Decart Raises $300M as Low-Latency AI Becomes Infrastructure

Decart, the San Francisco-based AI infrastructure startup founded by Dr. Dean Leitersdorf and Moshe Shalev, just raised a $300M Series B at a nearly $4B valuation. Radical Ventures led the financing, with participation from NVIDIA, Adobe Ventures, Toyota Ventures, Benchmark, Sequoia Capital, Valor Equity Partners, Zeev Ventures, Atreides Management, and eBay Ventures. Individual investors included Andrej Karpathy and members of the Yamauchi family connected to Nintendo.

The company develops low-latency AI infrastructure designed for real-time world models, AI-generated environments, and persistent interactive systems. Decart’s core products include DOS, an inference and training optimization stack, Lucy for real-time video transformation, and Oasis, a real-time AI-generated interactive environment. The funding matters because infrastructure is quietly becoming the most important layer in AI. Consumer applications capture attention while infrastructure captures economic gravity. Investors are increasingly separating companies demonstrating AI features from companies building the computational systems future AI products will depend on.

Decart’s rise also reflects a broader shift across enterprise AI, robotics, autonomous systems, gaming, and media infrastructure. The market is moving beyond static generative AI outputs toward persistent AI environments that require dramatically faster computational performance and lower latency.

What Happened

Decart announced a $300M Series B in May 2026, bringing total funding to more than $450M since launching in 2023. The financing pushed Decart’s valuation close to $4B less than a year after its previous $100M round valued the company at $3.1B. That kind of acceleration changes how investors evaluate a company because early-stage hype usually burns hot for 6 months before reality walks into the room carrying infrastructure costs and operating pressure. Decart appears to have crossed into a different category entirely, with the market treating the company less like an experimental AI startup and more like foundational infrastructure.

Dr. Dean Leitersdorf and Moshe Shalev built Decart around a thesis that now feels increasingly obvious in hindsight: latency is becoming one of the defining constraints in modern AI systems. Faster systems create more immersive experiences, more persistent environments, and more commercially viable AI applications. Most AI systems still operate like a vending machine where a prompt goes in, a response comes out, and a pause sits awkwardly in the middle. Decart is building systems designed to operate continuously in real time, and that distinction changes entire categories of software behavior.

Why Decart Matters

Decart’s products sit directly inside the infrastructure layer many companies still underestimate. DOS is Decart’s optimization stack for low-latency inference and training workloads tied to LLMs, agents, and world models. Lucy focuses on real-time video transformation with near-zero latency, while Oasis became the company’s breakout signal after reaching 1M users in 72 hours as a fully playable real-time AI-generated environment.

That growth matters because AI engagement is becoming harder to fake. The market has already entered the phase where novelty alone no longer sustains adoption. People experimented with AI image generators because they were curious, but sustained engagement now requires responsiveness, immersion, and utility. Decart appears to understand where the market is heading next: AI systems that feel less like tools and more like environments.

The company is also positioning itself at the intersection of several massive infrastructure markets simultaneously, including robotics, autonomous systems, enterprise AI, gaming, and interactive media. Few startups manage to establish technical relevance across multiple infrastructure categories without collapsing under their own ambition.

The Market Context Behind the Funding

The AI market spent the last several years obsessing over models, but attention is now shifting toward inference optimization, computational efficiency, and real-time AI systems. That shift explains why strategic investors like NVIDIA participated in Decart’s round because hardware companies understand something the broader market is slowly learning: model capability means very little if systems cannot operate efficiently at scale.

This is where infrastructure companies become dangerous competitors because they are not dependent on a single application trend. They become embedded inside broader computational ecosystems. Decart’s partnerships with Comcast, ElevenLabs, Amazon, and NVIDIA suggest the company is already moving beyond experimental deployments. Enterprise infrastructure relationships rarely happen because technology looks impressive in a demo. They happen because performance gains create measurable economic advantages.

Milliseconds matter in AI infrastructure because lower latency affects responsiveness, compute utilization, scalability, operating costs, and user retention simultaneously. That becomes especially important as enterprises move toward persistent AI environments rather than isolated chatbot interactions. There is also a geopolitical layer underneath this market shift. Decart reflects the growing influence of Israeli-founded AI infrastructure companies expanding into U.S. enterprise ecosystems, while the founders’ backgrounds in Unit 8200 continue a broader trend of defense-trained technical operators moving aggressively into AI infrastructure and computational systems.

What This Signals for AI Infrastructure

The market is beginning to divide AI companies into 2 categories. The first category builds features while the second builds dependencies. Features create attention cycles, but dependencies create market power. Decart increasingly looks like the second category.

That distinction matters because infrastructure companies tend to compound quietly before becoming unavoidable. Most people never think about computational plumbing until everything above it stops functioning correctly. Investors understand this dynamic extremely well, which is why infrastructure companies often command premium valuations long before the broader market fully understands their importance.

Decart is also entering the market at a moment when enterprise AI buyers are becoming more skeptical. Enterprises no longer want vague promises attached to oversized GPU budgets. They want measurable performance improvements tied directly to economics and operational capability. The companies winning this next phase of AI will likely be the ones capable of reducing latency, improving efficiency, and enabling real-time interaction at scale. That is the lane Decart is trying to own.