Bernstein Sees OpenAI Compute Spending Reaching $614 Billion Before 2030

Bernstein analyst Mark Moerdler said OpenAI’s announced computing-power purchase commitments could translate into about $614 billion in infrastructure spending by the end of 2030. According to Sina Finance, that is below the widely cited estimate of $750 billion; Bernstein said investors may be overstating OpenAI’s near-term funding pressure and that direct risks to Microsoft, Oracle, and Amazon are relatively limited. Bernstein’s estimate assumes all confirmed contracts proceed as planned. Data-center delays, project cancellations, and flexibility in hardware purchases could reduce or defer some spending. Costs would accrue as data centers enter service rather than becoming payment obligations immediately. The firm expects OpenAI’s annual infrastructure spending to peak at $184 billion in 2031. OpenAI’s agreement with Oracle involves about $300 billion in computing capacity, expected to be delivered in phases over five to six years. OpenAI accounts for less than half of Oracle’s remaining performance obligations. OpenAI has also committed to purchasing about $250 billion in Azure cloud services from Microsoft, with capacity to be provided in stages over six years. Its agreement with Amazon totals about $138 billion over seven to eight years and includes computing capacity related to Trainium AI chips. The agreement with CoreWeave has a ceiling of $22.4 billion, but that amount does not represent guaranteed revenue. Bernstein said OpenAI’s revenue growth could help fund some of its expanding computing needs. As of September this year, its annualized revenue was reportedly close to $50 billion, up 77% year over year. As computing supply constraints ease, industry risks could shift toward excess capacity. Specialist computing-rental companies that finance data-center construction at high cost and lack stable, long-term customers may face greater operating pressure.