Foxconn skips building its own robot brain, bets on NVIDIA GR00T 2.0 and factory data flywheel
lukas_m_ziegler · x · 2026-10-31
At HHTD in Taipei, the author interviewed Leo Guo, who leads physical AI and robotics at Foxconn, about the company's robotics strategy:
- No in-house foundation model: while half the robotics world races to train its own foundation model, Foxconn partners with NVIDIA and runs on GR00T 2.0.
- The edge is factories: Foxconn has a huge range of real industrial scenarios, collecting real production data and post-training the model on tasks that actually happen on a production line.
- A data flywheel: that data flows back to NVIDIA to improve GR00T, which returns stronger and collects even better data — an asset at a scale almost nobody else has.
The takeaway: Foxconn's advantage isn't the model, it's manufacturing scenarios and data.
More from Companies & People
- Google DeepMind-era research scientist Jason Baldridge moves to Substack — jasonbaldridge · 2026-10-31
- Tang Luong & Salakhutdinov headline Seoul's JoongAng Forum 2026 on AI and growth — rsalakhu · 2026-10-31
- Cursor Opens Ambassador Program with SpaceXSI, Fund AI Meetups — msharmas · 2026-10-31
- AI scholar Vincent Conitzer opens postdoc position for CoRe Safe AI — conitzer · 2026-10-31
- Gary Marcus re-ups 9 reasons OpenAI could become the WeWork of AI — GaryMarcus · 2026-10-31
- Anthropic wealth could seed a wave of centimillionaire angels, VCs predict — ashleymayer · 2026-10-31