World Labs vs Ami Labs

Architectural Philosophy: Generation vs. Representation

The deepest difference between World Labs and AMI Labs lies in what their models produce. World Labs builds generative models — systems that output 3D scenes you can see, navigate, and edit. Their Marble model takes a text prompt or image and creates a fully explorable 3D environment with geometry, lighting, and spatial coherence. This is spatial intelligence applied directly: the AI understands depth, occlusion, and physical structure well enough to generate convincing 3D worlds.

AMI Labs takes LeCun's position that generation is fundamentally the wrong objective. JEPA learns to predict in abstract representation space — capturing the meaningful structure of how the world changes without wasting capacity on unpredictable surface details. LeCun has argued for years that autoregressive generation (whether of text tokens or pixels) cannot scale to real-world intelligence. AMI's world models are designed to understand and predict, not to render. This makes them potentially more powerful for planning and decision-making in robotics and autonomous systems, but less immediately useful for creative 3D content generation.

Commercial Maturity: Shipped vs. Stealth

As of March 2026, World Labs has a significant execution lead. Marble reached general availability in November 2025 and has been iterating rapidly since — adding real-time generation via RTFM in October 2025, launching the Chisel 3D editor for coarse layout-based scene creation, and releasing the World API in January 2026 for programmatic access to 3D world generation. The Autodesk partnership, anchored by a $200 million investment, signals serious enterprise traction in design and entertainment workflows.

AMI Labs, by contrast, has no public product. CEO Alexandre LeBrun has candidly acknowledged it could take years for world models based on JEPA to move from theory to broad commercial deployment. The company's first practical deployment will likely come through its strategic partnership with Nabla, the clinical AI company co-founded by LeBrun, focused on FDA-certifiable agentic healthcare systems. For buyers evaluating tools today, World Labs is the only option with a shipping product.

Funding and Investor Signals

Both companies have raised over $1 billion, making this one of the most heavily capitalized races in AI history. World Labs raised $230 million at a $1 billion valuation in 2024, then closed a $1 billion round in February 2026 with backing from Autodesk, NVIDIA, AMD, and Fidelity. AMI Labs closed a $1.03 billion seed in March 2026 — the largest seed round in European history — co-led by Cathay Innovation, Greycroft, and Bezos Expeditions, with strategic backing from NVIDIA, Samsung, and Toyota Ventures.

The investor profiles reveal divergent market theses. World Labs' backers (Autodesk, AMD) are rooted in creative tools and GPU-accelerated rendering — a bet on near-term 3D content creation. AMI's backers (Toyota Ventures, Samsung, Temasek) reflect a bet on longer-horizon physical-world AI for automotive, industrial, and hardware applications. Both have NVIDIA backing, reflecting Jensen Huang's strategy of supporting every credible path to world model dominance.

Target Markets and Use Cases

World Labs is squarely targeting the creative and design economy. The Autodesk partnership positions Marble as a tool for architects, game developers, filmmakers, and virtual production studios. The World API enables developers to build applications on top of generated 3D environments. This is a clear product-market fit: creative professionals already work in 3D and need faster ways to generate environments.

AMI Labs is targeting domains where understanding and prediction matter more than visual generation — robotics, healthcare, industrial automation, and wearable intelligence. The Nabla partnership exemplifies this: healthcare AI needs to reason about consequences and constraints, not generate pretty scenes. AMI's JEPA-based world models could power AI agents that plan actions in the real world, a fundamentally different value proposition than 3D content creation.

Open Science vs. Proprietary Product

AMI Labs has made an explicit commitment to open research — publishing papers and open-sourcing code. This mirrors LeCun's longstanding philosophy and his track record at Meta AI (now FAIR), where he championed open models like LLaMA. LeBrun has stated that openness accelerates progress and builds ecosystem advantages. This approach could attract top research talent and foster a community around JEPA-based architectures.

World Labs takes a more conventional proprietary approach, shipping commercial products and APIs while selectively publishing research. This strategy prioritizes product moats and enterprise partnerships over ecosystem building. For the foundation model landscape, AMI's open approach could prove more influential long-term even if World Labs captures more near-term revenue.

The Leadership Factor

Both companies are led by towering figures in AI. Fei-Fei Li is the architect of ImageNet, the dataset that launched the deep learning revolution, and a leading voice on spatial intelligence and human-centered AI. Yann LeCun is a Turing Award laureate who pioneered convolutional neural networks and has been the most vocal critic of autoregressive language models among top AI researchers. The intellectual credibility of both founders has been instrumental in attracting billion-dollar funding on the strength of vision alone.

AMI Labs has assembled a notably deep executive team, with Laurent Solly (former Meta VP Europe) as COO, Saining Xie as Chief Science Officer, and Pascale Fung as Chief Research and Innovation Officer. World Labs' team is more research-focused around Li's Stanford network. Both teams have the talent to execute — the question is whose architectural bet proves right.