittent. A cloud passes; the turbines slow. Voltage dips. Frequency wobbles. GPUs hate that. They generate errors, lost computations, corrupted data. To compensate, you need large batteries or flywheels. More cost. More complexity.
From my audit experience, I know that every line of code has a hidden assumption. Here the assumption is that cheap, stable power will always be available. It is a bug in the whitepaper. The real world has no infinite gas limit.
Contrarian: The Hidden Costs of "Green" Compute
The prevailing narrative says AI datacentres can be green because they use renewable energy. That is half-truth. Yes, Scotland has wind. But the datacentre will not connect directly to a wind farm. It connects to the national grid. When the wind stops, the grid pulls from gas or coal. The carbon footprint is real, even if the power purchase agreement says "100% renewable." This is accounting, not physics.
Moreover, the datacentre competes with local communities for the same electricity. If the grid needs upgrading, costs are socialized. Residential bills rise. And the jobs created? Maybe a few hundred for a facility that consumes as much power as a small city. The trade-off is rarely discussed.
Security is not a feature; it is the foundation.
Power is security for a datacentre. Without it, the compute is dead. Investors are starting to realize this. CoreWeave's IPO ambitions hinge on this project. A delay or scaling back would slash its valuation. The math doesn't add up: a project that cannot get the power it needs is a liability, not an asset.
Takeaway: A Forewarning for Blockchain
This is not just an AI story. It is a story for every high-energy decentralized network. Blockchain validators, Layer-2 sequencers, and DeFi protocols running on heavy hardware face the same constraints. The blockchain trilemma—security, scalability, decentralization—has a fourth dimension: energy. We ignore it at our peril.
Trust the code, verify the trust. But also verify the transformer. The grid. The kilowatt-hour price. The next bull run will be fueled by more than hype. It will be fueled by electrons. And those electrons are not as cheap or as green as the brochures claim.
A bug fixed today saves a fortune tomorrow. The bug here is the assumption that power will simply show up. It won't. Not in Scotland. Not anywhere without a plan.