AI video startup Runway raised a $315 million Series E at a $5.3 billion valuation in February 2026 to redirect capital toward pretraining next-generation world models, according to TechCrunch. The funding, led by General Atlantic with participation from Nvidia, Fidelity, and others, signals a broader industry pivot-Yann LeCun left Meta in late 2025 to found AMI Labs, which raised $1.03 billion in March 2026, while Fei-Fei Li's World Labs and Google DeepMind have released competing products. For robotics and simulation teams, the shift means workloads increasingly demand new data pipelines, physics-based evaluation metrics, and long-horizon safety testing that standard LLM benchmarks do not cover.
The move transforms what practitioners must test for, not just what they build.
Runway's raise and the founding of AMI Labs
Runway's $315 million round included Nvidia, Fidelity Management & Research, Adobe Ventures, Mirae Asset, Felicis, Premji Invest, and AMD Ventures, TechCrunch reported. The company said it will use the funds to pretrain its next generation of world models, following a December release that shipped specialized variants for environment simulation, robotics, and digital avatars. Yann LeCun announced his departure from Meta in November 2025 and, in March 2026, launched AMI Labs with a $1.03 billion seed round to build JEPA-based world models for robotics and autonomous machines. Additional startups by ex-Nvidia and ex-DeepMind researchers are reportedly raising nine-figure rounds, though the claim comes from a single Threads excerpt and is worth confirming as more detail becomes public.
Competitors and backers align
Runway's pivot places it in direct competition with Fei-Fei Li's World Labs, which has raised about $1.2 billion and shipped the Marble 3D-world-generation product, and with Google DeepMind's Genie research. Nvidia is developing its own Cosmos world foundation models while also investing in rival labs-a signal that compute providers are hedging across the field rather than betting on one winner.
New testing demands for autonomous systems
World models are trained to simulate physical dynamics, spatial relationships, and cause-and-effect, unlike LLMs that act as next-token predictors. Evaluation moves from perplexity and human-preference scoring toward metrics for physical realism, collision fidelity, and long-horizon stability. Teams working on robotics or autonomous systems should expect to rely on simulated or instrumented real-world datasets, domain-randomized training scenarios, and closed-loop rollout testing-not text-scaling alone. This requires simulation engineering and physics-aware safety regimes that differ materially from LLM-centric deployment.
Why this matters for Science & Research
For research teams, integrating world models is not a model-class swap; it demands dedicated budgeting for new data pipelines, simulation tools, and safety evaluation frameworks beyond standard NLP benchmarks. As more labs release open-sourced datasets and evaluation suites, science and research practitioners should track which tools become standardized and whether funding consolidates around a few platforms. Teams that build simulation engineering capabilities and physics-based validation now will be better positioned to adopt these models as they mature.
Your membership also unlocks:
