Israeli AI startup Decart raises $300M backed by NVIDIA, Amazon and Radical Ventures, targeting real-time inference costs in the fast-growing world model market.
Decart has closed a $300 million funding round at a $4 billion valuation, making it one of Israel's most heavily backed artificial intelligence startups. Three years old and already past $450 million in total funding, the Tel Aviv company builds infrastructure for running AI at real-time speed.
Radical Ventures led the deal. New investors include NVIDIA, Adobe Ventures, Toyota Ventures, eBay Ventures, Atreides Management and Valor Equity Partners, with Sequoia Capital, Benchmark and Zeev Ventures returning from prior rounds. Individual backers span former Disney chief executive Michael Eisner, Andrej Karpathy, members of Nintendo's founding Yamauchi family and gaming investor Moritz Baier-Lentz, according to Israel National News.
Founded in 2023 by Dr. Dean Leitersdorf and Moshe Shalev, both veterans of Unit 8200, Israel's military intelligence unit, Decart develops what it calls "world models": systems capable of generating and responding to video, simulations and virtual environments in real time. Target markets include media production, advertising, robotics and industrial simulation.
What Decart actually builds
At the center of Decart's platform is DOS 2.0, an optimization layer that reduces the cost and latency of AI inference across NVIDIA GPUs, Google TPUs and Amazon Trainium chips. The company claims it can dramatically lower inference costs without sacrificing throughput, though no independent benchmarks have been published to support those claims.
Alongside the funding, Decart announced strategic partnerships with both Amazon and NVIDIA. Amazon will also function as a direct customer, giving AWS cloud users access to Decart's technology for media, commerce, advertising and robotics workloads. The scope matters: AWS serves hundreds of thousands of enterprise customers, and a native integration would give Decart distribution that most infrastructure startups spend years chasing. NVIDIA's decision to invest, given its direct visibility into GPU utilization across its partner network, is the clearest third-party validation Decart has offered so far.
Repricing from $3.1 billion to $4 billion on $300 million in fresh capital is a modest step. Decart raised $100 million at the earlier figure just last year. The narrow valuation increase relative to new capital raised signals that investors are pricing in meaningful execution risk despite the caliber of the cap table.
World models hit the mainstream
The fundraise lands in a loud week for the world model category. At Google I/O on Tuesday, Google unveiled Gemini Omni, a multimodal model that DeepMind CEO Demis Hassabis described as a meaningful step toward artificial general intelligence, according to TechCrunch. Google's Gemini 3.5 Flash is already the default powering Search's AI Mode and claims to run four times faster than comparable frontier models in output tokens per second, per Mashable.
Decart is not competing with Gemini on model capability. Its position is that whatever model wins the race, running it at real-time scale stays a structural engineering problem, and DOS 2.0 sits at that bottleneck. As Wired reported this week, proactive AI agents are replacing passive assistants faster than analysts projected, a transition that makes low-latency inference infrastructure more commercially valuable than it was 12 months ago.
That investor list reads as a forward order book. Toyota suggests automotive simulation, Adobe and eBay point toward creative and commerce workloads, and Karpathy's participation brings domain credibility from his years building real-time perception systems at Tesla. These are not passive checks.
Replication is the structural risk this round does not resolve. Amazon, NVIDIA and Google each run internal inference optimization teams, and a strategic partnership in 2026 does not prevent any of them from absorbing comparable capability natively by 2028. Whether Decart can embed itself deeply enough in customer workflows before that happens is the question its founders will spend the next two years racing to answer.
---
FAQ
What does Decart AI do?
Decart builds real-time artificial intelligence infrastructure, including DOS 2.0, an optimization platform that lowers cost and latency for AI workloads running on NVIDIA GPUs, Google TPUs and Amazon Trainium chips. The company also develops world models capable of generating video, simulations and virtual environments in real time.
Who invested in Decart's $300 million round?
Radical Ventures led the round, with NVIDIA, Adobe Ventures, Toyota Ventures, eBay Ventures, Atreides Management and Valor Equity Partners as new backers. Sequoia Capital, Benchmark and Zeev Ventures returned, alongside individuals including Andrej Karpathy and former Disney CEO Michael Eisner.
What is DOS 2.0?
DOS 2.0 is Decart's core optimization infrastructure, designed to improve the speed and cost efficiency of AI processing across multiple chip architectures, including NVIDIA, Google and Amazon hardware.
How has Decart's valuation changed?
Decart raised $100 million at a $3.1 billion valuation in 2025. This round reprices the company at $4 billion on $300 million in new capital, a modest increase that reflects both investor confidence and acknowledged execution risk.