The $2B Bet Against LLMs: The Biggest Signal in AI This Year
Two of the smartest minds in AI are betting against LLMs... and they just raised $2B to prove it.
Good morning AI entrepreneurs & enthusiasts,
Two of the most legendary minds in AI history just raised $1B each on the same day. And both are betting the current paradigm isn’t enough.
David Silver, the man behind AlphaGo, is building Ineffable Intelligence in London with a $4B valuation and a thesis that LLMs will never reach superintelligence. Fei-Fei Li’s World Labs just closed $1B to build AI that understands worlds, not just words. When researchers of this caliber walk away from the biggest labs on earth to pursue entirely new directions, pay attention. The frontier is shifting.
(And yes, Google just made AI music creation available to millions of people who’ve never heard of Suno...)
In today’s AI news:
AlphaGo creator David Silver raises $1B for London AI lab
Fei-Fei Li’s World Labs raises $1B for spatial intelligence
Google embeds AI music creation directly into Gemini
Today’s Top Tools + Quick News
🧠 AlphaGo Creator David Silver Raises $1B for London AI Lab in Europe’s Largest Seed Round Ever
News: Former Google DeepMind scientist David Silver is raising $1B for Ineffable Intelligence, a London-based startup led by Sequoia Capital at a roughly $4B pre-money valuation. Nvidia, Google parent Alphabet, and Microsoft are also reportedly in discussions to participate, making this Europe’s largest seed round in history.
Details:
Silver spent over a decade at DeepMind, where he led development of AlphaGo, AlphaZero, and AlphaStar, reinforcement learning systems that surpassed humans by abandoning human knowledge entirely. He also contributed to Google’s Gemini family of LLMs before departing late last year.
Ineffable Intelligence was incorporated in November 2025, with Silver appointed director in January 2026. The company has no product, no revenue, and no public roadmap.
Silver’s core thesis: LLMs alone will never reach superintelligence. In a paper co-authored with Richard Sutton titled “Era of Experience,” he argued that AI systems trained on human data are hitting a wall and that progress through supervised learning alone is slowing. His bet is on reinforcement learning, where AI teaches itself through direct experience.
Sequoia partners Alfred Lin and Sonya Huang reportedly traveled to London to meet Silver immediately after his exit, triggering intense competition among venture firms.
Silver joins a wave of elite researchers leaving Big Tech to pursue superintelligence independently. Mira Murati’s Thinking Machines Lab raised a $2B seed round, while Ilya Sutskever’s Safe Superintelligence has raised $2B at a $32B valuation, all before shipping products.
Why It Matters: This is the clearest signal yet that the world’s most elite AI researchers believe the current LLM paradigm has a ceiling. Silver is making a fundamental bet that experience-based learning will unlock what human-generated training data never can. The fact that Sequoia, Nvidia, Google, and Microsoft are all rushing to write ten-figure checks for a pre-product company in London tells you everything about the talent premium in this market: conviction in the right founder now commands billion-dollar capital on thesis alone. For Europe, this is a landmark moment. The last London AI company to attract this level of investor frenzy was DeepMind itself, which Google acquired for ~$500M in 2014. If Silver is right that reinforcement learning breaks through the scaling wall, the implications ripple across every major AI lab on the planet.
🌐 Fei-Fei Li’s World Labs Raises $1B to Bring AI Into the 3D World
News: Fei-Fei Li’s World Labs has secured $1B in new funding, including a $200M strategic investment from Autodesk, with additional backing from AMD, Nvidia, Andreessen Horowitz, Emerson Collective, Fidelity, and Sea. The round accelerates World Labs’ mission to build “world models” that understand, generate, and reason about immersive 3D environments.
Details:
World Labs emerged from stealth in 2024 with $230M at a $1B valuation. Reports from a month ago suggested the company was targeting a $5B valuation for this latest round.
The company launched its first product, Marble, in November 2025. Marble generates persistent, high-fidelity 3D environments from text, images, video, or 3D layouts, with export options including meshes and video.
