Gasgoo Munich- The frenzy in the primary market for embodied intelligence is fueling an "atypical" boom.
On one hand, financing data for July remained elevated, with valuations for some early-stage projects—buoyed by industrial capital and state investment—surging to levels typically seen in Series B or C rounds for traditional hard tech. On the other, global shipment forecasts of just tens of thousands of units stand in stark contrast to hundreds of product models already crowding the domestic market, creating a sharp disconnect.
Amid the influx of hot money and rising fears of a bubble, a core question looms: With valuation anchors drifting away from industrial fundamentals, how long can this capital feast—served well before its time—really last?
The hot money remains, but the playbook has changed
According to incomplete statistics from Gasgoo Embodied Intelligence, 36 funding rounds were disclosed in China's embodied robotics and core components sector in July, extending the momentum seen in the first half of the year.
Six of these rounds exceeded 1 billion yuan, secured by ROBOTERA, Morphi Robot, LimX Dynamics, Yimu Tech, Futuring Robot, and HiDream.ai.
In terms of funding stages, LimX Dynamics has reached a pre-IPO round, while Yimu Tech is at Series E and HiDream.ai at Series C. The remaining three are earlier than Series B. Notably, Morphi Robot's angel rounds alone surpassed 1 billion yuan—a scale approaching that of Series B rounds in traditional hard tech sectors.
Beyond the upward shift in early-stage valuations, rapid-fire, multi-round financing within short cycles has become standard industry practice.
This aggressive pace stems partly from the sector's nature—rapid tech iteration and heavy R&D spending—and partly from the strategy of top-tier firms, which are doubling down across rounds to quickly inflate project valuations.
Image source: Delta Intelligence
Take Delta Intelligence: in just six months since its inception early this year, it has completed six funding rounds—averaging almost one per month. Its investor roster includes robot makers like Zhiyuan Robotics, Leju Robotics, and GALAXEA; top financial institutions like Hillhouse Venture, Oriza Holdings, and Lenovo Capital; and industrial capital from the automotive and semiconductor sectors.
Additionally, ROBOTERA, HiDream.ai, Tashan Technology, Westlake Robotics, Futuring Robot, Nexforce, and EverWise have all closed at least three rounds in recent months. ROBOTERA has raised over 3 billion yuan in the past few months alone, while HiDream.ai's total funding has surpassed 2.1 billion yuan.
Upon closer inspection, the trend of "mega-rounds in early stages" within embodied intelligence is not mere capital hype—it is an inevitable outcome driven by multiple factors.
Structurally, embodied intelligence is a capital-intensive "algorithm + hardware + data + scenario" pipeline. Even in early stages, companies must validate the "algorithm-hardware-data" loop—a process requiring deep pockets that has raised the baseline for early-stage funding.
Compounding this is the absence of a unified technological endgame; the industry is exploring multiple paths in parallel, which further inflates early-stage capital requirements.
In terms of team quality, top startups boast "heavyweight" profiles: former executives from tech giants, elite academic backgrounds, and deep industrial resources. This naturally commands a premium for certainty. Capital markets have far higher expectations for these teams than for average startups and are willing to assign unicorn-level valuations and pour in massive funds at an early stage.
Regarding capital structure, Gasgoo data shows that in July, only LimX Dynamics was led by a U.S. dollar fund. The sector's dominant capital has shifted from traditional dollar VC to a mix of state investment and industrial capital. These two funding sources operate with different investment logics and pricing frameworks, directly elevating the overall magnitude of early-stage financing.
Moreover, as competition intensifies, massive funding is no longer just about bolstering cash flow—it is a strategic chip for securing talent, locking in supply chains, and capturing deployment scenarios. It has become a core tool for building competitive moats.
It is worth noting, however, that this model—borrowing against future valuations—depends heavily on consistent delivery in industrial deployment. If the pace of tech iteration and commercialization lags behind the rhythm of valuation inflation, the capital that once fueled the sector's rapid ascent could quickly turn into the pressure that bursts the bubble.
Behind the buzz, beware the valuation bubble
Undoubtedly, sustained financing fever has brought clear value to the embodied intelligence sector: ample capital has accelerated R&D iteration, attracted top cross-disciplinary talent, and fast-tracked the maturation of the entire supply chain.
