In the span of a week, a rift has cracked open in China’s fast-moving embodied intelligence sector.
On one side, startups like Tars and Spirit AI are riding high after announcing fresh funding rounds totaling hundreds of millions of RMB. On the other hand, GSR Ventures partner Allen Zhu has gone public with sharp skepticism, airing doubts about the sector’s commercial viability with an unusual degree of candor.
His critique touched a nerve. Responses poured in, and not all were friendly. But what stood out was what most of Zhu’s peers chose not to say. Industry voices like Matrix Partners China’s Zhang Ying and EngineAI founder Zhao Tongyang avoided pushing back on his central claim—that the sector’s path to commercialization remains murky. Instead, they chalked up the buzz to early-stage exuberance or a lack of long-term vision.
Zhu’s view isn’t exactly novel. Concerns that the space is “too early” or “commercially unclear” have been circulating for months.
ZhenFund partner Dai Yusen laid it out plainly at a mid-2024 event: general-purpose humanoid robots are too premature, which is why ZhenFund hasn’t invested in the category at all. Around the same time, Sinovation Ventures’ Lee Kai-Fu said much the same. Despite the hype, he argued, industry adoption would unfold at a crawl. Sinovation also opted out of investing in any humanoid robotics or embodied intelligence plays.
Most institutional investors and academic researchers expect the sector to reach maturity over the next five to ten years. That timeline has become something of a consensus—and a signal for venture capitalists to tread carefully.
“Embodied tech is highly uncertain. VCs aren’t comfortable making large or repeat bets,” a partner at a leading dual-currency fund told 36Kr. “In a USD 100 million or RMB 1 billion (USD 140 million) fund, backing two or three of these companies is already a stretch.”
A manager at a RMB-denominated fund with exposure to two embodied intelligence startups put it more bluntly:
“We placed multiple bets, but this is still a fuzzy space. There’s a real chance it’s a bubble.”
That caution has shaped the sector’s early fundraising dynamics. First rounds often rely on friends, family, and aggressive dollar funds. Then come corporate venture arms and strategic backers. But with limited new capital flowing in, state-owned funds are entering the fray earlier than usual.
According to 36Kr, some early-stage robotics investors are already looking to cash out. “Funds that backed projects in 2022 or 2023 are seeing 5x to 10x returns,” said one financial advisor. “Next year is all about product-market fit. That’s when the pretenders will be exposed.”
The dual-currency fund partner agreed: “Everyone’s watching DPI now. If you can’t show distributions, you can’t raise. Every time valuations jump a tier, you start thinking about partial exits—even with your own portfolio.”
Zhu might be accused of talking down the market on his way out, but privately, other investors echo his concerns.
Greed, fear, and a trillion-RMB ambition
It all kicked off on October 1, 2022, when Tesla unveiled Optimus, its humanoid robot. Like a butterfly flapping its wings across the Pacific, that reveal triggered a cascade of activity in China. Within a year, unicorns like Agibot and Galbot had emerged.
Embodied intelligence is often framed as the holy grail after artificial general intelligence: robots that can handle complex physical tasks with humanlike dexterity. Some analysts predict the global humanoid robotics industry could scale to a trillion-RMB market by 2035.
But high ceilings come with long runways—and enormous uncertainty.
Spirit AI co-founder Gao Yang said earlier this year that embodied intelligence is still at the equivalent of the “GPT-1 stage.” He expects it’ll reach “GPT-3.5” levels in about four years, implying that core intelligence is still a work in progress.
Others are even more skeptical. One USD fund partner suggested general-purpose humanoids may never fully materialize. “Matching 100% of human generalization might be impossible. But going one step beyond traditional AI—adding broader task flexibility—is probably within reach.”
A founder who has raised multiple rounds described the system architecture as having a “brain” and a “cerebellum.” The brain, responsible for decision-making and planning, is already deployable. The cerebellum—handling real-time motor control and path planning—is far less developed. “Foundation models probably won’t solve that part,” he said. “Maybe humans never will.”
So how are investors navigating a sector with sky-high risk and even higher upside?
For many, small checks are the answer. Backing early-stage rounds with modest capital allows them to stay in the game without overcommitting. Once the hype draws in follow-on capital, their early bets can be validated—at least on paper.
But rising valuations are starting to test that approach.
The commercialization bottleneck
Despite the hype, commercialization remains stuck in low gear—even as valuations continue to climb. Galaxea AI co-founder Xu Huazhe noted that investor sentiment has shifted dramatically: “Last year, they were curious but cautious. This year, the fear of missing out has blown up—they want skin in the game fast.”
Two flashpoints fueled that change, according to a partner at a dual-currency fund. “The embodied intelligence frenzy really took off after two key moments: the October 2024 product launch by US-based Physical Intelligence (PI), and Unitree Robotics’ breakout performance during the Lunar New Year.”
PI’s latest model showed off practical skills—folding clothes with a degree of generalization rarely seen before. It gave investors a concrete vision of what general-purpose embodied intelligence might look like, prompting many to double down on Chinese analogues. Unitree’s moment in the spotlight, meanwhile, was read as a sign of strong state backing for hard tech—especially humanoid robotics. That turbocharged expectations around fundraising and IPO potential, pulling investor attention toward other hardware startups in the space.
Still, while those events jolted sentiment, they didn’t solve the core problem: the industry has yet to find a clear path to revenue. As Zhu pointed out:
“The main buyers used to be academic research labs. Now we have a new customer profile: state-owned enterprises putting them in lobbies for display.”
Robotics expert Wang Tianmiao was even more direct in a September 2024 interview:
“[Despite] all this heat, people are starting to realize that humanoid robots today mostly just dance, flip, and patrol exhibition halls.”
Even supposedly practical applications are under scrutiny. Automotive factories are a favorite use case, partly due to Tesla’s own strategy. In 2023, several Chinese startups staged high-profile deployments in car plants. But insiders say many of these are more demo than deal. “They’re not sales—they are strategic tie-ups,” one investor said.
Spirit AI founder Han Fengtao told 36Kr that car factories aren’t ideal testbeds. “Embodied intelligence and humanoid robots are both nascent. Trying to combine two immature technologies and apply them to a very complex task is just too ambitious.”
What’s left is a familiar disconnect: ballooning valuations with little revenue to back them up.
“People are benchmarking across sectors,” said the dual-currency investor. “In most industries, a tenfold price-to-sales ratio is high. Some need price-to-earnings. But here, many of these companies haven’t even gone from zero to one. They haven’t found PMF. So how can they be worth tens of billions of RMB?”
Even so, not all bubbles are bad. Zhang argues they are a natural part of tech cycles. But without meaningful progress on commercialization and core tech, this bubble could go the way of the metaverse, community group buying, or coworking spaces—big on promise, short on delivery.
The real question, then, is whether to lean in or walk away.
And that may hinge on a single factor: whether foundation models in embodied intelligence can break through and reach its “GPT-3.5 moment.”
No one—not investors, not scientists—can say for sure. But the clock is ticking. And at this crossroads, the sector must pick a direction: forward or out.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Wang Fangyu for 36Kr.