Bernstein Analyst Puts OpenAI Compute Spending at $614 Billion by 2030

Bernstein analyst Mark Moerdler estimates OpenAI’s disclosed compute deals would drive about $614 billion in infrastructure outlays by 2030, below the $750 billion figure widely cited in markets.

The projection assumes every identified agreement proceeds as planned. Delays in data-center construction, canceled projects or changes to hardware purchases could reduce the total or push spending into later years.

The $614 billion estimate is cumulative, rather than an immediate payment obligation. Computing capacity is delivered in stages, with costs arising as hardware arrives and facilities begin operating. Annual spending is projected to peak at $184 billion in 2031, even though the cumulative estimate covers spending through 2030.

OpenAI’s agreements with Oracle, Microsoft and Amazon are valued at about $300 billion, $250 billion and $138 billion, respectively, and generally run for five to eight years. The agreement figures describe deals with multiyear terms; they do not mean the full amounts are due immediately. OpenAI’s agreements with CoreWeave have a ceiling of $22.4 billion, but that maximum does not guarantee revenue for the provider.

Bernstein sees limited short-term contract risk for large cloud providers. Specialized computing providers may face greater pressure if supply constraints ease, given their reliance on costly financing and continued leasing demand.