ABMedia has republished a lengthy analysis from 0xjacobzhao arguing that the biggest mismatch in embodied AI is not technological progress itself, but the fact that capital markets and industrial development are running on different clocks. The article breaks the robotics industry into six separate clocks and says the sector is still in a capital formation phase, while investors are pricing companies as if the market had already reached the stage where deployment economics are proven. Between those two points sit mass-production validation and supply-chain profitability, neither of which, the author says, has fully arrived.
The piece argues that over the next two to three years, the first signal that really matters is unlikely to come from upstream components. It is more likely to appear in a single vertical use case where unit economics work and can be replicated.
Six clocks, six different stages
The report separates the current state of the industry into six tracks: technology is in an early-to-mid stage, commercialization is still early, primary-market fundraising is accelerating, primary-market valuations are already late-cycle, IPOs are entering a concentrated realization period, and earnings transmission in the secondary market has not started yet. Mixing those clocks together, the author says, leads to serious timing errors.
It also argues that many of the stock-picking metrics commonly used for robotics companies — robot revenue mix, customer concentration, ASP multiplied by volume, gross margin, capex, and inventory — are metrics for a more mature phase. Before 2027 to 2028, those readings often do not have enough real operating data behind them. For the current Phase 3 stage, the article says the more relevant indicators are IPO supply cadence, lock-up expiration calendars, channel quality, and unit economics.
Fundraising is accelerating, but valuations have already peaked
The article says the market often describes the primary market as overheated and late-stage, but that framing is only half right. It splits the primary market into two lines — the fundraising cycle and the valuation cycle — and says they are now moving in opposite directions.
According to the report, domestic embodied AI financing reached RMB 93.5 billion in the first half of 2026, about five times the level of the same period a year earlier. In the author’s view, that is more dangerous than a standard late-stage primary market. In a normal late-stage setup, money is leaving. Here, fresh capital is still entering at top-of-cycle prices.
Capital allocation is also highly uneven. In the first half of 2026, 169 full-machine companies took in about RMB 33.2 billion, accounting for 72.2% of total industry financing. Downstream applications drew only about RMB 1.94 billion across 12 companies, or 4.2% of the total, even though the same dataset described that segment as having the smallest valuation bubble and the highest commercialization certainty.
Unitree’s listing became a test of scarcity pricing
The report treats Unitree Technology’s listing as a key market event. It says Unitree debuted on Shanghai’s STAR Market on August 19, 2026. The stock opened up 629% on the first day, taking market capitalization to RMB 444.9 billion. On the same day, however, the STAR 50 Index fell 6.89% and the humanoid robot segment broadly sold off. By the next day, Unitree’s market value had already dropped below RMB 280 billion.
A common market view is that scarcity premiums will disappear only after several more humanoid robot companies go public over the next three years. The article argues that the data from those first two trading days are consistent with a much earlier timetable: liquidity drain may begin on day one, and valuation reset may start on day two. The author also adds a caution. A single trading day is not enough to separate a Unitree-specific drain effect from a broader market decline, so this remains a hypothesis to be tested through similar event studies on future listings such as Yunshenchu and Leju, using excess returns of a robotics basket against the STAR 50.
The report also points to Unitree’s float structure. At the start of trading, unrestricted shares accounted for only 7.44% of total share capital. A price formed with a 7.44% free float and 85% turnover, it says, should not be treated as a stable valuation anchor. It is a product of liquidity structure.
A roughly 61% drawdown from the peak followed within two months
The article lays out the stock’s path in detail. On August 19, Unitree hit an intraday high of RMB 1,100, implying a market value of about RMB 444.9 billion, and closed at RMB 845. On September 2, it traded as low as RMB 546 intraday, roughly half the peak. On September 10, it fell below RMB 500. By October 13, it was around RMB 430, with market capitalization near RMB 174 billion, about 61% below the high. On October 19, after a policy boost, the stock rose about 17% in a single day and returned to roughly RMB 500 intraday, or about RMB 202 billion in market value, still less than half of the peak.
The report says both positioning and fundamentals were behind the decline. On the positioning side, the initial float was only about 7.4%, making the stock easy to push higher on sentiment and equally vulnerable when sentiment faded. On the fundamentals side, adjusted net profit in the first half of 2026 fell about 19.3% year over year, blended gross margin slipped from about 60% to about 56%, and Unitree was overtaken by Zhiyuan in first-half shipment share, about 44% versus 31%. At the then-current price, the stock was still trading at roughly 300 times earnings. The next lock-up expirations are listed as February 2027 and August 2027.
IPO review slowed and primary funding cooled at the same time
Citing September reporting from Reuters and other media outlets, the article says regulators used informal window guidance to slow the listing review process for humanoid robot companies. The focus, according to the report, was whether valuations were too high and whether revenue tied to local government projects represented genuine commercial demand. In Hong Kong, scrutiny of revenue authenticity and related-party transactions also became noticeably tighter.
