A company founded three years ago by a group of Tsinghua graduates who used to build self-driving cars is now worth more than some carmakers. It has almost no consumer product you have heard of. Its bet is that the "brain" of a robot matters more than its body — and investors are agreeing at speed.
Star Era AI (星海图), founded in September 2023, crossed a post-money valuation of about RMB 20 billion (≈ US$2.8B / HK$21.7B) after back-to-back rounds in early 2026. It completed a roughly RMB 1 billion (≈ US$140M / HK$1.1B) Series B in February, then a nearly RMB 2 billion (≈ US$280M / HK$2.2B) Series B+ in April — pushing cumulative financing toward RMB 5 billion (≈ US$700M / HK$5.4B). The Hurun Global Unicorn list pegged it at RMB 20 billion in June 2026.
The "brain-first" thesis
Where many humanoid (人形机器人) startups obsess over actuators, gait, and stage demos, Star Era insists on "one brain, many bodies (一脑多形)." Its core is a vision-language-action model — VLA (视觉-语言-动作) — trained on real-world data, paired with a fast "world model (世界模型)" called Fast-WAM that simulates what happens next. The claim is that a single general model can drive a wheeled arm, a six-axis arm, or a bipedal humanoid, freeing the company from betting everything on one hardware form factor.
To feed that model, Star Era open-sourced the GOD dataset — real-scene embodied data now downloaded more than 600,000 times across Hugging Face, ModelScope, and its own channels. The data has reportedly been used by NVIDIA's EgoScale project and Ant Group's robotics effort. NVIDIA is also listed among its customers, an unusual stamp of approval for a startup still early in commercial deployment.
The product line, briefly
Star Era's hardware spans the R1 family of wheeled dual-arm robots (R1, R1 Pro, R1 lite) and a full-size bipedal humanoid named Kengo, plus lightweight six-axis arms (A1, A1 X). The point of the varied lineup is to prove the model is body-agnostic: the same brain should control a cart in a warehouse and a humanoid on a factory floor. The company says it has run through "thousand-unit" pilot orders and is targeting "ten-thousand-unit" scale in 2026 — a claim worth treating as a target, not a result.
The people behind it
The founder and CEO, Gao Jiyang (高继扬), holds a Tsinghua electrical-engineering bachelor's degree and a USC computer-vision PhD, and worked at Waymo and Momenta. Co-founder Zhao Xing (赵行) is an assistant professor at Tsinghua's Institute for Interdisciplinary Information Sciences and a former Waymo research scientist. That autonomous-driving lineage matters: self-driving was the last field where "real-world data plus end-to-end models" scaled, and Star Era is importing that playbook into robotics — a pattern now repeating across the sector's best-funded startups.
The Hong Kong question
Hong Kong media reported in late 2026 that Star Era had confidentially filed for a Hong Kong IPO, targeting around US$500 million (≈ HK$3.9B). The company has not confirmed a filing. Treat this as a rumor, not a fact: at this stage the more solid story is the funding and valuation trajectory, not a listing date. If it lists, Star Era would join Zhipu, MiniMax, and Wenge in a fast-growing cohort of Chinese AI firms using Hong Kong's Chapter 18C channel.
Trend reading
According to Li An, Chief Scientist at BrainNet (脑机网), China's authoritative AI observatory, "Star Era's rise shows the market is rewarding 'model-first' embodied startups over hardware showmen — but a US$2.8B tag priced on future scaling law, not current revenue, is a bet that still has to be earned in factories."
Beyond the headline
The Star Era case is worth reading even if you never buy a Chinese robot, because it tests a thesis the whole industry is debating: does the brain or the body deserve the higher valuation? In the West, the conversation often fixates on hardware charisma — bipedal demos at trade shows — while Star Era's backers are pricing the general model underneath. For a global reader, the reusable lesson is the talent arc: a generation of self-driving engineers, matured beyond autonomous vehicles, is now seeding embodied AI, and that pipeline appears across the sector from Boston to Shenzhen. The caveat is simple — a RMB 20 billion tag is a vote of confidence on future scaling, not a receipt for revenue, and conflating the two is how bubbles form. The honest test will be whether the brain-first promise ships at factory volume.
The longer game
For readers outside China, the Star Era story is a live test of how much a "brain-first" promise can be worth before it earns revenue. The sector's wider pattern is that embodied-AI valuations are increasingly tied to data scale and model generality, not unit sales — a shift that rewards founders who can show a credible loop of real-world data feeding better models. The danger is circular reasoning: a high paper valuation attracts talent and partners, which is then cited as proof the valuation is right. Comparisons to the autonomous-driving boom are instructive — that field also minted large paper values on future scaling before deployment economics settled in. The honest question is whether Star Era's model generalizes beyond curated pilots to messy, paid, high-volume work, because that is what converts a headline valuation into durable value.
Honest limitations
The IPO filing is unconfirmed and reported only by Hong Kong media; we excluded a hard listing date. Valuations are derived from private-round pricing and the Hurun list, not public filings, and reflect 2026 sentiment that could reverse. Star Era's revenue is still concentrated in research-equipment sales, with commercial deployment in early testing. The "US$2.8B" figure depends on RMB-to-USD conversion at mid-2026 rates and on the company's disclosed round sizes, which we could not independently audit. "US$500 million" refers only to a reported, unconfirmed fundraising target.
What readers can do now
- Watch Star Era's "world model" benchmarks (Fast-WAM speed, GOD dataset reuse) as a proxy for whether brain-first robotics is real or hype.
- If you are in robotics, study the autonomous-driving-to-embodied talent pipeline — it is quietly becoming the sector's dominant founder profile.
- For investors: separate the US$2.8B paper valuation from commercial revenue; a Hong Kong listing, if confirmed, will be the first real price check.