The Price of Compute That Never Turns On

Microsoft’s cancellation of Nscale’s data center in Essex reveals what capital alone cannot buy. Nscale had planned 23,040 Nvidia GPUs at its Loughton site by early 2027, with 50 megawatts of initial capacity expanding to 90. The project anchored a reported $24 billion contract with Microsoft and supported the company’s $35 billion IPO target. U.K. Grid officials said power would not reach the site until the 2030s, so Microsoft walked, putting that IPO at risk, as SiliconANGLE reported on the Nscale deal.

Nscale’s relationship with capital looks nothing like its relationship with electricity. The company signed the Essex deal on a market price for compute that assumed power would materialize, permits would clear, and construction timelines would hold. Ludwig von Mises described this dynamic when he observed that one and the same piece of metal can even fulfil both purposes simultaneously. Ingots leave the vaults of banks for the workshops of goldsmiths when industrial demand bids the price high enough. Ornaments pass to the mint when the bullion price makes monetary use attractive. The price coordinates the flow, not the physical possibility at any single site.

The same force now drives AI infrastructure investment. A data center cannot be built where the grid will not serve it, the permit office will not approve it, or materials will not be delivered. Capital does not compress those constraints. What it does is bid up the price of compute that exists mostly on spreadsheets and in press releases. Every capex announcement, every valuation, every demand forecast draws money on its own momentum, pulling investment toward projects regardless of whether any given site can physically deliver them.

The market prices capacity it will never turn on. The winners hold not the largest checks or the highest valuations, but the land parcels where the power company already says yes and the permit is already signed. The premium accrues to sites where steel meets soil and electricity flows, not to the projects that look best on a balance sheet.