Waymo's 'Billions of Miles' of Virtual Tests Signal AI's Next Step
Morgan Stanley analysts declared that AI is moving beyond language toward models that can understand and simulate the physical world. A new report from the investment bank positions "World Models" as the next growth engine, addressing the shortcomings of Large Language Models (LLMs). While LLMs excel at tasks like writing and coding, they struggle with problems involving three-dimensional space, time, and physical laws. World Models function as an AI's "imagination engine," creating internal representations of an environment to predict outcomes and consequences.
This shift is already in practice. Waymo has utilized world models based on DeepMind's Genie 3 technology to conduct "billions of miles" of virtual road tests, training its systems on rare and dangerous edge cases. Similarly, Microsoft has demonstrated a fully AI-rendered, playable version of the 1997 game Quake II, where the model predicts each frame based on player input rather than relying on a traditional graphics engine.
AI Unicorns Emerge With Over $2.3B in Seed Capital
Investor capital is rapidly flowing into this new sector, creating instant unicorns led by the field's most prominent figures. Two startups exemplify this trend: World Labs, founded by Fei-Fei Li in 2023, and AMI Labs, co-founded by Yann LeCun, which came out of stealth in March 2026. Together, they have secured more than $2.3 billion in early-stage financing.
World Labs focuses on generating persistent, explorable 3D environments from text or image prompts. Its flagship product, Marble, launched in November 2025, aims to be a creative workstation for developers, allowing them to build and edit virtual worlds for gaming, design, and robotics. According to PitchBook data cited in the report, World Labs has raised approximately $1.29 billion, achieving a post-money valuation of around $5.4 billion after its February 2026 funding round.
AMI Labs is pursuing a different path based on LeCun's Joint-Embedding Predictive Architecture (JEPA). Instead of rendering photorealistic scenes, this approach learns to predict future states in an abstract, latent space, prioritizing efficiency and reasoning for applications in robotics and autonomous systems. In March 2026, AMI Labs announced a seed round of $1.03 billion, securing a valuation of over $4.5 billion, with Wit.ai founder Alex Lebrun serving as CEO.
'Imagination Engine' Limited to Minutes of Stable Interaction
While the potential applications in gaming, content creation, and autonomous systems are significant, Morgan Stanley's report offers a sober assessment of the remaining technical challenges. A primary obstacle is error accumulation over time; even advanced models like Google's Genie 3 can only maintain a stable, coherent simulation for a few minutes before objects and physics begin to drift.
Other major hurdles include a lack of precise control over the generated environments, difficulties in simulating complex social dynamics with multiple agents, and the absence of standardized benchmarks to measure progress. These constraints suggest that World Models will likely diffuse first into industries with high tolerance for error, such as digital content and gaming. Penetration into sectors requiring strict physical accuracy, like robotics and autonomous vehicles, will depend on solving these fundamental engineering problems.
