Bernstein: OpenAI's compute infrastructure spending through 2030 may reach $614 billion, below widely cited $750 billion

November 8 — according to Zhitong Finance, Bernstein analyst Mark Moerdler estimates that infrastructure spending corresponding to OpenAI's disclosed compute partnership agreements could cumulatively reach $614 billion by 2030, well below the widely cited $750 billion figure. The estimate assumes all identified contracts are executed as planned, but actual spending could be reduced or delayed by data center construction delays, project cancellations and adjustments to hardware procurement arrangements. Because the relevant compute is typically delivered in batches, costs will be incurred gradually as facilities come into operation rather than creating an immediate payment obligation, and annual spending is expected to peak at $184 billion in 2031.

Bernstein noted that OpenAI's agreements with Oracle, Microsoft and Amazon are worth roughly $300 billion, $250 billion and $138 billion respectively, typically with terms of five to eight years; the agreement with CoreWeave has a cap of $22.4 billion but does not represent guaranteed revenue. The firm believes large cloud service providers face limited contract risk in the near term, and that the real concern is that once the future tightness in compute supply eases, specialized compute providers that rely on high-cost financing and continued leasing demand could come under greater pressure.