Altman's Compute Thesis on Invest Like the Best: A Gigawatt of New Capacity Every Week, and Inference Revenue — Not Model Margin — Funds the Frontier

BigGo Finance·medium signalXBlueskyLinkedInCopy link

In the same July 28 episode, Altman laid out the economics behind the deceleration talk: OpenAI's plan to add a gigawatt of new capacity per week, the custom Broadcom-built Jalapeño inference chip he calls a potential 'huge competitive advantage,' and the early decision to buy compute 'at a scale nobody thought was rational.' His stated thesis is that compute fleet scale is the most durable moat, intelligence itself becomes a commodity, and massive inference volume — not high margins on model access — pays for frontier training. He also offered the line that traveled: 'We are about to create a genie that can grant any wish.'