OpenAI and Anthropic Projected to Account for 40–50% of Global AI Compute Growth Next Year

According to Dylan Patel of Semi Analysis, a significant portion of U.S. GDP growth is driven by AI infrastructure, with approximately one-third of newly operational compute this year being supplied to AI labs such as OpenAI.

Economic Structural Shifts

Annual capital expenditures (CapEx) by AI labs are expected to surpass $1 trillion this year and exceed $2 trillion by 2028. Moving away from a structure where losses were covered by past venture investments, Anthropic is expected to turn profitable in Q2, and OpenAI in Q3. This signifies a transition to a stage where businesses are sustained by their own revenue.

Profitability and Margin Improvements

Compute costs are approximately $100,000–$150,000 per megawatt, but service revenue from the latest models (such as GPT-5.6 and Opus 5) significantly exceeds this. In particular, Anthropic has recorded $500,000 in revenue per megawatt, establishing a virtuous cycle where profits generated from inference are reinvested into training.

Accelerating Market Concentration

30% of newly added compute this year has been allocated to AI labs, and next year, OpenAI and Anthropic are expected to account for 40–50% of incremental compute based on already signed contracts. This trend of compute concentration is accelerating, shifting the center of gravity of global AI infrastructure toward specific large labs.