Forget chatbots. The next trillion-dollar AI opportunity may not live in text at all—it lives in the physical world. As Large Language Models (LLMs) approach diminishing returns, a seismic capital rotation is underway toward AI systems that can see, navigate, and manipulate three-dimensional reality. The "Godfathers of AI" are leading the charge, and investors are following with checkbooks open.
What Happened: The World Model Moment
In a defining week for late 2025, Meta's Chief AI Scientist Yann LeCun—one of the founding fathers of deep learning—confirmed the launch of a new startup focused on "world models," reportedly seeking a valuation exceeding $5 billion. The venture aims to build AI that understands the physical world, not just text—a fundamental departure from the LLM paradigm that has dominated the past three years.
LeCun's move follows a clear pattern. Fei-Fei Li, Stanford professor and the "Godmother of AI," emerged from stealth with World Labs in September 2024, raising $230 million at a valuation exceeding $1 billion. Remarkably, Li bootstrapped the startup to unicorn status in just four months, with backing from Andreessen Horowitz, NEA, and even Geoffrey Hinton—another deep learning pioneer.
Meanwhile, Physical Intelligence closed a $600 million Series B at a $5.6 billion valuation in November 2025, led by Alphabet's CapitalG. The Jeff Bezos-backed company has now raised over $1 billion to develop Vision-Language-Action foundation models for robots—AI that processes camera feeds and natural language to output real-time control commands.
And the momentum keeps building. Skild AI, founded by former Meta AI researchers, reached a $4.5 billion valuation after its Series B, with SoftBank and NVIDIA now in talks to invest over $1 billion at a $14 billion valuation—nearly tripling its value in seven months.
Why It Matters: Beyond Text, Into Reality
The thesis driving this capital rotation is straightforward: LLMs are fundamentally limited. They predict the next word in a sequence but have no understanding of physics, causality, or the three-dimensional world humans inhabit. World models aim to fill this gap by teaching AI to perceive, reason about, and interact with physical space.
The market opportunity is massive. The spatial computing market is projected to reach $441 billion by 2030, growing at a 22% CAGR according to NextMSC. Other estimates from Research and Markets peg the market at $421.2 billion by 2030, up from $130.3 billion in 2024—a 21.6% compound annual growth rate.
For founders and investors, the implications are significant:
- Robotics becomes software: Physical Intelligence's $300/month per robot subscription model suggests a future where robot intelligence is delivered as SaaS, not embedded hardware.
- Platform plays emerge: Skild AI's "Skild Brain" operates across 100,000+ robot body types, positioning it as a potential "Android for robots."
- New application categories: World Labs targets artists, game studios, urban planners, and manufacturers—industries that need to simulate and interact with 3D environments.
The Competitive Landscape
The race for world models is attracting a unique coalition of talent and capital:
| Company | Latest Valuation | Key Backers | Focus |
|---|---|---|---|
| LeCun's Startup | $5B+ (target) | TBD | Physical world understanding |
| Physical Intelligence | $5.6B | CapitalG, Jeff Bezos, Thrive | Robot foundation models |
| Skild AI | $4.5B → $14B (talks) | SoftBank, NVIDIA, Sequoia | General-purpose robot brain |
| World Labs | $1B+ | a16z, NEA, Geoffrey Hinton | Large World Models, 3D spatial AI |
What unites these companies is a shared belief that the next AI paradigm requires embodied intelligence—systems that learn from physical interaction, not just internet text. Physical Intelligence's π0 model, for instance, already supports seven robot types across 50+ tasks, with success rates exceeding 90% on complex operations.
What's Next: The Embodied AI Era
For founders in the ecosystem, several trends deserve attention:
- Hardware-agnostic wins: The leading players are building models that work across robot embodiments. Startups locked to specific hardware may find themselves outflanked.
- Simulation becomes critical: Training world models requires massive simulated environments. Expect increased investment in synthetic data generation and physics engines.
- Enterprise adoption accelerates: Manufacturing, logistics, and healthcare are early targets. Companies that can demonstrate reliable, deployable systems will capture enterprise budgets.
- Talent wars intensify: The founding teams at these companies read like an AI hall of fame. Recruiting from this talent pool will become increasingly competitive.
The transition from LLMs to world models won't happen overnight. But the capital flows are unmistakable: over $2 billion has moved into this space in the past 12 months alone, with valuations climbing into the tens of billions. For investors and founders alike, the message is clear—the post-LLM era is arriving faster than expected.