A gigawatt of chips that had not shipped yet
NVIDIA and Thinking Machines Lab announced a multiyear strategic partnership on March 10, 2026. NVIDIA committed to deploy at least one gigawatt of its next generation Vera Rubin systems for Thinking Machines' frontier model training and its customizable AI platforms, with the first rollout targeted for early 2027. NVIDIA also made an investment in Thinking Machines, with the amount not disclosed in either company's announcement. [1][2]
Vera Rubin is NVIDIA's next generation architecture, meaning the hardware behind the commitment had not shipped to any customer when the deal was announced. The partnership also covers joint work designing training and serving systems for NVIDIA's architectures and broadening access to frontier and open models for enterprises, research institutions and the scientific community. [2]
Atlas interpretation: A gigawatt is the unit large labs use to describe compute at the scale of a power grid connection rather than a server order. Committing that scale of unreleased hardware to a company thirteen months old, alongside taking a stake in it, put Thinking Machines in the small group of labs NVIDIA treats as both a customer and an investment, ahead of the Vera Rubin generation actually existing anywhere. [1]
What followed: a shipped model, a smaller round
Four months later, Thinking Machines shipped its first product, the open weights model Inkling, a year after its record seed round with no product at all. [3]
By early September 2026, Yahoo Finance reported Thinking Machines was in talks to raise $5 billion to $6 billion at a minimum $40 billion pre money valuation, with NVIDIA discussing an investment of roughly $2.5 billion of that round. The reported target was down from an earlier $50 billion to $60 billion ambition that the article said collapsed earlier in 2026 after three co founders departed the company. The same report described NVIDIA's March commitment directly: capital and hardware supply, in its words, were no longer separate conversations at NVIDIA. [4]
Atlas interpretation: The March deal is easiest to read next to what came after it. NVIDIA priced its stake and its compute commitment into the same conversation before Thinking Machines had shipped anything, and the September reporting suggests that pattern held: the same investor was back discussing another check months later, after a product existed and after the company's own fundraising ambitions had come down from where they started. Whether NVIDIA's gigawatt commitment survives the lower valuation on the same terms is not addressed in either company's public account of the March deal. [4]
Sources
- Thinking Machines Lab and NVIDIA Announce Long-Term Gigawatt-Scale Strategic Partnership
Thinking Machines Lab · Mar 10, 2026
- NVIDIA and Thinking Machines Lab Announce Long-Term Gigawatt-Scale Strategic Partnership
NVIDIA · Mar 10, 2026
- Inkling: Our Open-Weights Model
Thinking Machines Lab · Jul 15, 2026
- Thinking Machines Lab Seeks $40B Valuation as Nvidia Extends Its Capital Allocator Role Into Model-Space
Yahoo Finance · Sep 4, 2026