Hyperscalers Outsource $12B Compute as Power Grids Tighten; Neocloud Model Validated

OpenAI's $12 billion CoreWeave deal and Nscale's $103.4 billion IPO contracts validate hyperscaler outsourcing of compute to specialized neoclouds, but power infrastructure emerges as the binding constraint on the $1 trillion capex cycle.

Hyperscalers are outsourcing GPU compute capacity at scale. OpenAI committed $12 billion to CoreWeave for inference infrastructure; Anthropic committed $11.6 billion to Akamai. Nscale's IPO filing disclosed $103.4 billion in contracted AI infrastructure demand—the highest visibility yet into committed neocloud capacity. CoreWeave alone signed $250 billion in new customer commitments in Q3. The economic logic is clear: building and operating GPU clusters costs more capital than paying a specialized neocloud to do it. Neoclouds have lower capex intensity per unit and absorb margin pressure hyperscalers cannot.

Pricing power is inverting in neoclouds' favor after two years of deflation. H100 rental rates bottomed at $3.38 per hour in 2026, but Nebius announced price increases of 17–21 percent across its lineup, effective October 1. The move signals supply tightening—the first material recovery since 2024. Nscale's IPO at ~$35 billion valuation despite ~$1 billion cumulative losses proves investor appetite for contracted demand. That pricing power is real and tightening.

Power infrastructure has become the binding constraint. Oracle invoked force majeure on Project Jupiter due to grid delays, stalling an $18 billion construction loan. The federal government responded with $1.9 billion in grid investment to unlock 23 gigawatts of data-center capacity. But OpenAI forecasts $856 billion in compute spending through 2030. Twenty-three gigawatts is a down payment, not a solution. Grid delays are now the primary bottleneck to hyperscaler capex, not capital or chip availability.

Rising capital costs compound the pressure. Bond yields hit 22-year highs this week. Rothschild Redburn downgraded CoreWeave and Nebius on leverage risk—a signal that neocloud financing is tightening even as contracted demand surges. The arbitrage only works if leverage remains cheap. A sustained shift in rates could force repricing or consolidation among overleveraged operators.

Watch: whether federal grid investment will scale linearly with projected demand, whether neocloud pricing holds firm (the Nebius move is the test), and whether rising rates force a reckoning on neocloud balance sheets. Hyperscalers have shifted capital risk to the neocloud layer—but that layer's viability depends on two things the market has always assumed: cheap financing and adequate power. Neither is assured.