Galaxea AI raises $291 million, valuation hits $2.9B with Apple supplier among investors.
Galaxea AI's newest humanoid robot can run on a single processor and costs less than a family sedan. It still has no idea when the embodied AI moment arrives.
The Beijing-based robotics startup closed a Series B+ funding round in early April, raising 2 billion yuan (roughly $291 million) from nearly 20 investors including Lens Technology (300433.SZ), a major Apple supplier listed on the Shenzhen exchange, and state-backed funds. The round pushed Galaxea's valuation above 20 billion yuan — about $2.9 billion — according to Caixin Global and Gasgoo, which reported the figures in Chinese. (A GNews wire story cited $29 billion — almost certainly a unit confusion in English-language reporting; the Chinese-language primary sources are consistent at 20 billion yuan or above.) Yicai Global, also cited as a source in this article, reported a valuation of CNY 200 billion (~$29 billion) — directly higher than the CNY 20 billion figure from Caixin and Gasgoo. This article uses the $2.9 billion figure based on primary-source reporting; the discrepancy is noted.
That makes Galaxea one of the most valuable robotics startups on the planet, three years after its founding, with no recorded revenue and a CFO who is publicly questioning whether the sector's foundational technology is ready. Luo Tianqi told Chinese outlet ebrun.com that China still lacks the physical data infrastructure for a true embodied AI breakthrough. "The data is not ready," he said. "The physical world is much more complex than language."
The company's own finance chief, in other words, is publicly tempering expectations. The valuation is not.
This is the central contradiction in the humanoid robotics race right now: investors are paying winner-take-all prices for companies whose own leadership is still arguing about when the game actually starts.
The hardware is further along than most peers. The timeline still isn't.
Galaxea is not purely a humanoid story. Its wheeled dual-arm robot platform entered large-scale deployment in the second half of 2025, with batch application feedback from automotive and logistics customers. Multiple media outlets reported at the end of 2025 that Galaxea had secured orders or letters of intent for thousands of units, according to CNMRA. That separates it from humanoid-only companies with zero shipped hardware.
The R1 humanoid, priced at $44,500 to $64,000 per unit according to Yahoo Finance, is designed to be cheap enough to deploy in volume. Its Fast-WAM action model achieves single-step inference in 190 milliseconds, versus 800 milliseconds for traditional wide-action models, according to Gasgoo. Galaxea says it can run the full neural policy on a single chip.
At those prices, Galaxea would need to sell roughly 48,000 to 65,000 R1 units to justify its current $2.9 billion valuation at today's revenue run rate — assuming no further dilution. Its target is 1,000 units by end of 2025, per Yahoo Finance. The gap between where the company is and where the valuation implies it will be is the bet investors are making.
Galaxea has raised nearly 3 billion yuan across approximately 10 funding rounds since September 2023, including a 1 billion yuan Series B in February 2026 that valued the company at 10 billion yuan, according to Caixin. The valuation nearly doubled in two months. That pace is not unusual in Chinese robotics right now, but it outpaces anything Galaxea has shipped.
The money is piling in before the data is ready
Galaxea's registration conversion in late January 2026 from a Limited Liability Company to a Joint-Stock Company (foreign-invested, unlisted) suggests preparation for a public listing. The company is building the financial infrastructure of a public company before it has the revenue of one.
Luo's candor about data readiness is notable because it came from the CFO, not a researcher whose job is managing expectations. "You can't scale what you can't measure," he told eBrun.com. "The physical world doesn't generate clean training data the way text does. Someone has to build those pipelines first."
That "someone" is the open question. Unitree, Fourier, and a dozen other Chinese robotics startups are burning capital at a similar pace, all betting on a market that currently consists of pilot programs and demonstration units. If the data infrastructure problem is years from solved, the unit economics of humanoid deployment are similarly distant.
Lens Technology's participation gives Galaxea a credible path to actually building at scale — Lens makes iPhone cover glass and has the manufacturing depth to be a serious production partner. That is not nothing. But producing a robot and deploying thousands of them into actual workflows are different problems. The automotive and logistics customers reportedly ordering Galaxea's wheeled platform are pilots, not volume buyers. The transition from pilot to production contract is where most automation plays slow down or stall.
If Galaxea ships at its targets and reaches profitability in 2026, it validates a new pricing model for robotics: cheap enough to buy, smart enough to adapt, deployed in sufficient volume to generate the data needed to get smarter. That is the virtuous circle every robotics investor is betting on.
If it doesn't, the $2.9 billion valuation becomes a ceiling, not a floor, and the correction flows to every pre-revenue humanoid startup priced on the same assumptions. The embodied AI GPT moment may come. It may come from Galaxea, Unitree, Figure, or someone who hasn't raised yet. What Luo's own words suggest is that it is not coming in 2026 — regardless of what the valuation says.
Yicai Global reported a valuation of CNY 200 billion ($29B), higher than the CNY 20 billion cited by other outlets.
Correction: An earlier version of this article cited a $29 billion valuation from a GNews wire story. The primary Chinese-language sources consistently report the valuation as above 20 billion yuan, approximately $2.9 billion. This article uses the $2.9 billion figure based on primary-source reporting.
Sources: