Thinking Machines Lab Secures Multiyear Nvidia Deal for 1 Gigawatt of Vera Rubin Chips

A gigawatt of compute is not a number that comes up in startup deals very often. But Thinking Machines Lab — the AI company founded by former OpenAI CTO Mira Murati — has locked in exactly that through a new multiyear agreement with Nvidia. The deal includes a significant investment from Nvidia and priority access to Vera Rubin chips, the next generation of Nvidia's AI accelerator hardware. It's the kind of infrastructure commitment that used to be reserved for hyperscalers.

Why This Deal Is Bigger Than It Looks

The AI industry has a compute problem that money alone can't fully solve. Even well-funded startups have struggled to secure enough GPUs to train and run frontier models at scale. Nvidia's chips are in short supply relative to demand, and getting allocation — especially for next-generation hardware — often comes down to relationships, timing, and strategic value. Thinking Machines Lab just solved that problem for the foreseeable future.

One gigawatt of compute capacity is a serious number. For context, large AI training runs for models like GPT-4 consumed somewhere in the range of tens of megawatts. A gigawatt puts Thinking Machines Lab in a tier of infrastructure typically associated with cloud providers and national labs, not a startup that hasn't yet shipped a public product. That's a deliberate choice — it signals that Murati's team is building for enterprise deployment at a scale that requires infrastructure commitments made years in advance.

What Are Vera Rubin Chips

Vera Rubin is Nvidia's next-generation AI accelerator architecture, named after the American astronomer. It follows the Blackwell generation and is designed to push the performance ceiling further for large-scale AI workloads — training massive models, running inference at high throughput, and handling the kind of multi-modal tasks that enterprise AI systems increasingly demand. Getting early, guaranteed access to this hardware gives Thinking Machines Lab a window of competitive advantage before Vera Rubin becomes widely available.

For Nvidia, the deal makes strategic sense too. Backing a startup led by one of the most credible figures in AI — Murati spent years at OpenAI overseeing some of the most significant model launches in the industry's history — is a smart bet. Nvidia has increasingly positioned itself not just as a chip supplier but as an active player in the AI ecosystem, and this partnership fits that pattern.

Compute Has Become the New Talent War

A few years ago, the defining competition in AI was over researchers and engineers. Companies paid extraordinary salaries to attract the people who could build and train frontier models. That race hasn't stopped, but a second front has opened up — the fight for compute. Access to enough high-quality hardware, at the right time, has become just as determinative as having the right team. You can hire brilliant researchers and still fall behind if you can't run the experiments fast enough.

Thinking Machines Lab appears to have internalized that lesson from the start. Securing a multiyear Nvidia deal this early — before the company has a widely deployed product — suggests Murati and her team are thinking about infrastructure constraints the way a general thinks about supply lines. You solve logistics before the battle, not during it.

What Comes Next

Thinking Machines Lab has been relatively quiet about its product roadmap, but the focus appears to be on enterprise AI systems — the kind of deployable, customizable intelligence that companies can actually integrate into their operations rather than general-purpose chatbots. That's a crowded space, but compute infrastructure at this scale gives the company real room to differentiate on performance and reliability.

The deal also raises the stakes for the company's eventual product launches. With this level of Nvidia backing and infrastructure commitment, the expectation from investors and the broader industry will be high. Murati built her reputation on shipping things that worked. The resources are now in place. The question is what Thinking Machines Lab actually builds with them.