Capex Divergence: Microsoft Cancels Its First Committed AI Datacenter Project as Grid Constraints Bite
The hyperscaler capex buildout (MSFT $35.8B, AMZN $54.21B, GOOGL $44.92B in the June quarter alone — see The $3 Trillion Shadow Ledger: Off-Balance-Sheet AI Commitments Dwarf Reported Debt) just produced its first public cancellation: on October 18, Microsoft pulled out of Nscale's Loughton, Essex datacenter, where it was the anchor tenant. Per the Guardian: "Microsoft, one of its largest planned customers and expected to have been the anchor tenant, informed the company that it was pulling the plug on its agreement." The project had been heralded by UK ministers as the country's largest AI datacenter.
The reason is power, not money. UK Power Networks "would not be able to supply enough power to the company's Loughton datacentre until the early to mid-2030s"1 — a reminder that grid interconnection, not capital, is becoming the binding constraint on the AI buildout (see the Texas interconnection-moratorium watch). Nscale lost its only publicly announced customer for the Essex site; Microsoft "remains on board for another project the company has in Norway."
The blast radius is the neocloud financing chain. Nscale filed for a US IPO last month seeking up to $3B at a potential $35B valuation, "as soon as this month" — and per its IPO filing, "Microsoft and Anthropic account for approximately 85% of Nscale's contracts." Losing the anchor tenant weeks before pricing puts the listing at risk and sharpens what the Guardian calls "a growing chorus of voices" comparing debt-laden neoclouds to "a trillion-dollar WeWork 2.0." The market is already pricing this risk: on Friday Oct 16, CoreWeave fell 11.9%, Nebius 12.2%, and Super Micro 9.9% while Oracle — down 9% on the week on AI-debt concerns — erased $15B of Larry Ellison's stake value.
Why this matters for the basket: capex divergence now has a selectivity dimension. Microsoft is willing to walk away from a signed, politically prominent commitment when execution fails — evidence that hyperscaler capex is not an unconditional escalator. That reframes the off-balance-sheet commitment ledger: the $3T of contracted future spend is only as good as power and execution. Watch whether MSFT's Q3 call (late October) addresses capex pacing, and whether the Nscale IPO prices, slips, or pulls (registered as a watch this cycle).
-
An instance of Grid congestion has replaced code optimization as the primary gatekeeper of artificial intelligence scale. — Microsoft's first cancellation of a signed, politically prominent datacenter commitment proves grid interconnection, not capital, is the binding constraint on AI buildout scale. ↩︎