Private companySeries A

Reactor

Data & simulation · US

reactor.inc

Ex-Apple Vision Pro and Luma AI founders building the real-time inference infrastructure layer for world models — a developer platform for running generative video, world models and robot policies at sub-50ms latency.

Reactor is the developer platform (SDK + API) for real-time generative media and world models. Its proprietary inference engine, globally distributed GPU cloud and low-latency streaming layer serve models at 60+ FPS and sub-40ms latency — against a 400ms+ industry average — via a single unified API. It hosts third-party models (MiniMax, Vidu, LTX, Alibaba, ByteDance) and powers real-time interactive video for media/entertainment customers. In physical AI, Reactor Robotics runs world models as closed-loop simulated environments to train/evaluate robot policies, and serves robot policies from a distributed GPU cloud with observations/actions streamed in real time, hosting policies from NVIDIA, Black Forest Labs, Robbyant and Xiaomi. Named customer: Overworld, a real-time world-model lab. Backed by Lightspeed, WndrCo, NVIDIA (NVentures), Amplify Partners, Sky9 Capital, Sapphire Ventures and FPV Ventures.

Research mapping

Physical AI / Sim-to-Real, Digital Twins & Validation (L9) / Data & simulation

Evidence boundary

Original analysis and independent corroboration remain under development. Commercial scale and durable technical advantage remain under review.

The sub-50ms real-time inference/streaming layer: frontier world-model and robotics labs can build the models, but serving them interactively demands purpose-built low-latency GPU infrastructure (Reactor’s custom kernels + global GPU cloud) — the toll-road between model and production.

Infrastructure layer for real-time world models (pa-9): runs world models as closed-loop simulated environments for robot-policy training/evaluation and serves robot policies via a globally distributed GPU cloud. Hard to copy because sub-40ms serving at scale requires in-house GPU kernels plus a global GPU network — not model weights.

Platform & Developer Teams

Overworld (named customer — real-time world-model lab)

Alberto Taiuti — Co-founder & CEO — Former CTO and co-founder of Luma AI; former technical lead on Apple Vision Pro.

Bryce Schmidtchen — Co-founder & CTO — Former technical lead on Apple Vision Pro (kernel optimization, computer vision for AR/VR).

Source
Last round
$59M Series A (May 2026)

Lightspeed Venture Partners (lead), NVIDIA (NVentures), Sapphire Ventures, WndrCo, Amplify Partners, Sky9 Capital, FPV Ventures

Mapped funds and investors

Direct relationships supported by published evidence.

Reported holdings as-of

No sourced relationships mapped yet.

Disclosed investment history

No sourced relationships mapped yet.

Showing up to four relationships in each group. ETF weights describe the fund’s portfolio, not ownership of this company. Private investment history does not establish current ownership.

  1. Product

    Reactor cuts Wan VAE decoder time per chunk from 133.8 ms to 60.2 ms (2.2x) at equal output quality for real-time world models

    Source
  2. Product

    Reactor publishes the FastH3-Omni engineering deep dive (real-time video + synchronized audio; 5s 1344x768 clip 42s -> 14.7s on 4xB200)

    Source
  3. Financing

    NVIDIA (NVentures) and Sapphire Ventures join Series A; $74M total raised

    Source
  4. Financing

    $59M Seed + Series A led by Lightspeed (Amplify Partners, WndrCo, Sky9 Capital, FPV Ventures); emerges from stealth

    Source
The weekly brief

Get a weekly digest of what’s changing.

Research updates, thesis changes and company developments. One email a week.