Deep Research

Decart AI (private)

4 min read

Snapshot

  • Status: Private — no 13F position; this is a private book holding
  • Sector: Real-time generative world models / interactive AI video, plus GPU-inference optimization software
  • Latest known valuation: **~300 M Series C announced May 18, 2026, led by Radical Ventures
  • Atreides involvement: Confirmed — named participant in the May 2026 Series C (CTech/Calcalist; Decart’s own announcement). Not in the disclosed list for the Aug 2025 round at $3.1 B — the “2025 round” attribution sometimes repeated is unsupported.
  • HQ: Tel Aviv, Israel (founded 2023 by Dean Leitersdorf, CEO, and Moshe Shalev, CPO — Unit 8200 alumni)
  • One line: an AI lab building world models that generate and transform interactive video in real time (milliseconds, not minutes), funded by a profitable inference-optimization software business.

Business Overview

Decart runs three product lines on one technical foundation — extreme low-latency inference. DOS is a GPU/accelerator optimization layer licensed to cloud providers, AI labs, and hyperscalers (the Series C coverage emphasized using it to run workloads on non-NVIDIA silicon, including Amazon’s chips — “breaking the NVIDIA lock-in,” even as NVIDIA invested in the round). Lucy is a real-time world-editing model that transforms live video streams (try-on, advertising, streaming). Oasis — the 2024 viral demo of a fully AI-generated playable Minecraft-like world — has evolved toward robotics and physical-AI simulation. MirageLSD (Jul 2025) claimed the first infinite, zero-latency real-time video-to-video generation.

Unusually for a frontier lab, Decart says it has been roughly profitable from early on via DOS licensing, having spent under 53 M seed/A (2024, Sequoia, Zeev, Benchmark; 100 M Series B at 300 M Series C at ~450 M.

Atreides’ Involvement

  • Series C, May 2026 — confirmed. Atreides Management is named among participants in the 4 B valuation. Check size not disclosed; Radical Ventures led.
  • Aug 2025 Series B (3.1 B) — no evidence. Disclosed participants were Sequoia, Benchmark, Zeev, and Aleph. Reports placing Atreides in a “2025 round at ~$3.1 B” appear to conflate the two rounds; the verifiable entry is May 2026.
  • Vehicle — inferred, not disclosed. Special Circumstances Series X ($82.6 M Form D, filed May 14, 2026 — four days before the round was announced) is the closest timing match, though the same series window also fits Cognition’s Series D; the Valor Atreides AI JV is the other candidate. SPV underlying assets are never disclosed — inference only.

Why Atreides Owns It

Decart is the game-engine thesis one layer earlier. Atreides’ public book holds Unity (#3 equity position, calls stacked on common) and Roblox as levered bets that real-time 3D engines win from AI — generative content collapses creation cost while engagement monetization compounds. Decart attacks the same end-state from the generative side: if world models can render interactive, persistent environments in real time, they either become the next engine layer or get absorbed into the current one — both outcomes Atreides is positioned for. There is a “wafers” angle too: DOS is a software wedge against accelerator scarcity, making non-NVIDIA silicon usable for frontier workloads — directly adjacent to Baker’s Amazon/Trainium enthusiasm (“the most undervalued AI chip”). The capital-efficiency profile (near-profitable, <$10 M burned pre-2025) is a rarity among AI labs and limits downside relative to typical frontier-lab burn.

Risks

  • Frontier-lab collision: Google (Genie), OpenAI (Sora), and World Labs are all building world models with vastly more compute; Decart’s latency edge must persist against scale.
  • Two-business sprawl: an inference-optimization licensing business and a consumer-facing world-model lab share infrastructure but not go-to-market; focus risk for a ~200-person company.
  • DOS dependency: profitability rests on licensing deals with a handful of hyperscalers/labs who are all building in-house optimization and could internalize the capability.
  • Demo-to-revenue gap: Oasis/Mirage virality has not yet demonstrated durable consumer or enterprise monetization; run-rate revenue is undisclosed.
  • Geopolitical/talent concentration: core R&D in Israel; founder-centric technical leadership.
  • Late, small, illiquid: Atreides entered at ~$4 B (an 8x markup in ~18 months) via undisclosed vehicles, with no mark until an exit event.

Sources

Backlinks

Graph View