The article says there has been no public written suspension order. Market descriptions range from a slower pace to a de facto temporary freeze. Even so, the effects are already visible:
- Zhiyuan started its Hong Kong listing process in July, but a new pre-IPO financing round triggered Hong Kong Exchange rules, meaning the earliest possible listing is now early 2027.
- Huandong Technology withdrew its STAR Market application on September 21, becoming the first robot IPO termination of 2026. Its parent company said the reason was that the market environment had changed significantly.
- Hikrobot’s review was suspended on September 30 because its financial documents had expired.
- Leju’s prospectus showed that data-collection-related revenue accounted for 44.94% of 2025 revenue, and its largest customer was a shareholder-related party, exactly the kind of revenue structure regulators are watching.
Primary-market funding cooled as well. The report says humanoid robot financing in September was about RMB 10.5 billion, down about 73% from RMB 38.7 billion in August. Capital shifted toward upstream segments such as chips, vision, joints, and dexterous hands, as well as leading companies that had already validated products. Startups with only demo prototypes found fundraising materially harder.
The author does not describe this as a full retreat. The article notes that Zhiyuan still launched a new financing round in October at an implied valuation of about RMB 50 billion. The better description, it says, is that capital is repricing body makers based on revenue quality rather than abandoning the sector.
Why the report says an “AI-style earnings transmission” signal will not arrive
The article pushes back on a common framing that treats AI as capex and robotics as opex substitution. For factories, robots are usually capex as well, or they are deployed through leasing and Robotics-as-a-Service. The real difference, the author says, is not accounting treatment but decision structure.
AI compute buyers are a small number of hyperscale cloud companies, and their spending can carry a defensive, fear-of-missing-out element. Robot buyers are thousands of factory managers and production-line operators. They always have a ready substitute sitting in front of them: human labor. As long as people can still do the job, the burden of proof stays with the robot.
That is why the report says investors should not wait for EPS transmission. The real anchor to watch is whether a vertical scenario can prove repeatable unit economics, with payback shorter than 18 to 24 months and evidence of repeat purchases.
Planned supply is already approaching the scale of demand projected a decade out
To address whether the industry is overheating and duplicating capacity, the article cites a survey by an international investment bank covering nine A-share robotics supply-chain companies. None of them confirmed large, firm orders. None could provide a clear mass-production timetable. Yet their combined capacity plans corresponded to roughly 100,000 to 1 million robot-equivalent units per year, while the same institution’s own forecast for global humanoid robot shipments in 2035 was 1.38 million units.
The report also says China now has more than 150 humanoid robot-related companies, with more than half of them startups or cross-sector entrants. Nearly 50 companies have launched mass-production models in a concentrated wave, triggering a broad price war. Profit on entry-level models has been compressed to below 5%, and some companies are taking orders below cost. The National Development and Reform Commission has publicly called for entry and exit mechanisms.
Even under the most conservative reading, the author says, supply planning is still far ahead of current demand. Which historical analogy fits best depends on one variable: where subsidies are directed. Solar and new energy vehicle subsidies were aimed at the demand side, creating capacity while also pulling real demand. Robotics subsidies, by contrast, are mainly aimed at the supply side through local industrial funds, production-line subsidies, and innovation centers. They create capacity, but not demand.
A nine-layer map of the robotics stack
The article divides the robotics value chain into nine layers, from L0 to L8, and evaluates each one separately.
- L0 edge compute: the only layer structurally similar to the AI compute chain, because demand does not depend on a “million humanoids” assumption. All robot forms need low-power multimodal inference at the edge. The risk is being absorbed by full-stack in-house development at body makers; the article notes that Tesla has already developed a dedicated robot chip.
- L1 embodied foundation models: the highest technical value and also the highest valuations. The valuation case assumes winner-take-all dynamics, but robotics still lacks the unified benchmarks and interfaces that helped create that structure in large language models.
- L2 data, simulation, and evaluation: the article says the real distinction is not real versus simulated data, but service versus asset. The question is whether the data can be sold a second time. Customized data-collection services are hard to reuse; simulation asset libraries and evaluation standards can be reused across customers.
- L3 deployment software: one of the author’s long-term favorites, but also a layer that can die by arriving too early. Platform value depends on reuse across robot bodies, while body makers are still building full stacks themselves. The report gives three entry signals and says investors should wait for at least two: a de facto hardware interface standard, second-tier body makers willing to adopt third-party stacks, and a single customer with more than 1,000 robots in operation. Teleoperation is described as the only investable subsegment right now.
- L4 body OEMs: the most crowded layer in the entire chain. Secondary-market pricing is driven by float structure and IPO supply, not by fundamentals.
