The Person
Forty Milliseconds to Another World
Right now, if you open Decart's Oasis demo, a neural network somewhere on an H100 GPU starts conjuring a Minecraft-like world - fields, caves, sky - from your mouse movements alone. No game engine. No pre-rendered assets. No physics rules coded by hand. Just a model predicting the next frame, over and over, at 20 frames per second. Each frame takes 40 milliseconds. This is what Dean Leitersdorf has been building toward since he was doing his PhD at 2 in the morning after a full day at the Israeli military.
Leitersdorf is 27. He is the Co-Founder and CEO of Decart, a company that in May 2026 closed a $300 million Series C round at roughly a $4 billion valuation - with NVIDIA, Adobe, Toyota, Andrej Karpathy, and the former CEO of Disney on its cap table. Decart went from stealth launch to unicorn status in under twelve months. The company processes over 100 frames of high-definition video per second and runs inference at 1,600+ tokens per second, eight times the industry average. These are not marketing numbers. They are the reason NVIDIA backed a competitor rather than watch it succeed without them.
The pitch Leitersdorf makes is straightforward, and he makes it without flinching: the text box is not where consumer AI ends. "Today's ChatGPT isn't what consumers will keep on their phones long-term," he has said. "Something new is coming - an experience that will make the current one look primitive." He is describing his own product. He is also probably right.
"We could have built a great company to sell to Nvidia, but we're building an app for a billion users."- Dean Leitersdorf, Co-Founder & CEO, Decart
Origin
Three Degrees, One Military Unit, Zero Shortcuts
Leitersdorf grew up between Israel, Switzerland, and Silicon Valley - Palo Alto specifically, where he finished high school in two years before most of his classmates had decided on a major. He then enrolled at Technion, the Israel Institute of Technology in Haifa, simultaneously pursuing a BSc, MSc, and PhD in Computer Science. He completed all three in approximately five and a half years.
During those years, he also served in Unit 8200, the IDF's elite signals intelligence division - the one that produced Check Point, Waze, CyberArk, and about a dozen unicorns you've heard of. The schedule was not theoretical: 9 a.m. to 7 p.m. at the base, then home to do research. He finished his PhD at 23, winning the ACM PODC 2023 Dissertation Award for work on fast distributed algorithms and sparsity-aware computation. His doctoral supervisor was Prof. Keren Censor-Hillel.
His younger brother Orian later completed a Technion PhD at 21, breaking Dean's record. Their older brother Yoav runs YL Ventures, a prominent Israeli cybersecurity VC. Their parents are doctors and researchers. The Leitersdorf family is, by any reasonable measure, an outlier.
Dean Leitersdorf received his PhD at age 23 - one of the youngest in Technion's history - for research on fast distributed algorithms and sparsity-aware computation. His younger brother Orian later broke the record at 21.
At Unit 8200, Leitersdorf met Moshe Shalev, who would become Decart's Co-Founder and Chief Product Officer. They spent time after graduation - Leitersdorf also did a postdoctoral stint at the National University of Singapore - before co-founding Decart in late 2023. The $21 million seed round from Sequoia Capital and Zeev Ventures came almost immediately.
Career Arc
From Haifa to a $4B Company
The Products
Three Products, One Stack
Decart describes itself as "a fully vertically integrated AI research lab." That phrase earns its weight here: the company builds its own inference infrastructure, its own world models, and its own consumer applications - all the way from GPU-level optimization to the frame you see on your screen. The three pillars of this stack are DOS, Lucy, and Oasis.
Ultra-fast inference and training infrastructure. Chip-agnostic - runs on NVIDIA GPUs, Amazon Trainium, and Google TPUs. Already licensed by cloud providers and AI labs.
Live video transformation in under 30ms. Powers virtual try-on, live advertising, gaming, and social platforms. The only real-time world model in production at scale today.
The first real-time AI-generated open world game. Inspired by Minecraft. Takes keyboard and mouse input; a 500M parameter model generates every frame. No game engine - just inference.
DOS 2.0 solves a problem that increasingly matters as companies diversify away from single chip providers: it lets AI workloads switch between NVIDIA, Amazon Trainium, and Google TPUs without rewriting the model. This chip-agnostic capability is why cloud providers and AI labs are licensing it. The revenue it generates is what funded Decart's path to profitability three months after launch - an unusual milestone for a company at this stage.
Lucy's real-time video transformation is harder to appreciate without seeing it. Less than 30 milliseconds means the transformation responds before a human can consciously perceive the delay. Decart deploys Lucy across commerce (virtual try-on), live advertising, streaming platforms, and social media. It is, by their account, the only model doing this in production at this scale.
Oasis is the product that went viral - millions of people loaded up an AI-generated Minecraft-like world and discovered that a neural network could hold a coherent interactive environment together at real-time speeds. The 500 million parameter model was trained on Minecraft gameplay videos. It has no rules about gravity or collision. It just learned what comes next.
"Decart is trying to deliver delightful AI experiences - really trying to let people interact with their imaginations, and other people's imaginations, in a way that's never been possible before."- Dean Leitersdorf
Funding
From $21M Seed to a $4B Round
The funding history of Decart is not a gradual climb. It is an escalating series of bets by people who rarely agree on the same thing. Sequoia Capital led the seed. Then came NVIDIA, Adobe, Toyota, eBay Ventures, Andrej Karpathy (former OpenAI co-founder and Tesla AI director), Michael Eisner (former Disney CEO), and members of the Nintendo founding family. Radical Ventures led the $300M Series C. The total sits above $453 million.
What makes the Series C unusual is not the size - $300M rounds happen. It is the roster. NVIDIA investing in a company that makes chip-agnostic inference software is a statement. Adobe and Toyota investing in a world model startup suggests they see Lucy or DOS in their own product pipelines. Karpathy backing Decart personally - after spending years at OpenAI and Tesla building exactly the kind of systems Decart is trying to supersede - is harder to read as anything other than conviction.
The Ambition
Kilocorn or Bust
Leitersdorf has said publicly that he wants to build "a kilocorn" - a trillion-dollar company. He has also said he wants to build "a billion-user consumer app." These goals are not separate. His thesis is that the interface for AI has not been invented yet - that the text chat box is a placeholder, not a destination - and that the company that ships the right consumer experience at the right latency will capture the kind of user base that currently belongs to Netflix, YouTube, and TikTok.
"When we founded Decart, we decided we wanted to solve a truly massive problem," he said. "It took us a year to admit that this wasn't going to be a two-year startup. It would take five years to build something that, when people look at the world before and after, there's a real difference."
That framing matters. Leitersdorf is not pitching a feature. He is describing a category shift - the move from AI that answers questions to AI that generates persistent, interactive, sensory experiences. Decart's vertically integrated approach (owning the inference stack, the world model, and the application layer) is the structural bet that makes this conceivable. You cannot get to 40ms per frame by stitching together third-party components. You have to own the whole pipeline.
The company has 110 employees and offices in San Francisco, Palo Alto, and Tel Aviv. It is not chasing OpenAI or Anthropic on language models. It is not competing on benchmarks. It is competing on milliseconds.
Sidebar