Within a week a rift opened up in China’s fast moving embodied intelligence industry.

On the one hand, startups such as Tars and Spirit AI have announced new funding rounds totaling hundreds millions of RMB. GSR Ventures’ Allen Zhu, a partner at the firm, has made his company public.

Sharp skepticismwas expressed in a manner that was unusually candid.

His critique touched a nerve. The responses poured in and they were not all friendly. What stood out, however, was what Zhu’s peers chose to not say. Industry voices such as Matrix Partners China’s Zhang Ying, and EngineAI founder Zhao Tongyang chose not to challenge his main claim: that the sector’s path towards commercialization is murky. They attributed the buzz instead to early-stage enthusiasm or a lack long-term vision. Zhu’s opinion is not new. Since months, concerns that the space is ‘too early’ or ‘commercially unclear’ have circulated.

ZhenFund Partner Dai Yusen stated it clearly at a mid-2024 conference: general-purpose robots are premature. This is why ZhenFund didn’t invest in the category. Lee Kai-Fu, of Sinovation Ventures, said the same thing around the same time. He argued that despite the hype, industry adoption will unfold at a crawl. Sinovation has also decided not to invest in humanoid robots or embodied intelligent plays.

Most academic researchers and institutional investors expect the sector will reach maturity in the next five to 10 years. This timeline has become a consensus, and a signal to venture capitalists that they should tread carefully.

Embodied tech is highly unpredictable. A partner at a dual-currency fund said that VCs don’t like to make large or repeated bets. “In a fund of USD 100 million (USD 140 million), backing two or even three of these companies would be a stretch.”

One manager at an RMB-denominated RMB fund who had exposure to two embodied Intelligence startups said it more bluntly.

We placed multiple bets but this is still a hazy space. It’s possible that it’s just a bubble.

This caution has shaped early fundraising dynamics in the sector. First-round funding is often dependent on friends, family and aggressive dollar funds. Next, corporate venture arms and strategic investors are brought in. State-owned funds have stepped in earlier than usual due to the limited amount of new capital.

According to 36Krsome early-stage robots investors are looking to cash out. Financial advisors report that funds that backed projects for 2022 or 2023 have seen 5x to 10x return. Next year, it’s all about product-market match. It’s then that the pretenders are exposed.

According to the dual-currency funds partner, “Everyone is watching DPI right now.” You can’t raise money if you don’t have distributions. You start thinking about partial withdrawals every time valuations jump up a tier–even for your own portfolio.

Zhu may be accused of talking the market down on his way out but other investors share his concerns.

Fear, greed, and a trillion RMB ambition

All of this began on October 1, 2022 when Tesla unveiled Optimus. It was a butterfly flapping its wings across the Pacific. This announcement sparked a cascade in China, like a butterfly flapping across the Pacific. In less than a year, unicorns such as Agibot and Galbot were born.

Embodied Intelligence is often framed after artificial general intelligence as the holy grail: robots with humanlike dexterity that can perform complex physical tasks. Some analysts predict that the global humanoid robots industry will reach a trillion RMB market by 2035.

But with high ceilings comes long runways — and enormous uncertainty.

Spirit AI’s co-founder Gao Yang stated earlier this year that the embodied intelligence was still at “GPT-1” stage. He expects that it will reach “GPT 3.5” levels in four years. This implies that core intelligence is a work-in-progress. Others are even more sceptical. One USD fund partner said that general-purpose humanoids might never materialize. “Matching 100 percent of human generalization may be impossible.” It is possible to go beyond traditional AI and add broader task flexibility. The cerebellum, which is responsible for real-time motor control, path planning and decision-making in real time, is much less developed. “Foundational models probably won’t be able to solve that part,” said he. “Maybe we will never.”

How do investors navigate a sector with high risk and even greater upside?

Small checks are the solution for many. They can stay in the game by backing early-stage rounds using modest capital. Once the hype brings in additional capital, their initial bets will be validated – at least on paper.

But rising values are beginning to test this approach.

Commercialization bottleneck

In spite of the hype, commercialization is stuck in low gear – even as valuations continue climbing. Xu Huazhe, co-founder of Galaxea AI, noted that investor sentiment had changed dramatically: “Last Year, they were curious but conservative. The fear of missing out is exploding this year. They want to get in on the action fast.

According to a partner with a dual currency fund, two flashpoints have fueled the change. “The embodied-intelligence frenzy really took hold after two key moments, namely the October 2024 launch of the US-based Physical Intelligence and Unitree Robotics’ breakout performance during Lunar New Year.” Investors were able to see a concrete example of what embodied intelligence could look like. Many jumped on Chinese equivalents. Unitree’s moment of fame was interpreted as a sign that the Chinese government is strongly supporting hard tech, especially humanoid robots. This boosted expectations for fundraising and IPOs, attracting investors to other hardware startups.

While these events jolted the sentiment, they did not solve the core issue: the industry still hasn’t found a clear way to revenue. Zhu noted:

The main buyers used to academic research laboratories. We have a new type of customer: state-owned companies displaying them in their lobbies.

In a September 2024 article, robotics expert Wang Tianmiao said it even more directly:

[Despite] with all this heat, many people are beginning to realize that most humanoid machines today do nothing but dance, flip and patrol exhibition halls. Tesla’s strategy is partly to blame for the popularity of automotive factories as a use case. In 2023, a number of Chinese startups will stage high-profile deployments at car factories. Insiders claim that many of these deals are more like demonstrations. Investors said that these are not sales, but strategic partnerships. Spirit AI founder Han Fengtao said to 36Kr

that car factories weren’t ideal testing grounds. “Both embodied intelligence and humanoid robotics are in their infancy. It is simply too ambitious to try to combine two immature technology and apply them to such a complex task.

The result is the same old disconnect: inflated valuations without any revenue to support them.

The dual-currency investment said that people are benchmarking across industries. In most industries, a price-to-sales rate of tenfold is high. Some companies need to know the price-to-earnings ratio. Many of these companies are still at zero. They haven’t discovered PMF. How can they be worth 10s of billions RMB?

Despite this, not all bubbles have a bad effect. Zhang says they are a normal part of tech cycles. This bubble could end up like the metaverse, group buying communities, or coworking spaces – full of promise but short on delivery.

Ultimately, the question is whether you should lean in, or walk away.

This may depend on one factor: whether the foundation models of embodied intelligence will break through to reach their “GPT-3.5 Moment.”

Nobody can say for certain, neither investors nor scientists. The clock is ticking. At this crossroads the sector has to choose a direction – forward or out.

KrASIA Connection presents translated and adapted material that was originally published in 36Kr. This article has been written by Wang Fangyu, for 36Kr.