The Autodesk partnership will explore how World Labs’ models can work alongside Autodesk’s industry-standard design tools, starting with entertainment use cases. Autodesk will also serve as an adviser and collaborate at the research and model level.
Autodesk is simultaneously developing “neural CAD,” a generative AI model trained on geometric data that can generate working 3D models with an understanding of how designs function in the real world.
World Labs plans to expand Marble into robotics and scientific discovery, moving beyond creative content generation into physical-world AI reasoning.
Why It Matters: Fei-Fei Li captured it perfectly: “If AI is to be truly useful, it must understand worlds, not just words.” That single line explains why $1B just flowed into spatial intelligence. Most AI investment is still chasing better language models. World Labs is betting the next frontier is generating 3D environments governed by real physics, geometry, and dynamics. The Autodesk partnership is the strategic unlock. It gives World Labs a direct channel into the professional workflows of architects, engineers, filmmakers, and game designers who collectively represent trillions in economic output. When you combine AI-native world generation with Autodesk’s install base across every creative and industrial vertical, you get a picture of what “spatial intelligence” looks like at scale.
🎵 Google Embeds AI Music Creation Directly Into Gemini
News: Google just launched Lyria 3 inside Gemini, DeepMind’s newest AI music generation model that lets users transform a text prompt, photo, or video into a fully customized 30-second track complete with auto-generated lyrics and cover art.
Details:
The model handles everything from genre and tempo to vocal style and full lyrics autonomously, with users able to feed text, images, or video as creative input. Lyria 3 produces 48kHz stereo tracks with significantly more musical complexity than previous generations.
DeepMind has been developing Lyria since 2023, but this marks the first time the technology reaches mainstream consumers through the Gemini app.
Every generated track is tagged with Google’s SynthID watermark, and Gemini now lets users upload any audio file to verify whether it was AI-created.
YouTube creators gain global access to Lyria 3 through Dream Track for Shorts, enabling customized audio tracks to pair with video content.
Why It Matters: AI music platforms like Suno and Udio have quietly gotten scary good, producing full tracks that fool most listeners. But they’ve remained locked in a niche. Embedding Lyria directly inside Gemini changes the game: Google is putting AI music creation one prompt away for millions of users who’ve never touched a dedicated music model. For content creators and marketers in your audience, the competitive landscape for AI audio just shifted overnight. The distribution advantage Google holds here is enormous.
🛠️ Today’s Top Tools
🎶 Lyria 3 - Google’s new AI music generation model now live inside Gemini
🗣️ Phoenix-4 - Tavus’ real-time human avatar AI with genuine emotional intelligence and sub-600ms latency
⚡ Grok 4.20 - xAI’s upgraded 4-agent collaboration model with native multi-agent reasoning, 2M token context, and the only AI profitable in live stock trading competitions
📰 Quick News
OpenAI hired Meta’s Charles Porch as VP of global creative partnerships, poaching Instagram’s 15-year celebrity dealmaker to win over a skeptical Hollywood following its $1B Disney deal in December.
Tavus launched Phoenix-4, a real-time avatar rendering model with 10+ emotional states, active listening behavior, and sub-600ms conversational latency, backed by a $40M Series B led by CRV and Sequoia.
Perplexity is abandoning advertising on its AI search platform entirely, with executives telling the Financial Times that sponsored content undermines trust in AI-generated answers. The company reached $200M in annual recurring revenue betting on subscriptions instead.
OpenAI acqui-hired enterprise AI startup Nerve, whose “AI chief of staff” product connected data sources and internal tools to AI agents, with the team joining to scale ChatGPT’s search and agentic capabilities.
The trailer for ‘The AI Doc: Or How I Became an Apocaloptimist’ is going viral ahead of its March 27 theatrical release, featuring sit-downs with Sam Altman, Dario Amodei, and Demis Hassabis from the Academy Award-winning team behind Navalny and producers of Everything Everywhere All at Once.