Bernstein projects OpenAI infrastructure spending to hit $614b by 2030, lower than prior estimates

Bernstein analyst Mark Moerdler estimates that the infrastructure spending associated with OpenAI's disclosed compute agreements could total $614 billion by 2030. This projection is notably lower than the $750 billion figure previously reported by The Wall Street Journal. Moerdler noted that while the calculation assumes all identified contracts proceed as planned, actual expenditures might be reduced or delayed due to factors like data center construction postponements or hardware procurement adjustments. The firm anticipates that annual spending will peak at $184 billion in 2031, as costs are tied to facility operationalization rather than immediate payment obligations. The analysis details several key agreements: contracts with Oracle, Microsoft, and Amazon are valued at approximately $300 billion, $250 billion, and $138 billion, respectively, typically spanning five to eight years. Additionally, the agreement ceiling with CoreWeave is set at $22.4 billion, though this does not represent a guaranteed minimum. Bernstein further commented that while large cloud service providers face limited short-term contractual risks, the greater concern lies with specialized compute suppliers who rely on high-cost financing and continuous leasing once the future supply crunch eases.