2026-10-14 09:52

Lead-in: Technical pathways remain unconsolidated, while components have become a safe haven for capital. A group interview at this year’s Bund Summit left a lasting impression.

A group interview at this year’s Bund Summit left a lasting impression. On that day, all interviewees were AI-focused investors from prominent institutions. During the self-introduction segment, at least three investors proactively stated that embodied intelligence was one of their firm’s primary investment focus areas.

It wasn’t until Most Talk FunTalk mentioned recent rumors that regulatory authorities had tightened IPO scrutiny on embodied intelligence startups that the atmosphere shifted dramatically.

The room went almost instantly awkward.

Investors then began reassessing the development trajectory of embodied intelligence, acknowledging that the industry remains in its very early stages and that technology has yet to converge.

This question became a watershed moment in the entire interview. That day would later be regarded as a turning point for the entire sector.

Indeed, just that morning, news broke that regulators had issued informal “window guidance,” including companies with recurring revenue, narrowing losses, or genuine technological innovation, within the scope of humanoid robot IPO review.

Before this, the industry had already endured at least a year of intense speculative fervor—startups sprouted overnight, funding valuations skyrocketed like sesame seeds growing taller each day.

In the first half of 2026 alone, total fundraising in China’s embodied intelligence sector surpassed 90 billion RMB—five times the amount raised during the same period in 2025. The “hundred-billion club” in this space continued expanding, with over 20 companies achieving valuations exceeding 10 billion RMB within just six months.

For a sector still in its nascent technical phase, such enthusiasm in primary markets was largely driven by imagination about future potential. But as market sentiment kept rising, two pivotal events occurred.

In May, UBTECH launched a companion robot marketed exclusively to adults—humanoid robots venturing into the adult entertainment sphere shattered public imagination around embodied intelligence products. In August, after Yujie Robotics listed on A-share, its market cap nearly halved after five trading days, further puncturing the bubble surrounding embodied intelligence company valuations.

These two incidents appeared independent and unrelated, but when internal pressure within the bubble becomes too great, their sequential occurrence may have been inevitable.

Expectations declined sharply, and the primary market for embodied intelligence financing cooled down. Recently, multiple investors confirmed to us that since the watershed moment, an increasing number of investors are eager to sell existing shares.

This is understandable: with tighter IPO access, the exit timeline via public listing has lengthened and risks increased, prompting many to seek earlier monetization.

Zhang Xin (pseudonym), one such investor, invested in a world model startup earlier this year at a post-money valuation of 3 billion RMB. By October, the company’s external valuation had surged to 22 billion RMB.

Zhang Xin recently attempted to sell his stake at a 12 billion RMB valuation—but so far, he hasn’t found a buyer.

01

A team of 40 people, more than half part-time, presenting only a demo, yet seeking funding at a 10 billion RMB valuation.

Does that sound absurd? So did Sun Bo, founder of Lingjing Intelligence, who witnessed this scene firsthand at this year’s WRC.

Though Lingjing Intelligence itself is an embodied intelligence startup focused on cognitive core development, this year, Matrix Partners became its angel, angel+, and pre-A round investor.

But Sun Bo shared this story not just to illustrate a case—he clearly signaled that at some stage, the industry exhibits signs of bubble formation.

He also forwarded a news article about Shao Tianlan to us. The article detailed several social media posts by Shao Tianlan, founder of Meicamand—the so-called “first stock in embodied vision, brain, and hand”—that exposed the mystique surrounding the commercialization of embodied intelligence.

Shortly before, Shao Tianlan posted multiple times publicly accusing embodied intelligence firms of fabricating revenue through so-called data collection centers.

Shao deduced the operational logic: local governments or state-owned platforms fund the construction of data collection centers; robotics companies sell equipment to these centers and recognize revenue; subsequently, the centers resell data back to enterprises, enabling capital recycling.

The value of data collection centers is self-evident. Unlike large language models, whose training data primarily consists of text—readily available from vast online repositories—embodied intelligence requires far more diverse data types, with longer and more complex data pipelines.

High-quality real-world data is extremely scarce, becoming a critical bottleneck obstructing industry progress.

