Galbot's Wang He: Embodied AI's "ChatGPT Moment" Could Arrive by End of 2028

A quantifiable claim

Per IThome citing CCTV Finance, at the 2026 World AI Conference (WAIC 2026) Galbot founder and CTO Wang He gave a strikingly concrete forecast: trained on massive integrated data, embodied-AI foundation models can already hit 70%–80% success on tasks with no task-specific training — a level he likens to "the conversational quality of early digital models."

The key enabling conditions, in his view: a strong pretrained model plus an efficient post-training paradigm. On current data-accumulation speed and model-convergence trends, the field could reach embodied AI's "ChatGPT moment" before the end of 2028.

What the "ChatGPT moment" means

The "ChatGPT moment" is a popular metaphor in tech and investment circles for a technology that suddenly breaks through, jumps out of its niche, and reaches the public fast enough for ordinary people to feel its value directly. Applied to embodied AI, it means robots stop being special-purpose lab or production-line equipment and enter broader real-world scenarios with "good-enough" general capability.

Against peers' views

The view echoes others in embodied AI — Unitree's Chen Li has also discussed embodied AI's "ChatGPT moment," framing it as "two 80%s to be met." Different teams independently anchoring the inflection with a "success-rate threshold + a date" signals that the field is shifting from "can we build a demo" to the more engineering-minded "can general success rate cross the usable line."

Why it matters

Two actionable signals for anyone tracking AI deployment: first, the yardstick for embodied AI is converging on "general success rate on untrained tasks," not single-point stunts; second, the timeline is now pinned near end-2028 — if it holds, the window to position around data, post-training paradigms, and real-scenario data loops is the next two to three years. Whether it lands still hinges on data scale and post-training efficiency converging on schedule.