OpenAI rewrites its Microsoft deal, David Silver raises $1.1B, and humanoid robots land at Tokyo airport

Two companies. One partnership that was supposed to last until the machines became smarter than us. That clause is now gone - and what replaced it tells you everything about where the AI industry's center of gravity is shifting.

David Silver, who just raised $1.1B at his new lab, was the lead researcher behind one of AI's most famous milestones. Which game did his AI system master at DeepMind?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

For years, buried inside the OpenAI-Microsoft deal was a clause that would automatically reshape the partnership the moment OpenAI achieved AGI. That clause is now gone. The two companies renegotiated their agreement Monday, and Microsoft dropped its legal objections to OpenAI's blockbuster $50B deal with Amazon's AWS.

In exchange, Microsoft will remain OpenAI's primary cloud partner, and OpenAI products will continue to be sold through Microsoft's channels. Microsoft also picks up a revenue-share component in the restructured deal - meaning it gets a financial cut as OpenAI grows. The old deal was increasingly strained as OpenAI expanded beyond Microsoft's infrastructure. This rewrite essentially formalizes a looser, more commercial relationship between two companies that still need each other, but less exclusively than before.

Read the full story →

David Silver - the researcher whose work on AlphaGo and AlphaZero redefined what reinforcement learning could do - has raised $1.1 billion at a $5.1 billion valuation for his new British lab, Ineffable Intelligence. The company is only a few months old. That funding round is a statement in itself.

The core thesis is that current AI is too dependent on human-generated data - and that this is a ceiling, not a feature. Silver's approach leans heavily into self-play and environment-based learning, the same principles that let AlphaGo master Go without ever studying a human game record. If Ineffable Intelligence can scale this to general domains beyond games, it would represent a fundamentally different path to advanced AI than the LLM training paradigm everyone else is following. It's a contrarian bet, and a very well-funded one.

Read the full story →

Japan Airlines is launching a trial of humanoid robots as baggage handlers at Tokyo's Haneda airport - one of the world's busiest. It's a direct response to two converging pressures: a surge in inbound tourism and a deepening labor shortage that Japan's aging workforce simply cannot fill.

The trial is genuinely significant because baggage handling is physically demanding, high-volume, and time-critical - exactly the kind of environment where robotics have historically struggled. The fact that Japan Airlines is willing to run a live airport trial, not just a controlled demo, suggests confidence in the hardware. Japan has been one of the most aggressive adopters of robotics in industrial settings, driven by necessity as much as enthusiasm. If this works at Haneda, expect it to spread fast.

Read the full story →

Beijing has ordered Meta to unwind its $2 billion acquisition of Manus, a Chinese-founded developer of autonomous AI agents. China is now requiring domestic tech companies to seek explicit government approval before accepting US investment - a policy tightening that makes cross-border deals like this one increasingly difficult to close.

Manus had made a splash earlier this year as one of the more capable autonomous agent platforms, and Meta's acquisition was seen as a fast-track move to catch up in the agentic AI space. The block is a setback not just for Meta but for any US company eyeing Chinese AI startups. The geopolitical dimension of AI development is getting harder to separate from the technical one - and this decision is a sharp reminder of that.

Read the full story →

More than 600 Google employees, many from the DeepMind lab, signed an open letter to CEO Sundar Pichai this week demanding that Google block the Pentagon from using its AI models for classified purposes. The signatories reportedly include more than 20 principals and directors - this isn't a junior engineer petition.

The letter surfaces a tension that has been simmering since Google's Project Maven controversy in 2018. Back then, employee pressure led Google to decline a contract renewal. The current push comes as the AI industry moves deeper into military AI territory, with competitors like Microsoft and Palantir actively expanding their defense footprints. Whether Pichai responds publicly or quietly sidesteps the demand will be worth watching closely.

Read the full story →

If you've been watching AI companies spin up products at a pace that feels impossible to match, here's one less excuse: 60sec.site lets you build a fully functional AI-powered website in - you guessed it - about 60 seconds. No code, no designer needed. Worth a look if you're shipping something soon.

The answer is Go! David Silver led the AlphaGo project at DeepMind, which in 2016 became the first AI system to defeat a world champion Go player - a feat many experts had expected would take another decade. His follow-up, AlphaZero, taught itself to master Go, chess, and shogi from scratch using only self-play. That same philosophy of learning without human data is the foundation of his new $1.1B company, Ineffable Intelligence.

The OpenAI-Microsoft AGI clause being dropped is a pretty quiet way to mark a major moment - the two companies have essentially decided that "AGI" is no longer a useful legal trigger. Do you think we're closer to AGI than most people assume, or further away? Hit reply and give me your honest take - I read every single response.

That's it for today. More tomorrow - and as always, you can find our full archive at Daily Inference. See you then.