Thinking Machines Lab Lands NVIDIA Investment
▼ Summary
– NVIDIA and Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, have formed a multiyear partnership involving a compute commitment worth tens of billions of dollars.
– The partnership includes NVIDIA supplying at least a gigawatt of its next-generation Vera Rubin systems and making a significant, undisclosed investment in the startup.
– Thinking Machines Lab, which has raised over $2 billion, aims to build highly customizable AI systems, distinguishing itself from competitors that offer more fixed products.
– The deal reflects the intense industry race to secure vast computing power, as securing it early is seen as a key to gaining a durable competitive advantage in AI.
– For NVIDIA, this investment generates revenue and secures a strategic stake in a promising AI lab, while Murati’s refusal of a Meta acquisition shows her intent to build an independent company.
A major strategic alliance has been formed between NVIDIA and the artificial intelligence startup Thinking Machines Lab, founded by former OpenAI executive Mira Murati. This multi-year partnership centers on a compute commitment of staggering scale, with the startup set to deploy at least a gigawatt of NVIDIA’s forthcoming Vera Rubin AI systems to power its model development. The financial terms of the accompanying investment remain confidential, but the chip supply agreement alone is reported to be valued in the tens of billions of dollars, underscoring the immense resources now being directed toward a select group of frontier AI labs.
Murati established Thinking Machines Lab in early 2025, and in just over a year, it has secured more than $2 billion in funding from a notable consortium. Backers include premier venture firms Andreessen Horowitz and Accel, alongside strategic investments from both NVIDIA and, interestingly, the venture arm of its chief rival, AMD. The company has rapidly expanded its team from approximately 30 to about 120 employees, signaling its transition from a nascent idea to a serious contender in the AI landscape.
The startup’s core philosophy distinguishes it from other major AI firms. Its mission is to develop AI systems that are more widely understood, customizable and generally capable. This focus on customizability appears to be a deliberate strategic positioning. While companies like OpenAI and Anthropic primarily offer fixed product offerings, Thinking Machines is building foundational infrastructure designed to be shaped and adapted by other companies and developers to meet their specific needs.
The collaboration with NVIDIA extends far beyond a simple purchase order. It involves deep technical cooperation to optimize Thinking Machines’ software stack for NVIDIA’s hardware architecture. This level of chip-level integration has historically been a powerful accelerant; it was a key component in the rapid advancement of models like GPT. Murati emphasized this point, stating that NVIDIA’s technology is the foundation of the modern AI field and that the partnership will hasten their ability to create malleable, user-directed AI.
This deal is a clear signal of the intensifying compute race within advanced AI research. Thinking Machines is among a handful of labs securing compute resources at the gigawatt scale, often through agreements signed well before the physical hardware is operational. The industry-wide bet is that securing massive computational capacity early will confer a lasting competitive edge in developing the next generation of models. For NVIDIA, such partnerships serve a dual purpose: they guarantee substantial revenue from hardware sales while also providing an equity stake in potential industry leaders, building a portfolio that aligns with the cutting edge of AI development.
Murati’s path to this point involved turning down a notable acquisition offer from Meta’s Mark Zuckerberg last year. The scale of the NVIDIA partnership demonstrates her commitment to maintaining an independent trajectory, backed by the formidable resources required to make that ambition plausible. The central question now is whether a compact, 120-person laboratory can effectively compete with organizations orders of magnitude larger. With this partnership, one significant hurdle is removed: she is no longer short of compute to try.
(Source: The Next Web)
