The archive · AI & Models · Technical decision · 2023–2026
Decart's efficiency-first bet on real-time world models draws $300M at a $4B valuation
Founded in 2023, Decart bets real-time video and world models need a hardware-deep optimization stack; a $300M round and 100K developers followed
Decart
What the business is
Decart builds real-time generative video and world models — Lucy, Oasis, and Oasis 3 — plus DOS, an optimization stack that runs them efficiently on Nvidia, Amazon, and Google hardware.
How it started
Decart was founded in late 2023 in Israel by Dr. Dean Leitersdorf and Moshe Shalev, both veterans of the Israeli military's Unit 8200. Its thesis was that the next AI platform would be real-time and interactive — generative video and simulated worlds — and that this only becomes a product when running it stops costing a fortune.
What happened
Decart raised $100 million at a $3.1 billion valuation in August 2025, then closed a $300 million round in May 2026 at roughly $4 billion, led by Radical Ventures with Nvidia, eBay Ventures, Adobe Ventures, and Toyota Ventures joining existing backers Sequoia Capital, Zeev Ventures, and Benchmark. Amazon became a strategic customer for enterprise AI applications in media, commerce, advertising, and physical AI. The company says it already generates revenue from contracts with large cloud providers, AI labs, and hyperscale companies.
How it ended up
On June 10, 2026, Decart launched Oasis 3, a photorealistic driving world model available via API at $0.02 per second, targeting autonomous-vehicle companies that need rare scenarios simulated at scale. TechCrunch's hands-on test found striking realism and unlimited generation but also fading scene coherence and imperfect physics — limits the company says it is working on. Decart is still live and expanding into robotics and physical AI.
Background
Decart was founded in late 2023 in Israel by Dean Leitersdorf and Moshe Shalev, veterans of Unit 8200, on a contrarian bet: the next big AI platform would be real-time generative video and world models, and the winner would be the company that makes running them cheap enough to build on, not the one with the biggest model.
That efficiency thesis is embodied in DOS, Decart's optimization stack, which the company says makes its models more than an order of magnitude cheaper to run than competitors' by optimizing all the way down to Nvidia, Amazon, and Google hardware. The approach helped it close a $100M round at $3.1B in August 2025 and a $300M round at nearly $4B in May 2026, with Nvidia as investor and Amazon as strategic customer.
By June 2026 Decart had more than 100,000 developers building on its real-time video model Lucy and shipped Oasis 3, a photorealistic driving world model sold through an API at $0.02 per second. TechCrunch's testing praised the realism but documented the field's open problems — scene coherence decays over time and physics is imperfect — making Decart a leading but not yet proven bet on real-time AI as a developer platform.
What has to be true
- Decart bet that efficiency, not raw scale, decides who wins real-time AI, and proved the point by staying dramatically cheaper than rivals
- API access from day one treated world models like the early LLM API moment, aiming to seed a developer ecosystem rather than sell demos
- Strategic investors who are also potential customers — Nvidia, Toyota, Adobe, eBay, Amazon — signal the bet is shared by the companies that would consume the technology
- Releasing Oasis 3 with honest, documented limitations kept the bet testable: if coherence and physics improve, the platform claim becomes credible
What can be applied
When the bottleneck is cost, efficiency can be the moat: Decart drew strategic investors and 100K developers by making real-time AI an order of magnitude cheaper, not by claiming a better model.
Aftermath
As of June 10, 2026, Decart is live and scaling: it has raised more than $450 million in total, employs roughly a hundred people, and generates revenue from contracts with large cloud providers, AI laboratories, and hyperscale companies. Its three product lines — DOS, Lucy, and Oasis — are aimed at e-commerce, livestreaming, autonomous vehicles, and robotics, with Oasis 3 newly available to developers via API. The company's next milestones are improving long-horizon coherence and physics in its world models and growing the developer community it believes will emerge around them.
Sources
- Decart's new world model can simulate hours of photorealistic driving — with some caveats
- Nvidia backs Israeli AI unicorn Decart in $300 million funding round at $4 billion valuation
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card