TLDRs;
- Ex-OpenAI CTO Mira Murati’s startup tackles AI output randomness for enterprise reliability.
- Thinking Machines Lab raises $2 billion, attracting top-tier investors and elite AI talent.
- Research on GPU orchestration aims to make AI responses reproducible and predictable.
- Startup plans open research series, challenging the trend of closed AI development.
Former OpenAI CTO Mira Murati is leading a bold new initiative to address one of the most persistent challenges in artificial intelligence, output consistency.
Her US-based startup, Thinking Machines Lab, is conducting pioneering research aimed at reducing randomness in large language models (LLMs), particularly during inference stages where GPU orchestration often introduces variability.
This focus on determinism is critical for enterprise adoption. While many AI systems generate high-quality responses, their unpredictability has limited practical deployment in professional settings, scientific research, and reinforcement learning applications. By aiming to make AI outputs reproducible, Thinking Machines Lab is positioning itself to meet the real-world needs of businesses that demand reliable, repeatable results.
$2B Funding Fuels Early Ambitions
In a statement released last week, Thinking Machines Lab confirmed it had raised an astonishing US$2 billion in seed funding, valuing the company at roughly $12 billion.
The round was led by Andreessen Horowitz, with participation from Nvidia, Accel, ServiceNow, Cisco, AMD, and Jane Street. Remarkably, the company has yet to release a product or generate revenue, underscoring the market’s confidence in Murati’s leadership and the startup’s elite team, nearly two-thirds of whom are former OpenAI employees.
This level of investment reflects broader trends in AI venture capital. Global AI startup funding surged to $100 billion in 2024, up 80% from the previous year, making AI one of the hottest investment sectors worldwide. In this environment, leadership pedigree and technical expertise often outweigh traditional business metrics like revenue or product traction.
Open Research Series “Connectionism” Launched
Thinking Machines Lab has also announced a commitment to transparency in AI development. The company plans to regularly publish research findings and code in a series called “Connectionism,” offering rare insight into its work.
This approach stands in contrast to the increasingly closed strategies of other AI labs, positioning the startup as a hub for developers and researchers who value open collaboration.
A key focus of the research is understanding how randomness in AI models arises at the GPU kernel level. According to researcher Horace He, better control of these processes could lead to more deterministic outputs, significantly improving reinforcement learning and enterprise customization of AI systems. While it remains unclear if these techniques will be part of the startup’s first commercial product, the research itself is already reshaping conversations around AI reliability.
Ex-OpenAI Talent Shapes New Startup
Murati joins a growing trend of former OpenAI executives launching high-profile AI ventures. Others include Dario Amodei with Anthropic and Ilya Sutskever with Safe Superintelligence. This diaspora of talent is redistributing expertise across the AI ecosystem, creating a more decentralized landscape and fostering competitive innovation. The presence of top-tier researchers, including John Schulman, underscores the startup’s technical credibility and its ability to attract both capital and intellectual resources.
The early momentum of Thinking Machines Lab illustrates how AI investment is increasingly a bet on talent, leadership, and vision rather than immediate commercial outcomes. With its dual focus on determinism and transparency, the company could set new standards for enterprise-ready AI systems.