As Zhao Fuheng, Investment Director at Jiuhe Capital, explained, even if all Egocentric data were aggregated today, it might still be insufficient for large-scale pre-training in embodied intelligence. Thus, data collection centers represent a key lever to overcome the industry’s data bottlenecks.

The model itself isn’t flawed—problems arise in whether embodied intelligence firms should classify such revenue as commercial orders, especially when used as reference data for investors. Does it reflect true operational performance? Does it demonstrate a validated commercial loop?

After all, “concerted arrangement” data collection center deals—where funds move between affiliated parties—carry clear risks of revenue inflation.

In embodied intelligence, consensus remains elusive, including on what constitutes a legitimate commercial order. Beyond data collection centers, some firms also count intent orders as commercial revenue.

Intent orders and actual revenue are separated by multiple layers.

A telling data point: in January 2026, domestic humanoid robot intent orders exceeded 120,000 units, while global humanoid robot shipments in 2025 totaled only 18,000 units. Moreover, a significant portion of those orders were for technology validation and interactive demonstrations.

If we strip away the inflation from commercial order claims in embodied intelligence, many companies’ commercialization narratives collapse into a hollow proposition.

In Sun Bo’s view, true commercialization in embodied intelligence hinges on solving three core challenges simultaneously: reliability, real-time performance, and cost control.

Yet currently, OEM manufacturers have not delivered strong solutions. Consequently, revenue generation remains their most pressing challenge.

No revenue → no IPO → no exit for investors → reduced willingness to invest → cooling of funding frenzy. This feedback loop gradually extinguishes the heat in the embodied intelligence funding wave.

Currently, the greater challenge lies with companies already mid-way through their journey—they carried high valuations from the last round, face difficulty securing the next round, yet now operate at substantial scale. R&D costs must be sustained, salaries paid, rent covered, and buyback clauses from financing agreements inch closer to triggering.

They’re navigating a path full of thorns.

02

The fundamental reason embodied intelligence cannot achieve commercialization is that technology has not yet converged.

Yan Qianhang, Vice President at PeakBridge Capital, told Most Talk that the technology is still in early R&D phase—embodied intelligence won’t grow linearly year-on-year as expected, and practical deployment remains premature.

He drew a parallel with autonomous driving: in 2012, no one knew where it would land—mining operations? Logistics? No clear answer existed. But as technologies matured and converged, deployment strategies evolved dramatically.

Looking back at embodied intelligence, numerous technical paths remain contested.

For instance, on foundational paradigms, multiple independent lines exist. VLA was once dominant, but now world models are gaining traction—seen as a way to fill the causal reasoning gaps inherent in VLA.

Similarly, debate persists over whether human-like form is necessary—or whether bipedal or wheeled locomotion is optimal—leading to a proliferation of competing designs.

Early-stage deployments often rely on half-baked solutions easily outpaced by superior future technologies. Thus, Yan believes, one should wait until core technical maturity is achieved before evaluating how to deploy.

In contrast, upstream component suppliers enjoy relatively clearer commercial pathways. Regardless of which technical path ultimately dominates—or which OEM emerges victorious—core components like dexterous hands, sensors, and joint modules will remain indispensable.

For example, Lingjing Intelligence operates in the “digging the well” business model. Its self-developed cognitive foundation has already integrated with over 300 embodied intelligence firms, achieving industry coverage exceeding 70%.

Multiple investors confirmed to us: the current embodied intelligence sector isn’t uninvestable—it’s simply more attractive to invest in high-technical-barrier, proprietary-knowledge (now-how) components.

Based on primary market trends in the first half of this year, over 70% of capital flowed into midstream segments of the embodied intelligence supply chain. Though upstream received only 20% of funding, 30 component firms leveraged this capital to raise nearly 9.6 billion RMB—averaging 320 million per company, significantly higher than the midstream average.

Meanwhile, previously hyped OEMs—backed by China’s robust supply chain—undoubtedly possess promising product roadmaps. Yet precisely because of standardized, socially enabled supply chains, differentiation among OEMs will shrink.

Nothing new under the sun. At the turn of the last century, PC OEMs faced the “wintel dilemma.” With Windows and Intel deeply intertwined, they jointly defined a unified open standard. Downstream PC assemblers could only perform assembly and design tweaks within this rigid framework.

