What are the AI lab compute commitments?

AI lab compute commitments could result in $614 billion in infrastructure spending through 2030, according to Bernstein analyst Mark Moerdler. This figure is lower than the $750 billion often cited, as construction delays, cancellations, and flexible hardware purchases could delay or reduce payments. Annual infrastructure spending may peak at $184 billion in 2031, with costs rising gradually as data centers become operational.

OpenAI reportedly achieved $50 billion in annualized revenue in September, with growth at 77% year-over-year. This revenue growth could help cover the expanding infrastructure needs, though projections remain uncertain. Oracle's relationship with OpenAI involves about $300 billion in contracted capacity, spread over five to six-year agreements.

OpenAI represents less than half of Oracle's remaining performance obligations, limiting the concentration of risk. Microsoft has a separate $250 billion Azure services commitment from OpenAI, with capacity delivered in six-year stages. Microsoft's diversified cloud operations may provide better cushioning against delays compared to smaller infrastructure specialists.

Amazon has commitments totaling $138 billion across seven- and eight-year terms, including its Trainium processors. CoreWeave's OpenAI agreements amount to up to $22.4 billion, but these are contract ceilings rather than guaranteed revenue. The bigger industry risk may surface when computing shortages ease, exposing speculative construction projects and providers reliant on expensive financing.

Existing contracts with established cloud companies seem more resilient than future demand for specialized computing rentals.