Guanglun Intelligence

Beijing embodied-AI simulation & synthetic-data "data factory" — physics-solver training grounds for robots; world's first embodied-data unicorn [5].

1. Core Product / Service

Guanglun Intelligence (光轮智能, "Lightwheel AI"; Beijing Haidian, founded January 2023) is a physical-AI simulation and synthetic-data infrastructure company — it does not build robots. It operates a "digital parallel factory" where robot and embodied-AI companies train models thousands of times in parallel on a self-developed physics solver and simulation stack [4][6].

Its three product layers:

• Simulation — a high-precision multi-physics solver reproducing conditions from −40°C to over 1,000°C and surfaces from smooth glass to rugged mining terrain [4][6].
• Data — a synthetic-data engine plus a human-video data engine, described as the world's largest non-ontology data engine; the company reports more than 80% of leading embodied-AI teams use its simulation assets and synthetic data [4][5].
• Evaluation — RoboFinals, billed as the world's first industrial-grade simulation evaluation platform for robotics [6].

Its LeIsaac simulation workflow is included in Hugging Face's official documentation, and it has partnered with Siemens on SimFoundry, RoboFinals and RoboStack for industrial embodied simulation [6][7].

2. Target Users & Pain Points

• Embodied-AI / robot companies — reported customers/partners include ByteDance, Alibaba, agibot|AGIBOT, Galbot, NVIDIA, Google, figure-ai|Figure AI, 1X, Toyota, Bosch, BYD and Geely [4][6].
• Autonomous-driving-adjacent and industrial players migrating to robotics.

Pain solved: robot data is scarce and expensive — the company argues embodied AI needs at least 1,000× the data of autonomous driving. Guanglun claims to cut robot development cycles from 3–6 months to 2–3 weeks and training costs by over 90% via synthetic simulation [4][6].

3. Competitive Landscape

Guanglun's differentiation is vendor-neutral synthetic-data + evaluation infrastructure for embodied AI specifically, rather than being a model or hardware vendor [6].

4. Unique Observations

• The Mercor analogy was a category error — and that's the point. Jimmy's 2026-09-24 research first analogized 光轮智能 to mercor (human expert labor), but the facts show it is synthetic simulation data, not human labor — the correct analogs are NVIDIA Isaac and scale-ai's synthetic data. The confusion itself signals how new the "embodied-data infrastructure" category is [3][4].
• "World's first embodied-data unicorn" is a picks-and-shovels bet. The ¥1B A++/A+++ round crowns Guanglun as the first pure-play embodied-data unicorn, validating data — not models — as the embodied-AI bottleneck (the 1,000× data-demand gap) [5][6].
• The synthetic-vs-real data divide. Guanglun's synthetic-first approach sits opposite galaxea's 100M-hour real-world data strategy; whichever wins, both treat data volume as the moat. The Siemens partnership suggests Guanglun is also targeting industrial buyers beyond pure robotics [4][7].

5. Financials / Funding

• Founded: January 2023, Beijing Haidian [4].
• Funding: ~¥1B (10亿元) across A++ and A+++ rounds, making it the world's first embodied-data unicorn; investors include New Hope Group, Dingbang Innovation, AUX Group, Capstone, JIC Technology Investment, and Guofang Innovation [5][6].
• Revenue: 2025 revenue reportedly grew ~10×, with Q1 2026 revenue expected to surpass all of 2025 [6].

6. People & Relationships

• Founder / CEO: Xie Chen (谢晨) — former autonomous-driving simulation lead at NVIDIA, Cruise and NIO [4][6].
• President / co-founder: Yang Haibo (杨海波) [6].
• Partners: Siemens (industrial simulation), Hugging Face (LeIsaac documentation) [6][7].
• Customers / peers: galaxea, agibot, figure-ai and other embodied teams; NVIDIA, Google, Toyota, Bosch, BYD, Geely [4][6].