Firms like OpenAI and Meta claim a compute shortage while also exploring selling compute capacity. This isn't a contradiction but a strategic evolution. They are buying all available supply to secure their own needs and then arbitraging the excess, effectively becoming smaller-scale cloud providers for AI.

Financing large-scale compute infrastructure requires treating compute like a utility rather than backing a single tenant model. Pooling different labs and public-sector entities smooths out demand volatility and diversifies counterparty risk for institutional debt providers.

Instead of directly funding AI data centers, India's national AI mission uses a demand-side strategy. It subsidizes compute access for users like startups and researchers, creating a guaranteed market that incentivizes private companies to build and offer compute capacity competitively.

The advertised per-hour GPU cost is misleading. Because research workloads are spiky and unpredictable, labs over-provision compute. This rampant underutilization means the effective price paid is often 10 times higher than the marketed rate, creating massive deadweight loss.

Escalating compute requirements for frontier models are creating a new market dynamic where access to the best AI becomes restricted and expensive. This shifts power to the labs that control these models, creating a "seller's market" where they act as "kingmakers," granting massive competitive advantages to the highest corporate bidders.

For leading AI labs like Anthropic and OpenAI, the primary value from cloud partnerships isn't a sales channel but guaranteed access to scarce compute and GPUs. This turns negotiations into a complex, symbiotic bundle covering hardware access, cloud credits, and revenue sharing, where hardware is the most critical component.

A major shift in behavior among top AI labs is their move from three-year to five-year take-or-pay contracts for GPU infrastructure. They are locking in capacity at massive scale for longer durations, signaling extreme confidence in sustained, long-term demand for compute.

The question of who pays for large-scale open source model training has a clear answer: chip manufacturers. For companies like NVIDIA, funding a multi-billion-dollar training run is a negligible marketing expense to fuel the ecosystem and drive massive, high-margin hardware sales.

Well-capitalized commercial AI labs can expend immense compute budgets to rush solutions to famous conjectures ahead of university researchers. Meanwhile, academic departments face funding pinches where even providing monthly $200 cloud accounts for PhD students competes with hiring human talent.