Mira Murati left OpenAI in September 2024 without saying much about what came next. Eighteen months later, the answer is a multiyear partnership with Nvidia worth potentially tens of billions of dollars and enough computing power to run 750,000 homes.
The two companies announced on Tuesday that Thinking Machines Lab, Murati's AI startup, will deploy at least 1 gigawatt of Nvidia's next-generation Vera Rubin processors, with hardware scheduled to come online early next year. Nvidia has also made what the companies describe as a "significant investment" in Thinking Machines, though neither side disclosed the amount. The Financial Times reported the chip supply portion alone is worth tens of billions of dollars.
For Murati, it is a signal that she is not building another OpenAI wrapper. She is competing at the frontier.
What a gigawatt actually means
Until recently, only the largest AI labs in the world, such as Google, Meta, and Microsoft, operated at that scale. For a company that had roughly 30 employees a year ago and released its first product only last October, reaching that threshold puts Thinking Machines in a category occupied by very few.
The sheer scale of compute also shapes what kind of AI a company can build. Training the most capable models requires vast amounts of processing power run continuously over weeks or months. Without it, companies are limited to refining or adapting models built by others. With a gigawatt behind her, Murati is signaling she wants to train from scratch.
Nvidia's Vera Rubin platform, the successor to its current Blackwell chips, is the company's most advanced hardware and is expected to ship in the second half of this year. The deal makes Thinking Machines one of its earliest major customers.
Who Mira Murati is, and why it matters
Murati joined OpenAI in 2018 as VP of applied AI and research, and rose to chief technology officer before the company's most turbulent moment. In November 2023, OpenAI's board abruptly fired CEO Sam Altman. Murati was appointed interim CEO, a role she held for roughly two days before the board reversed course and reinstated Altman under pressure from staff and investors. The board was subsequently overhauled.
She left the company almost a year later, in September 2024, and founded Thinking Machines in February 2025. Within five months, the startup was valued at $12 billion after raising $2 billion in seed funding from investors including Andreessen Horowitz, Accel, and, notably, Nvidia itself.
Thinking Machines is developing multimodal AI systems focused on collaboration with users rather than full autonomy, a deliberate contrast to the direction most large AI labs have taken. Its first product, an API called Tinker, launched in October 2025 and is designed to make it easier to fine-tune open-source models including Llama, Qwen, and DeepSeek.
The company has not been without turbulence. In January, Murati fired cofounder Barret Zoph, reportedly over a relationship with an employee. Within hours, Zoph and fellow cofounder Luke Metz had announced they were rejoining OpenAI. Despite that, Thinking Machines has grown from about 30 employees a year ago to roughly 120 today, with more hires arriving from top AI labs than leaving for rivals.
The circular money problem
The Nvidia-Thinking Machines deal is significant on its own terms. It is also another data point in a controversy that has been building across the AI industry for more than a year.
The pattern works like this: Nvidia invests in an AI startup. That startup uses the capital, in part, to purchase Nvidia chips. Nvidia books the revenue. Its stock rises. Repeat.
The risk with these arrangements is that they can create skewed incentives and magnify losses if demand for AI fails to match today's lofty expectations. Critics have drawn comparisons to the late 1990s, when telecom companies like Lucent extended vendor financing to struggling customers who then used the money to buy Lucent equipment, a structure that eventually collapsed.
Nvidia addressed the criticism directly, publishing a seven-page document rebutting claims that it invests in customers to inflate revenue. The company argued it does not rely on vendor financing arrangements to grow sales and that its customers pay within 53 days of purchase, unlike classic vendor financing deals where repayment stretches over years.
Prominent short-sellers are not convinced. Jim Chanos, who correctly called the collapse of Enron, has argued that Nvidia is putting money into loss-making companies in order for those companies to order its chips. Bernstein Research analyst Stacy Rasgon wrote that the pattern will clearly fuel circular concerns, even if it falls well short of constituting fraud.
The Thinking Machines deal fits the template precisely. Nvidia has invested in the company and will supply it with hardware. The investment predates Tuesday's announcement, but the new chip commitment deepens the relationship in exactly the way critics have flagged across the industry.
Nvidia's dealmaking spree
The Thinking Machines announcement is one of several large partnerships Nvidia has struck in recent months. In February, the company announced a major multiyear deal with Meta. In early March, it announced agreements with optics companies Coherent and Lumentum. And last year, Nvidia agreed to invest $30 billion in OpenAI as part of that company's $110 billion fundraising round.
Nvidia's financial results give some context for why it continues making these moves. In its third quarter, the company posted earnings per share of $1.30 on revenue of $57 billion, comfortably ahead of analyst expectations. Its data center business generated $51.2 billion in sales against estimates of $49.3 billion. For the fourth quarter, Nvidia guided for $65 billion in revenue, well above Wall Street's $62 billion forecast.
Those numbers reflect genuine demand. The question the circular financing debate raises is not whether the demand is real today, but whether a portion of it is being manufactured by the investing relationships Nvidia has cultivated, and what happens to the growth story if those relationships ever unwind.
For now, Murati is focused on what that compute can build. "This partnership accelerates our capacity to build AI that people can shape and make their own," she said in Tuesday's statement, "as it shapes human potential in turn."
That is a vision worth watching. Whether the financial structure supporting it is as durable as the ambition is a separate, harder question.