Star on trust: how Mira Murati attracts investment in a startup without a product

Mira Murati - former CTO of OpenAI and co-founder of startup Thinking Machines Lab - raised $2 billion from investors in July. As noted by TechCrunch, this is one of the largest seed rounds of funding in Silicon Valley history. Thinking Machines Lab's total valuation has reached $12 billion - the company is less than a year old, it hasn't brought a single product to market yet, and Murati hasn't even told much about what it's going to do yet, other than that it will be a new AI. Still,  Thinking Machines Lab is one of the few startups that investors see as a real threat to leading AI modelers like OpenAI, Anthropic and Google DeepMind. Thinking Machines Lab is a level of credibility that many startups can only dream of. Let's try to understand how Albania-born Mira Murati achieved this and what she is going to shake up the market with.

Ermira "Mira" Muratiwas born in the city of Vlore in southern Albania in late 1988. At the age of 16, she received a scholarship to attend Pearson College UWC, a Canadian private college, and completed the International Baccalaureate program there. She then continued her studies in the United States, completing a dual program: in 2011 she became a Bachelor of Arts at Colby College and in 2012 a Bachelor of Engineering at Dartmouth College. A year later, she was already working at Tesla as a senior product manager, participating in the development of the Model X. Her interest in human-computer interaction led her to Leap Motion in 2016, where she was already working on augmented reality technologies as vice president of product and engineering. In 2018, Murati joined the OpenAI team.

It has to be said, it's an unobvious career choice: with Tesla on her resume, Murati could have applied for positions at other major Valley tech companies, but as she said in an interview with Wired magazine, she's always been interested in AI, believing that general-purpose AI (AGI) "will be the newest and most important technology that humanity will create." At the time, she says, OpenAI was the only organization that was both interested in developing AI capabilities and making sure that development was going in the right direction. At OpenAI, Murati began working on supercomputer strategy and also "led several research teams" as vice president of applied AI and partnerships.

As she told in another interview with Fortune magazine, OpenAI was at the time a little-known nonprofit with just 52 AI enthusiasts. Mira quickly got to know its main co-founders, Sam Altman, Ilya Sutzkever, Greg Brockman, and others. Apparently, she was a good leader, because less than four years later, in May 2022, she  became the CTO of OpenAI and led the teams that created the now famous ChatGPT, as well as the Dall-E image generation AI.

And then we all remember what happened: in late November 2022, ChatGPT was made publicly available, and in just two months it gained over 100 million users, becoming the fastest-growing app in history. The AI revolution had begun.

However, despite Murati's significant contributions, the public face of the company remained, of course, OpenAI CEO Sam Altman. Until an attempted corporate coup in November 2023, when the OpenAI board of directors fired Sam Altman and appointed Murati as interim CEO. The moment of glory didn't last long - Altman returned to his post a few days later. And in September 2024, Murati wrote on social media X that she was leaving "to pursue her own research." Again, an unobvious career move - leaving a company that had grown from a tiny lab to a staff of 2,500 employees and a estimate of $80 billion. For what? Maybe for a better work-life balance to please her parents? When Mira's mom was introduced to ChatGPT, the first thing she asked was: "When is Mira getting married?", reported Business Insider. Maybe for this one, too - according to data from Kosovo's Gazeta Express, she did get married in June 2025, with the ceremony allegedly held in Tuscany in the highest secrecy. The source of information, however, does not look very reliable, the newspaper is obviously yellowish.  For us, it is more interesting to another - in December 2024, Mira Murati wrote on her LinkedIn that she became co-founder and CEO of a startup Thinking Machines Lab, no less secretive than her alleged wedding.

In February of this year, the company officially announced itself on the X network, albeit in the most general of words: "We are the scientists, engineers, and developers behind some of the most popular AI products and libraries, including ChatGPT, Character.ai, PyTorch, and Mistral. Our mission is to make artificial intelligence work for you, creating a future where everyone has access to the knowledge and tools that enable AI to meet their unique needs."

