Runway started by helping filmmakers — now it wants to beat Google at AI

2026-05-15 · Source: TechCrunch · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Runway, an AI video-generation startup founded in 2018 by NYU Tisch School of the Arts alumni, is shifting its focus from text-to-video models to developing "world models" that learn from observational video data. Valued at $5.3 billion, Runway has built a reputation with models like Gen-4.5, powering workflows for major media players like Lionsgate and AMC Networks, and generating $40 million in ARR in Q2 2026. The company launched its first world model in December and plans another this year, aiming to create AI systems that simulate environments to predict behavior. This strategic pivot positions Runway against well-funded competitors like Google, Luma, and World Labs, all pursuing AI that can solve complex problems in fields ranging from robotics to drug discovery. Runway has raised $860 million, including a $315 million round in February, and emphasizes its unique, non-Silicon Valley culture as a competitive advantage.

Key takeaway

For research scientists and entrepreneurs evaluating next-generation AI development, Runway's pivot to world models trained on observational video data suggests a significant shift from language-centric AI. You should investigate the potential of physics-aware video models for applications in robotics, drug discovery, and climate modeling, recognizing that this approach aims to accelerate scientific progress by compressing experimental cycles. Be aware that securing dedicated, large-scale compute resources will be critical for success in this highly competitive domain.

Key insights

Training AI on observational video data and world models, not just language, is the next frontier for advanced intelligence.

Principles

• AI intelligence can emerge from understanding world mechanics, not just human descriptions.
• Less biased data, like direct observational data, can lead to AI beyond existing human knowledge.

Method

Develop AI world models by training on diverse sensory data (text, video, voice) to create digital twins of environments, enabling faster scientific experimentation and problem-solving.

In practice

• Explore video-to-world model conversion for interactive entertainment and robotics.
• Consider multi-modal training for AI systems to achieve compounding effects.

Topics

• AI Video Generation
• World Models
• Runway AI
• AI Competition
• Scientific Infrastructure

Best for: Research Scientist, Entrepreneur, AI Scientist, Director of AI/ML, Investor

Related on AIssential

• Runway Raises $315M for AI World Models — World models are critical for advanced AI video generation, enabling complex environmental understanding and simulation.
• Can you *really* train AI to "get" videos just by showing it a million of them? — Current video models excel at generation but lack systematic evaluation for genuine spatiotemporal reasoning.
• The Sequence Knowledge #812: The Sora Moment: When Video Models Became Physics Engines — Video generation models like Sora are evolving into data-driven physics engines, simulating real-world dynamics.
• Is AI video just a prequel? Runway’s CEO thinks world models are next — AI-generated video is evolving into a versatile creative tool with broader applications beyond traditional content creation.
• Exclusive: Runway launches $10M fund, Builders program to support early-stage AI startups — Runway is expanding beyond AI video generation to cultivate an ecosystem for its "video intelligence" and "general world models."
• AI Is Crossing the Frontier of Human Knowledge | Kevin Weil — AI models are now solving problems beyond human knowledge, poised to accelerate scientific discovery and reshape industries.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by TechCrunch.