During that era, the wintel alliance captured most of the industry’s profit margin. PC OEMs lacked bottom-up influence and earned only low-margin manufacturing profits.

Embodied intelligence OEMs may soon face a similar crisis.

For instance, Jia Yueting raised 70 million USD in two months for his FF robot venture. But upon examining product details, critics questioned whether the entire robot was manufactured in China, shipped to the US, re-certified, and rebranded—FF has not directly addressed these allegations.

In FF’s official communications, 242 units were shipped between March and June, with annual shipment targets now raised to 2,000 units.

How did Jia obtain these robots? Through OEM production. China’s robotics contract manufacturing ecosystem is now highly mature. In Huaqiangbei, nearly every component and factory can be sourced—many offer ODM services, supporting small-batch production, rapid prototyping, and iterative development.

Like traditional PCs, robots have entered the “one-meter container” era.

03

While demystifying the embodied intelligence sector is essential, it hasn’t diminished consensus on its long-term potential.

But in the absence of technical convergence, we must acknowledge the existence of bubbles—and distinguish between types.

Generally, there are two kinds: value-based bubbles—where expectations of future capabilities diverge significantly from actual outcomes—and scale-based bubbles—where the expectation of 100 top-tier companies materializing, but only 10 actually emerge.

If one fails to grasp the difference, reconsider the “Hundred Group War” of 2014: the assessment of local lifestyle services was correct, but having 100 companies competing in Groupon and food delivery was clearly excessive.

Evidently, today’s embodied intelligence bubble is scale-based. Everyone knows many companies will disappear—but as Zhao Fuheng noted, investors remain willing to keep investing, hoping the sector might eventually produce something akin to large language models.

This statement subtly reflects the prevailing sentiment in primary markets: better to invest incorrectly than miss out. This mindset, to some extent, continues fueling the embodied intelligence sector.

Secondary markets, of course, fare better—since assets are always accessible, secondary markets, though influenced by sentiment, tend to revert to mean over time.

This explains why a primary-secondary inversion is now emerging in the embodied intelligence space.

Multiple investors pointed to the case of dexterous hands. Currently, the market darling is Lingxin Dextrous Hand—a startup specializing in dexterous hands. Market sources indicate the company’s previous round valuation was around 20 billion RMB, with the next round targeting over 40 billion RMB. Another report suggests Lingxin is preparing for a Hong Kong IPO, with optimistic investors projecting a post-listing market cap exceeding 100 billion RMB.

Interestingly, the market isn’t short of dexterous hand makers. For example, Zhaowei Electric (003021), widely seen as the most representative manufacturer, recently launched a 20-degree-of-freedom dexterous hand—supplier to Tesla. Yet its market cap stands at only ~17 billion RMB.

Of course, a company’s worth is moot if it can’t list. One investor judged that post-regulatory changes, Lingxin’s IPO plans are likely to be significantly delayed.

Naturally, Lingxin isn’t the only firm affected—regulatory tightening has stretched investment recovery cycles across the sector. Throughout this process, multiple risks emerge: first, valuation markdowns; second, competitive failure; worst of all, startup collapse.

This is precisely why investors like Zhang Xin want to sell existing shares. The benefit? These sellers entered the sector early—most investments made last year or early this year—during a year-long capital frenzy. Even if valuations drop, the returns remain sufficient to generate profits.

Harder hit are those who joined around mid-year—within a short timeframe, they’ve become the final recipients in a musical chairs game. To some extent, these investors will become steadfast followers, waiting patiently for the industry’s true inflection point.

As for the embodied intelligence startups themselves, none suffer more. Objectively, the industry still needs extensive technical and commercial validation—so these companies will require prolonged capital support.

But facing a cooling market, fundraising has become vastly harder. Primary market share liquidity adds another layer of difficulty. Take Zhang Xin’s investment: when a 12 billion RMB stake is already on sale, who would agree to inject capital at a 22 billion RMB valuation?

One investor told us: to restrict old share transfers, many embodied intelligence firms have stopped pursuing equity restructuring.

From a business perspective, this is rational. As Yan Qianhang said in the interview: before the inflection point arrives, everyone should recalibrate expectations and maintain relative rationality.

Source: Most Talk FunTalk