A somewhat more detailed manifesto on the company's own website also sheds little light on its plans. It only shows that the lab will be engaged in machine learning of AI models that will be really useful and convenient in cooperation with humans.

"Guessing what Murati's company will do and how much money it can raise has been a favorite game in Silicon Valley since she left OpenAI in September,"  wrote Bloomberg in February, while also recalling that Thinking Machines Lab is seeking a $9 billion valuation. The publication listed Mira Murati and four other principal co-founders as potential AI billionaires, with each valuing their stake at about $1.4 billion. That's an impressive rise, considering that in 2004 her wealth as CTO of OpenAI was estimated  "just" $5 million. In addition, Bloomberg mentioned in March that Thinking Machines Lab was seeking $1 billion in funding.

In mid-July, as I noted, it became clear that the plan had been oversubscribed - the company raised $2 billion at a total valuation of $12 billion. The round was led by investment fund Andreessen Horowitz and was also joined by Nvidia, Accel, ServiceNow, CISCO, AMD and Jane Street, wrote TechCrunch.

"We're building a multimodal AI that works with the way you naturally interact with the world - through conversation, through vision, through our out-of-the-box interactions. We're excited to be able to unveil our first product in the next couple months," promised Murati, announcing the deal.

Interesting insights about the new startup shared by The Information in June, citing its sources. According to him, investors who have spoken to Murati call Thinking Machines Lab's business model "RL for business." RL in this case stands for reinforcement learning. It's a common machine learning method for developing AI that, using a special algorithm, "rewards" the model with points  for achieving certain goals and "punishes" it by reducing points for "wrong" or undesirable behavior. Thus, in much the same way a human learns, the model explores its environment and looks for ways to solve problems that maximize rewards. Thinking Machines Lab plans to customize models based on certain business metrics (KPIs) tracked by its clients, which are typically related to revenue or profit growth.

"Thinking Machines Lab may be counting on AI customers being willing to pay more for models customized for their industry, such as customer support, investment banking or retail," The Information writes.

Speaking of retail. In late June, AI startup Anthropic presented the results of an interesting experiment: they assigned a Claude-based AI agent to manage a small retail store in their company's office. The agent was named Claudius and was given a pretty wide range of powers: it could monitor inventory, independently determine assortment, set prices, order from suppliers, and communicate with its customers (Anthropic office employees) via a special channel in corporate Slack. Read it, it's a very entertaining story. To make a long story short, Claudius turned out to be a bad businessman: in just one month he, as they call it, haggled and suffered losses. The most dramatic drop in revenue was due to the purchase of a large number of metal cubes (at the request of the company's employees), which were then sold at a price lower than what Claudius had paid because he was too "soft-hearted" and easily gave large discounts when asked. Still, Anthropic is optimistic about the experiment, and believes that AI can be made into a successful middle manager with some adjustments. "In the long term, it may be possible to fine-tune enterprise management models, perhaps through an approach such as reinforcement learning, in which smart business decisions are encouraged and selling heavy metals at a loss is discouraged," Anthropic writes.

Looks like that's exactly what Mira Murati is going to do.

Her startup also plans to develop a product for consumers, possibly a chatbot that could compete with OpenAI's ChatGPT, sources told The Information. 

In order to catch up with the competitors, Murati plans not to create its AI completely from scratch, but to take individual "layers" from already existing open-source models and combine them;

"When you send a message to an artificial intelligence application like ChatGPT, it passes through many layers of virtual neurons in a neural network. Each layer has a different purpose, from identifying certain patterns to deepening the artificial intelligence model's understanding of your text and forming a response. Murati reportedly believes that by combining certain layers, her company can create powerful artificial intelligence models much more efficiently than larger competitors,"  explains Inc.

All in all, competition is a good thing. Let's wish Mira Murati and her team success, especially if she drew (hopefully) the right conclusions from her favorite movie, 2001: A Space Odyssey, in which the ship's AI went crazy and killed the crew.

This article was AI-translated and verified by a human editor