The Brief

Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has struck a multiyear partnership with Nvidia that includes a significant equity investment and access to at least one gigawatt of next-generation Vera Rubin computing systems. Deployment is targeted for early 2027, giving the thirteen-month-old company compute capacity on par with the largest frontier-model laboratories.

The Report

Nvidia and Thinking Machines Lab announced a long-term strategic partnership on Monday that will see Murati’s startup deploy at least one gigawatt of Nvidia’s forthcoming Vera Rubin platform for training frontier AI models. The deal includes a direct equity investment from Nvidia, though neither company disclosed the size of the stake.

The Vera Rubin systems — comprising Rubin GPUs built on 336 billion transistors and Vera CPUs with 88 Arm-based cores — represent Nvidia’s next-generation architecture. Deployment is scheduled for early 2027. At Jensen Huang’s own estimate of roughly $50 billion to build a gigawatt of AI infrastructure, the hardware commitment alone places Thinking Machines among the most aggressively resourced AI ventures in the world. Approximately two-thirds of that cost is GPUs.

The partnership extends beyond raw compute. The two companies will collaborate on designing training and serving systems optimised for Nvidia architectures, with a stated aim of broadening access to frontier and open-source models for enterprises, research institutions, and the scientific community. Thinking Machines’ existing product, Tinker — a cloud service for fine-tuning open-source large language models using LoRA — supports more than twelve models, including Meta’s Llama series.

“AI is the most powerful knowledge discovery instrument in human history,” Huang said in a statement. “Thinking Machines has brought together a world-class team to advance the frontier of AI.” Murati described Nvidia’s technology as “the foundation on which the entire field is built,” adding that the partnership “accelerates our capacity to build AI that people can shape and make their own.”

Thinking Machines was founded in February 2025 by Murati alongside several former OpenAI researchers, including reinforcement learning pioneer John Schulman and Barret Zoph. The company secured a $2 billion seed round in July at a $12 billion post-money valuation — one of the largest early-stage raises in technology history — with participation from Nvidia, AMD, ServiceNow, and others. The company has since grown from approximately 30 to 120 employees. In January, Murati announced the departure of co-founder and CTO Zoph, naming Soumith Chintala as his replacement. Reports in late 2025 indicated the company was seeking additional funding at a valuation approaching $50 billion, though no such round has been confirmed.

The Nvidia investment arrives as compute access has become the binding constraint on frontier AI development, with the largest laboratories securing multibillion-dollar infrastructure commitments to maintain their positions.

The Angle

The numbers here tell a structural story that the partnership language obscures. A year ago, Murati left the organisation that defined the frontier of AI research. Thirteen months later, her startup has secured enough compute to match it. The speed of that rearmament says less about Murati — whose capability was never in question — than about what a gigawatt of Nvidia silicon has become in the current landscape: not a resource but a credential. The chip deal is the moat now. Everything else — the team, the research agenda, the product — follows from whether you can get the hardware.

What is worth noticing is the direction of Nvidia’s strategy. Huang is not picking winners. He is making them. A significant equity stake plus privileged access to next-generation silicon is not a commercial transaction — it is a kingmaking mechanism. Nvidia supplies the substrate on which every frontier lab depends, and it is now using that position to select which new entrants get to compete. The company that controls the picks and shovels has started buying the mines.

Thinking Machines’ emphasis on open models and customisable AI is the other detail that bears weight. The largest labs have consolidated around proprietary systems accessible only through API. Murati appears to be building in the opposite direction — toward models that enterprises and researchers can shape directly. Whether that is conviction or positioning is unknowable from the outside. What is observable is that the compute to make it real just arrived, and it arrived with Nvidia’s name on both the hardware and the cap table.

The company’s turbulence — Zoph’s departure, the reported staff discussions about returning to OpenAI — is real and would normally signal instability. In this case it may signal something closer to the opposite: that the venture has passed the point where early co-founder dynamics determine the outcome. At a gigawatt of committed compute, Thinking Machines is no longer a team. It is an infrastructure position.

The competitive map of frontier AI just added a new entrant with resources that, twelve months ago, only three or four organisations on Earth could claim. The gap between founding a company and fielding a credible challenge to the incumbents has compressed from a decade to a year — provided you can get the chips.

The race to build the systems that will define this century now has another entrant resourced to win it, and the timeline between ambition and capability has never been shorter.