Nvidia pauses its $36B GPU financing deal over antitrust fear

Nvidia's own employees are what stopped it. The AI Compute Partnership, launched 1 July 2026 and carrying $36 billion in disclosed commitments by late August, is on pause after staff warned that one clause could pull antitrust scrutiny.

The money was never the sticking point. Nvidia offered cloud providers a revenue floor plus half the upside in exchange for GPU capacity commitments on roughly six-year terms. The catch: providers could only lease those chips to Nvidia-approved customers, and Nvidia signalled it preferred capacity spread across many smaller AI firms rather than parked with one large tenant. That is a dominant supplier deciding who gets compute. The two launch signatories — Sharon AI (up to 40,000 GPUs) and Firmus Technologies, building a 360MW Indonesian campus for up to 170,000 — covered about 210,000 chips between them.

Nvidia won't call it a pause. Its statement: the July model "is still in place and continues to evolve due to high demand." Reporting says otherwise, and the walk-back landed less than two months after launch.

Why it matters: whoever finances the GPUs also gets to pick who computes — and Nvidia just learned that's a legal position, not a commercial one.