AI researcher and co-founder Andrew Tulloch has officially left Thinking Machines Lab (TML) to rejoin Meta, marking one of 2025’s biggest talent moves in artificial intelligence.
Tulloch, best known for his work in machine learning, pretraining, and reasoning systems, is returning to the company where he spent over a decade building scalable AI infrastructure.
Why AI Talent Wars Are Heating Up
The global race for AI supremacy isn’t just about models, it’s about people.
From Silicon Valley to Singapore, top researchers are being poached with billion-dollar offers and stock-heavy incentives.
For startups like Thinking Machines Lab (TML), co-founded in 2025 by former OpenAI CTO Mira Murati and Tulloch himself, retaining talent has become a serious challenge.
When Tulloch’s move to Meta was confirmed on October 11, 2025, insiders described it as both “expected and inevitable.”
Meta’s aggressive hiring strategy underscores how AI expertise has become the new oil, scarce, strategic, and fiercely competed for.
Who Is Andrew Tulloch
Think of Tulloch as a bridge between mathematical theory and large-scale engineering.
An Australian computer scientist and Cambridge-trained mathematician, he has spent more than a decade designing the backbone systems that power modern AI.
Before co-founding TML, Tulloch worked at:
- Meta’s FAIR Lab (2012–2023) – Building scalable AI training infrastructure.
- OpenAI (2023–2025) – Contributing to GPT-4, especially its reasoning and pretraining systems.
His deep focus on “large-scale problems in machine intelligence” turned him into one of the few engineers fluent in both math and model architecture, a rare combination that major labs value immensely.
How Thinking Machines Lab (TML) Changed the Game
Tulloch co-founded Thinking Machines Lab (TML) in February 2025 with Mira Murati and other ex-OpenAI veterans.
The lab focused on bridging research and production through open-model tools, essentially making AI fine-tuning accessible to smaller teams and enterprises.
Their biggest innovation:
Tinker, a Python API for distributed GPU fine-tuning using LoRA adapters.
This tool allows researchers to:
- Train open models efficiently without needing massive GPU clusters.
- Scale training from laptops to supercomputers seamlessly.
- Focus on results, not resource management.
TML raised $2 billion in early funding, proving how much the market believed in its “research-to-production” vision.
But Tulloch’s exit marks a shift, both for the company and for the broader AI ecosystem.
The Meta Comeback: $1.5 Billion Offer and Strategic Reversal
In August 2025, reports surfaced that Meta CEO Mark Zuckerberg tried to acquire TML entirely, a deal that never materialized.
When that failed, Meta shifted strategy and made a personal offer to Tulloch, reportedly worth up to $1.5 billion over six years, including performance-based stock grants.
At the time, Tulloch declined. But two months later, he changed course, accepting a new Meta role that insiders believe will shape Meta’s next-generation reasoning and pretraining systems.
Meta has since hired over 50 top AI researchers, signaling its determination to compete with OpenAI, Anthropic, and Google DeepMind in the superintelligence race.
Key Highlights of Tulloch’s Career
- 2012–2023: Meta (FAIR) – Built scalable AI infrastructure.
- 2023–2025: OpenAI – Contributed to GPT-4 reasoning and pretraining.
- 2025: Co-founded TML – Developed Tinker API for open-model fine-tuning.
- Oct 2025: Returned to Meta – Role undisclosed, focus likely on large-model optimization.
Why Tulloch’s Move Matters to the AI Industry
For everyday observers, this isn’t just a career switch, it’s a signal.
AI’s future is being shaped by a few dozen researchers whose work defines what systems like GPT, Llama, or Claude can do.
Here’s what Tulloch’s move means for you and the industry:
- For Meta: A major leap toward more advanced reasoning and open-model ecosystems.
- For Startups: A reminder that retaining AI talent is now as important as raising capital.
- For Developers: Access to Meta’s upcoming tools could mean faster, smarter model fine-tuning capabilities.
- For AI Policy Watchers: More consolidation of top minds within tech giants, raising ethical and competition concerns.
The Future: AI’s Brain Drain or Brain Merge?
Tulloch’s journey, from Meta to OpenAI to TML and back to Meta, mirrors a larger trend in AI:
A revolving door of elite researchers moving between startups and giants, accelerating innovation but also centralizing power.
In his new role, Tulloch is expected to apply his mathematical optimization and reasoning insights to Meta’s upcoming AI systems, possibly shaping the foundation of the next generation of open or hybrid models.
One thing is clear:
The AI revolution isn’t slowing down; it’s just getting smarter about who leads it.
