- Published on
Bernstein: OpenAI's compute infrastructure spending could reach $614 billion by 2030, below earlier $750 billion rumor
- Authors
- Name
- DeepHorus
Bernstein analyst Mark Moerdler estimates that the infrastructure spending corresponding to OpenAI's disclosed compute cooperation agreements could cumulatively reach $614 billion by 2030, equivalent to about RMB 4.11 trillion. That figure is notably lower than the $750 billion previously reported by The Wall Street Journal. The market therefore gains a new reference point for evaluating OpenAI's long-term compute investment.
The estimate assumes that all identified contracts are executed as planned. Bernstein cautions that actual spending may be reduced or delayed because of data center construction delays, project cancellations, and adjustments to hardware procurement. Compute is usually delivered in batches, with costs incurred gradually as facilities are commissioned, rather than immediately creating payment obligations.
In terms of annual spending, Bernstein expects related investment to peak at $184 billion in 2031, equivalent to about RMB 1.23 trillion. This means the infrastructure bill will not be concentrated in the near term. The market's focus is shifting from total contract value to actual delivery progress.
By partner, OpenAI's agreements with Oracle, Microsoft, and Amazon are approximately $300 billion, $250 billion, and $138 billion, respectively, usually with terms of five to eight years. Its agreement with CoreWeave has a cap of $22.4 billion but does not represent guaranteed revenue. The above contracts form the main framework for its future compute spending.
Bernstein believes that large cloud service providers face limited short-term contract risk. The real concern is that after future tightness in compute supply eases, specialized compute suppliers that rely on high-cost financing and continuous leasing demand may come under greater pressure. The focus of risk thus shifts to suppliers' financing and leasing models.
From the perspective of industry impact, OpenAI's long-term compute bill remains high, but the $614 billion estimate is somewhat lower than earlier rumors. The pace of data center construction, hardware procurement arrangements, and project execution will determine actual spending. The divergence between cloud vendors and specialized compute suppliers may also deepen.
Looking ahead, Bernstein's estimate provides a baseline scenario for observing OpenAI's infrastructure expansion. If contracts are executed on schedule, spending will peak around 2031; if construction or procurement is delayed, spending may be pushed back or contract. The market will focus on facility commissioning progress and contract execution details.
