Mira Murati is no longer letting Thinking Machines Lab be defined only by its funding round. Her return to public view puts the company's technical bet under a brighter light.
Mira Murati has spent much of the past year and a half doing something unusual in artificial intelligence: saying very little. That worked while Thinking Machines Lab was hiring, raising money, and building away from the daily noise around OpenAI, Anthropic, Google, and xAI. It works less well once a company has a $12 billion valuation, a major Nvidia compute commitment, and a public claim that AI should feel less like a chatbot and more like a collaborator.
That is why her Bloomberg appearance in San Francisco on June 4 matters. As TechCrunch reported, it was Murati's first major media appearance in roughly 18 months, and she used it carefully. She did not promise a consumer launch or declare victory over incumbents. She described a direction. Thinking Machines is building what it calls interaction models, systems designed to process audio, video, and text continuously rather than waiting for a user to finish a neat prompt.
This is not just a user interface detail. The current AI market is full of systems that can write, summarize, code, and reason, but most still operate like call-and-response machines. You speak. They wait. They answer. Thinking Machines is betting that the next competitive layer is the quality of collaboration itself: whether an AI can follow interruptions, read pauses, handle overlapping speech, and respond while a human is still shaping the thought.
Murati's low profile carries weight because of where she came from. She left OpenAI in September 2024 after helping oversee core model development and product launches including GPT-4. During the five-day leadership crisis in November 2023, when Sam Altman was briefly removed as CEO, she also served as interim CEO. That episode made her more visible, but not in a way any technical leader would choose.
Thinking Machines Lab was announced in February 2025 with a deliberately broad mission: build AI systems that are more understandable, customizable, and useful to people. In July 2025, Reuters reported that the company raised about $2 billion at a $12 billion valuation in a round led by Andreessen Horowitz. That is a remarkable number for any seed-stage company, especially one whose public product story was still forming.
Since then, the company has put more substance behind the valuation. Its first product, Tinker, began as a private beta for fine-tuning open-weight models and later moved to general availability with no waitlist. For researchers and developers, that is a practical wedge. It does not require Thinking Machines to beat OpenAI in a general benchmark on day one. It gives technical users a reason to build with the platform now.
The more ambitious claim came in May, when Thinking Machines previewed its interaction models. Reports on the release described systems built to handle live streams of audio, video, and text, with the company arguing that interactivity should be native to the model rather than bolted on through external turn-taking tools. That distinction will sound technical, because it is. But the business implication is simple: if AI is going to sit inside meetings, design sessions, customer calls, classrooms, labs, and trading desks, it cannot behave like a form field with a voice attached.
Compute Shows The Size Of The Bet
The other clue is infrastructure. In March, Axios reported that Thinking Machines had committed to use at least a gigawatt of Nvidia-powered compute beginning early next year, through a multiyear partnership built around Nvidia's Vera Rubin systems. Nvidia also made what the companies described as a significant investment, though the size was not disclosed.
A gigawatt is not a casual startup purchase. It places Thinking Machines in the category of companies that intend to train and serve frontier-scale systems, not merely wrap someone else's model in a sharper product. That matters for enterprise buyers because durable AI platforms need control over performance, cost, safety, and customization. It matters for investors because compute access has become one of the clearest signals of who can compete at the top end of the market.
There are still real questions. Thinking Machines has seen high-profile departures, including co-founder Andrew Tulloch and former CTO Barret Zopf. Murati downplayed the concern in her latest public comments, arguing that building a frontier lab compresses years of normal volatility into months. That may be true, but talent movement is no small matter in AI. These companies are built on scarce people, scarce chips, and scarce trust.
The more interesting question is whether Murati's deliberate silence was a weakness or a strategy. For a while, it looked like stealth by necessity. Now it looks more like controlled sequencing. First came the team and capital. Then came Tinker, a product for researchers. Then came interaction models, the broader technical thesis. Now comes Murati herself, stepping back into the conversation just enough to give the market a shape to follow.
That shape is different from the race to build the loudest chatbot or the most autonomous agent. Thinking Machines is trying to argue that the next leap in AI will come from systems people can steer naturally and adapt deeply. If that is right, the winners may not be the companies with the biggest model alone, but the ones that make powerful systems easier to work with in real time.
For now, Murati has bought Thinking Machines something valuable: attention without overpromising. The next phase will be harder. The company has to turn careful language, expensive compute, and impressive research demos into products that developers and enterprises choose when the incumbents are already everywhere.
Also read: Monako Glass puts AI coding agents on a developer's face • Apex shows investors are moving deeper into the space supply chain • Raspberry Pi shows how far the AI hardware boom has spread
