Thinking Machines Lab has received $2 billion in seed - round financing, and talent has become the most important factor in the AI industry.
The AI startup Thinking Machines Lab, founded by Mira Murati, the former CTO of OpenAI, has received a $2 billion seed round of financing led by a16z, with a valuation reaching $12 billion. Well - known institutions in various fields such as NVIDIA, Accel, ServiceNow, CISCO, AMD, and Jane Street participated in the financing. This is also the largest seed - round financing in the history of technology.
Founded in February 2025, Thinking Machines Lab initially planned to raise $1 billion with a valuation of $9 billion. However, over the past few months, the amount has increased steadily and finally settled at $2 billion.
The only publicly available information about this company currently is its luxurious founding team and a relatively vague technological direction. There is no specific product yet, let alone any revenue.
This financing, along with Safe Superintelligence, founded by Ilya Sutskever, which is already valued at $32 billion (also without products or revenue), and Meta's "crazy" poaching of top AI research talents, represent that the importance of talent in the AI industry has been raised to an unprecedented height.
Valued at $12 billion, no products yet, only an ultra - luxurious founding team
Mira Murati joined OpenAI in 2016 and held senior positions. Later, she served as the company's CTO. Under her leadership, the team developed a series of groundbreaking technologies such as GPT - 3, GPT - 4, DALL - E, and ChatGPT. Before that, she worked as a senior product manager at Tesla, responsible for developing the Model S, Model X, and Tesla's Autopilot advanced driver - assistance system.
Mira Murati
AI experts including John Schulman, Barret Zoph, Bob McGrew, Alec Radford, Alexander Kirillov, Jonathan Lachman, and Lilian Weng are all members of the founding team of Thinking Machines Lab.
John Schulman (Image from his personal official website)
John Schulman is a co - founder of OpenAI. He studied under Pieter Abbeel, a well - known AI expert at the University of California, Berkeley. In the early days, he led the reinforcement learning team at OpenAI, developing the famous PPO (Proximal Policy Optimization) algorithm and the TRPO (Trust Region Policy Optimization) algorithm. He also co - led the Super Alignment team with Ilya Sutskever, responsible for the post - training of ChatGPT. He briefly joined Anthropic in August 2024 and then joined Thinking Machines Lab as the chief scientist.
Barret Zoph (Image from his personal GitHub page)
Barret Zoph is the current co - founder and CTO of Thinking Machines Lab. He joined OpenAI in September 2022 as the vice - president of research and played a key role in the post - training of ChatGPT. Previously, he joined Google Brain in 2016 as a research scientist, focusing on training large sparse language models.
Bob McGrew (Image from a video screenshot of the podcast Startup Archive)
Bob McGrew is the former chief research officer of OpenAI and a key figure in the development of models such as GPT - 3, GPT - 4, and others. Before joining OpenAI in 2016, he held key technical positions at Paypal and Palantir for a long time.
Alec Radford (Image from his social media avatar)
Alec Radford is the first author of the GPT paper, which made OpenAI firmly follow the path of the Scaling Law. He is also a co - author or key contributor to key models such as GPT - 2, CLIP (text - image model), DALL·E, and Whisper. He is one of the technical souls of OpenAI.
Alexander Kirillov led the multimodal research team at OpenAI and was deeply involved in the development of projects such as Advanced Voice Mode and GPT - 4o.
Jonathan Lachman (Image from his social media avatar)
Jonathan Lachman is the former head of special projects at OpenAI. He was previously the chief operating officer of Leap Motion and also served as the head of strategic operations at Blend, driving the company's revenue from $20 million to $250 million. He promoted multiple strategic partnerships at OpenAI.
Lilian Weng (Image from her social media avatar)
Lilian Weng served as the head of the AI security system at OpenAI, committed to ensuring that AI systems are safe, controllable, and in line with human values.
Focusing on AI security, the first product will be released within a few months
In terms of technology and products, the information revealed by Thinking Machines Lab is relatively vague. Mira Murati said on social media that their first product will be released within a few months, which will include important open - source components and bring practical help to researchers and startups developing customized models.
According to a report by foreign media The Information, Thinking Machines Lab will develop AI solutions customized around KPIs for enterprises. This business is called "reinforcement learning for businesses" by investors. In addition to reinforcement learning, according to the information revealed by Murati to early investors, Thinking Machines Lab will use a new technology, which is to selectively "extract" specific layers from open - source models and combine them.
