OpenAI has abandoned its original $500bn joint venture model for Stargate, pivoting to fragmented bilateral agreements to secure computing power. The shift began when the company halted data centre projects in the UK and Norway, scaled back in Abilene, Texas, and saw key personnel move to rivals like Meta. Originally tied to Oracle, SoftBank, and MGX, Stargate now revolves around leasing capacity from third parties like Oracle, AMD, and Nvidia. Oracle alone has committed $300bn over five years for 4.5 gigawatts of compute, reflecting a market where OpenAI’s aggressive spending outpaces rivals.

The $500bn target has morphed into a $600bn forecast by 2030, with OpenAI claiming 8GW of capacity secured. This aggressive approach contrasts with competitors like Anthropic, which face tighter power constraints. OpenAI’s strategy hinges on renegotiating deals or abandoning projects deemed unprofitable, raising concerns about reliability. For example, the UK’s Nscale abandoned plans after OpenAI blamed regulations and energy costs—a move criticized by the UK AI minister. Microsoft has stepped in at Abilene and Nscale’s Narvik site, suggesting OpenAI’s partners are increasingly uneasy. The company’s “venture mindset,” led by CEO Sam Altman, prioritizes speed over long-term partnerships, a tactic that has both advantages and risks.

OpenAI’s infrastructure push may give it a structural advantage, but financial sustainability remains unproven. The firm missed internal revenue and user growth targets, triggering stock drops for Oracle and SoftBank. Critics argue its reliance on short-term deals undermines trust, while allies point to its ability to outmaneuver rivals in securing compute. As OpenAI claims to be “ahead of schedule,” the $600bn spending trajectory highlights the existential stakes for AI dominance. Whether this model endures or collapses under its own ambition remains to be seen.