The AI boom has a hidden bottleneck: electricity. And it's not just a data centre problem—it's a mirror for crypto's own scaling challenges.
CoreWeave, the GPU cloud provider that pivoted from crypto mining to AI, is staring at exactly this crisis. Its £8.2 billion data centre project in Scotland faces power supply concerns. The headline is neutral. The undertow is a warning for every builder betting on centralised scale.
I’ve been here before. In 2017, during the ICO bubble, I audited 150 whitepapers. Back then, the hype was about trustless protocols. Now, the hype is about AI compute. But the pattern repeats: we assume infrastructure will magically scale to meet demand. It won’t.
When I wrote my thesis ‘Code as Covenant,’ I argued blockchain wasn’t just a database—it was a mechanism for enforcing trustless social contracts. CoreWeave’s Scottish struggle is a covenant failure. The code (the GPU clusters) is ready. The community (the grid, the regulators, the local residents) is not.
Let’s unpack the technical reality.
CoreWeave plans a 500MW to 1GW data centre. That’s the power draw of a small nuclear reactor. Scotland has plenty of wind—over 13GW installed—but transmission capacity is the choke point. The grid in northern Scotland, managed by SSEN, can’t simply absorb a 500MW load. Upgrading transmission lines takes 6 to 9 years in the UK. CoreWeave’s project timeline? Probably 2 to 3 years. The mismatch is foundational.
Power quality is another hidden tax.
GPU clusters are hypersensitive to voltage flicker and frequency drift. A 50ms glitch can corrupt a training run costing millions. Scotland’s grid is heavily renewable, meaning intermittent power and voltage instability. To compensate, CoreWeave would need dynamic voltage restorers or massive battery banks—adding hundreds of millions to CAPEX. The cost of stable power isn’t just the kWh price; it’s the engineering required to make the grid dance in sync with H100s.
During DeFi Summer in 2020, I watched yield farming protocols exploit users through opaque incentive structures. I resigned because I couldn’t stomach the moral dissonance. Now I see a similar dissonance: AI companies claim to be building the future, while their power demands could push local communities into energy poverty or force new gas peaker plants. The narrative of progress hides the externalities.
The commercial stakes are brutal.
CoreWeave’s edge over AWS and Azure is price—about 30% lower. That margin comes from cheap power. If Scottish electricity becomes unreliable or expensive (due to capacity upgrade fees or renewable intermittency), the unit economics break. A 20% rise in power cost wipes out most of their profit advantage. Investors understand this. The article notes ‘investor confidence’ is shaken. I’d go further: this single site risk could slash CoreWeave’s IPO valuation by 20-30%.
I mentored junior developers during the 2022 bear market. One lesson I drilled: never build a castle on a single foundation. CoreWeave’s Scottish bet is the ultimate single point of failure. If the grid doesn’t deliver, the entire revenue stream for that site collapses. That’s not resilience; it’s hubris.
Now, the contrarian angle everyone misses.
Most analysis says the bottleneck is chips—NVIDIA’s supply constraints. I disagree. The bottleneck is grid capacity. The AI industry is so focused on compute efficiency (flops per watt) that it forgot the supply side: watts per location. As data centres scale to gigawatt levels, they compete with cities for power. The result will be a ‘grid crunch’ that forces AI companies to either pay astronomical premiums for ready capacity or delay projects indefinitely.
This is where crypto’s decentralised compute model offers a structural escape.
Platforms like Akash Network, Filecoin (with its retrieval market), and even Ethereum’s staking infrastructure prove that you can distribute compute across thousands of independent nodes. No single power outage kills the network. No single grid upgrade holds back progress. The covenant between code and community is enforced by geographic dispersion, not by a central power purchase agreement.
Bulls react. Bears reflect. We build.
But building centralised AI datacentres is just reactive scaling—throwing money at the same old model. The reflective builder asks: what if the future isn’t a few monolithic datacentres but a mesh of sovereign, power-diverse nodes? Each node could tap local renewables, battery storage, or even behind-the-meter industrial power. The network becomes antifragile.
I retreated to a cabin in rural Virginia during the 2022 crash. I spent 400 hours rereading Hayek and Turing. Hayek’s insight against central planning applies here: no single entity can know the local conditions for power supply everywhere. A centralised AI cloud decides where to build based on a few parameters. A decentralised compute network lets thousands of participants optimise locally—the ultimate Hayekian solution.
Tech changes. Values remain.
The value that guides me is sovereignty—the ability of individuals and communities to control their own infrastructure. Centralised AI datacentres concentrate power (both electrical and economic) in the hands of a few. Decentralised compute distributes it. CoreWeave’s Scottish struggle isn’t just a business hiccup; it’s a philosophical case study.
If we keep building in the shadow of the grid, we will always be hostage to transmission lines, regulatory approvals, and NIMBY campaigns. If we build with the grain of decentralization—small, local, redundant—we free compute from geographical tyranny.
The takeaway is not to short CoreWeave. It’s to recognise that the next golden age of compute will be defined not by how many GPUs we hoard, but by how intelligently we distribute power demand.
I run a crypto education platform now. We teach people to think in first principles. The first principle here is simple: resilience comes from distribution, not from size. CoreWeave’s £8.2 billion bet is a bet on size. The grid is saying ‘no.’ The market is starting to listen.
Verify the code, trust the community. The code for decentralised compute is already written. The community—global, diverse, voluntarily contributing power—is the only grid that can scale without breaking.
The Scottish hills won’t be covered in data centres. They’ll be covered in wind turbines. And the compute will be woven between them, node by node, not block by block.
That’s the future I’m building toward. Not a single £8.2B castle. A thousand sovereign cabins.
Bulls react. Bears reflect. We build.