On the official website of Thinking Machines Lab, they also emphasize multimodal capabilities and effective AI security measures, including alignment, red - team exercises, and post - deployment monitoring. This concept coincides with that of Ilya Sutskever's Safe Superintelligence.
The talent war among giants and top AI companies: talent becomes the most important factor in the AI industry
Anthropic, Safe Superintelligence, and Thinking Machines Lab are the three most powerful companies among the OpenAI Mafia (startups split from OpenAI). Their common feature is that they are led by top AI researchers and are full of senior talents.
The huge amount of financing and popularity they have received (especially Safe Superintelligence, as Thinking Machines Lab has not yet launched a product) show a new trend in the AI industry, that is, talent has become the most important factor in the AI industry.
The three important factors in the AI industry are talent, computing power, and data. Among them, computing power and data are directly related to the Scaling Law. The richer the data and the more computing power, the more likely it is to train a powerful model. For example, Grok3 and Grok4 of xAI are still following this approach. However, on the other hand, Ilya Sutskever predicted last year that the Scaling Law would gradually become ineffective, and the benefits brought by the inference calculation of the inference models represented by the o - series models are also gradually weakening.
The emergence of DeepSeek has shown the industry that they have trained the open - source Sota model R1 under the premise of limited computing power. In contrast, Meta has the second - to - none data and computing power in the industry, but after three generations of accumulation, the Llama 4 model has suffered a setback. This makes the industry realize that among the three factors of talent, computing power, and data, talent should be the core.
As a result, we have witnessed the talent war led by Mark Zuckerberg. He not only acquired 49% of the equity of ScaleAI for $14.3 billion and appointed Alexandr Wang as the leader of Meta's new "Super Intelligence" department but also poached former GitHub CEO Nat Friedman and Daniel Gross, the co - founder of Safe Superintelligence, to jointly lead this department.
In addition, he poached Shuchao Bi and Huiwen Chang, the developers of the voice and image modules of GPT - 4o, Trapit Bansal and Jason Wei, the representatives of the "Chain - of - Thought" technology, and important R & D members of the GPT series of models such as Ji Lin, Shengjia Zhao, Hongyu Ren, and Jiahui Yu from OpenAI.
He also poached Jack Rae and Pei Sun from DeepMind and Ruoming Pang, the head of Apple Foundation Models, from Apple.
Earlier, Google's deal with Character.AI was also targeted at talent. After the $2.7 - billion deal, Noam Shazeer (the main author of the Transformer paper) and Daniel De Freitas, the two co - founders of Character.AI, and about 30 core research team members have joined the Google DeepMind team, dedicated to the development of AI projects such as Gemini.
In addition to acquisitions targeting talent in the field of foundation models, the same is true in the application field. Also Google, reached a licensing agreement with AI coding unicorn Windsurf for $2.4 billion, "poaching" Varun Mohan, the CEO of Windsurf, co - founder Douglas Chen, and some core R & D team members. Another AI coding unicorn, Cognition, acquired the remaining assets and team of Windsurf.
Why do both those engaged in foundation models and those engaged in AI applications start to scramble for talent? Ultimately, it lies in the high cost of model training and application. When the computing power resources are determined, the results obtained by different people in training and applying models will be very different. And with each iteration, the computing power cost to be paid is getting higher and higher (while Zuckerberg is recruiting talents, he is also building a super - computing center with a capacity of 5GW). So talent with countless valuable practical experiences and know - how has become a strategic resource.
Fortunately, according to the latest data from the Paulson Institute at the University of Chicago, 47% of the world's top 20% AI researchers are from China. The emergence of a group of world - class language, vision, and robot models and numerous excellent AI applications in China also proves this point.
In the US market, according to a report by Pitchbook, the financing amount of US startups soared by nearly 76% in the first half of 2025, reaching $162.8 billion, of which investments in the artificial intelligence field accounted for about 64.1% of the total transaction volume.
In China's venture capital industry, this wave of artificial intelligence fever is also in progress. We hope that those top talents who are still in universities, research institutes, and large companies, understand technology, can innovate, and are willing to start a business will participate in this wave.
This article is from the WeChat public account "Alpha Commune" (ID: alphastartups). Author: Discovering Extraordinary Entrepreneurs. Republished by 36Kr with authorization.