- L5 components: this layer has to be split into five sub-buckets. The risk in actuators is not that orders never come, but that orders arrive without gross margin. Dexterous hands are overheated. Force control and tactile sensing are defined by bottlenecks. Precision parts with demand independent of humanoids have real cash flow plus a free humanoid option. Batteries and thermal management could move from overlooked to critical if industrial scenarios require 24-hour operation.
- L6 motion control and industrial robots: real cash flow plus a humanoid option. In the two- to three-year gap between Phase 3 and Phase 5, the report says this is one of the few ways to gain robotics exposure backed by actual earnings.
- L7 vertical deployment and RaaS: the best non-consensus direction, in the author’s view. The first real signal for the whole industry is likely to show up here. To tell the difference between a company that has been unfairly sold off and one with a low ceiling, the article suggests checking whether a second company in the same scenario can also raise money.
- L8 safety, certification, standards, and insurance: the real gate to commercialization. Possible triggers include mandatory regulatory certification, implementation of ISO and national standards, insurance underwriting requirements, and, least predictably, a major safety accident.
The article also revises a common entry rule for component makers. A single threshold such as “robot revenue exceeds 10% of company revenue” can generate false signals. If actuator revenue enters during a price war at margins below the company average, then a 10% robot revenue mix is dilutive, not re-rating. The report says the correct threshold is a double condition: robot revenue must be at least 10% of total revenue, and segment gross margin must be at least as high as the company’s blended gross margin.
Where durable value actually settles
After compressing the nine layers into four broader groups, the article reaches a direct conclusion: value becomes durable only when capability passes through the deployment layer and enters customer economics.
The strongest narratives and the most crowded capital today are concentrated in embodied foundation models, world models, and general-purpose humanoid bodies. The report says those are also the areas where the mismatch between valuation and evidence is most severe. More attractive areas, in the author’s framework, include evaluation and observability, deployment infrastructure, precision parts protected by time-based barriers, and vertical deployments with proven economics. Their shared feature is that value does not disappear depending on which body maker wins, and their duration can be matched more cleanly with investment tools.
In Phase 3 to Phase 4, the article says, deployment intensity and reliability explain cash flow better than model capability. The market, however, is paying the highest premium for model capability.
Re-anchoring the investment framework
The report closes by framing the issue around timing, valuation, and industry reality.
On timing, it says the primary market is trading 2030, the secondary market is waiting for 2027 financial statements, and the variables that will actually drive prices over the next 12 months are lock-up calendars and IPO supply cadence.
On valuation, the article uses Unitree as an example and says that if the company is valued in segments based on the real product and end-market mix disclosed in its prospectus, the option component accounts for about 95% of market value. The question, then, is not simply whether 200 times sales is expensive. It is whether investors are willing to pay about RMB 270 billion for an option with an uncertain maturity and exercise conditions that may take two to 10 years to satisfy.
On industry reality, the author says humanoid robots have already proven they can be useful. Success rates above 99% in narrow tasks are not stage performances. But the scale is still tiny. The article gives one example: for a global shipment leader, revenue from humanoid robots actually used in intelligent manufacturing and inspection was RMB 15.7 million, about 2.6% of humanoid revenue.
The report also says investment opportunities are not the same as business opportunities. Precision parts fit better as secondary-market allocations, certification and compliance services look more like private equity, and transitional capacity plus dual-source validation cash flow behaves more like a bond than an option. If a fund mandate only allows early-stage equity, the reachable opportunity set is much smaller than the opportunity set already identified.
Seven reverse checks the author says investors should use
In its section on investment discipline, the article argues that during the capitalization phase of Physical AI, the rule should be to invest before earnings transmission becomes obvious, but only after deployment evidence has become factual. It then lists seven categories of false indicators and the reverse evidence investors should demand instead:
- Do not pay for shipments without deployment. Ask for human intervention rates instead, with the article citing the YD/T 6770 standard as an example.
- Do not pay for leaderboard rankings. Ask for real utilization, mean time between failures, and SLA execution hours confirmed by customers.
- Do not pay for stacked degrees of freedom. Ask for cross-year repeat purchases, especially independent delivery orders from the same customer in successive batches.
- Do not pay for low-quality government or data-collection orders. Ask for ROI and payback periods released by actual buyers, not sellers.
- Do not pay for offshore structures that only hide risk. Ask whether deployment cost at the Nth site has fallen below 30% of the first site.
- Do not pay for component narratives detached from earnings. Ask for segment-level financial disclosure, including specific robot revenue and segment gross margin.
- Do not pay for body or model valuations supported only by long-dated TAM projections. Ask whether reusable software revenue is growing faster than headcount tied to delivery.
The article ends with a disclaimer saying it used AI tools including Claude Opus 5, ChatGPT-5.6, and Gemini 3.6 Flash during the writing process. It says the content is for information integration and academic or research exchange only, and does not constitute investment advice or a recommendation to buy or sell.