But when billion-yuan rounds become a monthly norm, and angel-stage companies barely a year or two old join the unicorn ranks, the risks accumulating behind the hype are impossible to ignore.
"Many companies currently lack mature products or clear commercialization paths, yet their valuations hit the 10-billion-yuan mark. This is fundamentally the result of scarce high-quality assets and abundant market liquidity," an industry insider who requested anonymity told Gasgoo Embodied Intelligence.
Beneath the veneer of capital prosperity, multiple industry concerns are quietly fermenting—chief among them the glaring disconnect between valuations and industrial fundamentals.
Image source: Unitree
The embodied intelligence sector is still in its infancy. Even for leaders like Unitree and Zhiyuan Robotics, mass production volumes have only just broken the 10,000-unit mark. Looking at the full year, market forecasts predict global shipments of humanoid robots in 2026 will reach only the tens of thousands.
Deutsche Bank, for instance, recently raised its 2026 global shipment forecast to nearly 50,000 units, with China accounting for about 40,000. Morgan Stanley, meanwhile, upgraded its China forecast from 28,000 to 50,000 units.
In stark contrast, the Ministry of Industry and Information Technology reports that China already had 400 humanoid robot models in the first half of this year—more than half the global total. The proliferation of models essentially reflects an industry where technological routes have not yet converged; it remains in a phase of parallel trial and error across multiple paths.
This also implies that the vast majority of players and products have yet to achieve true scaled shipments. Even though many companies claim deployment across multiple scenarios, actual volumes often remain at the pilot level—far from the scale required for commercial replication.
Yet on the funding side, numerous companies with short track records and no mature products have raised billions in yuan, with some valuations breaching the 10-billion-yuan threshold.
By conventional hard tech valuation logic, funding in the hundreds of millions typically corresponds to the Series A to B transition—implying scaled mass production, a clear profit model, and an end to pure cash-burning. Rounds of 1 billion yuan or more mark the watershed between Series B and C, signaling that a company has secured a solid market share.
By this benchmark, the funding and valuations of numerous early-stage embodied intelligence projects are approaching traditional Series B levels—a clear sign that valuation anchors have detached from industrial fundamentals.
In the view of the aforementioned insider, this valuation bubble not only distorts companies' self-perception but also overdraws the industry's long-term growth potential.
"The downside of this hype is that it drives companies to burn cash on market expansion at high costs—an unhealthy development model. Many teams feel that spending 1 billion yuan is fine as long as they can raise tens of billions in the next round. This completely disconnects from business reality and fosters a dangerous misconception—that reaching the current stage somehow justifies that valuation."
Even more alarming is that many companies mistake the zero-to-one technical validation—or even just a proof-of-concept demo—for having completed 80 or 90 percent of the work.
"But solving the long-tail problems of real-world deployment may require ten times the time and manpower, plus a collective effort from the entire industry and upstream supply chain," the insider added.
If the industry maintains this breakneck pace, few will be willing to settle down and do the unglamorous grunt work of engineering adaptation and scenario refinement—which will ultimately inflict severe damage on the sector. "Many industries decline not due to a lack of market demand, but because of the shortsightedness of practitioners. Teams think they are inherently worth that much, but in reality, it's largely a dividend of the times."
Moreover, both state capital and industrial capital have their own performance metrics: state funds have KPIs for industrial deployment, while corporate investors have timelines for supply chain integration. If scaled results fail to materialize over the long term, the capital retreat could happen far faster than the initial influx.
Embodied AI will not have a "ChatGPT moment"
Across the industry, despite the continuous influx of capital, truly maturing the embodied intelligence sector is no easy feat. Behind it lies a systemic barrier formed by overlapping bottlenecks in algorithms, hardware, engineering, and commercialization.
The most fundamental constraint is the insufficient generalization capability of algorithms.
Most current embodied intelligence models are confined to standardized scenarios within their training distributions, possessing only a shallow understanding of the physical world's spatial rules and causal logic. Once deployed in complex real-world factories or homes, variables like changing lighting, shifted object positions, or slight terrain variations can cause model failure.
Image source: Lingxin Qiaoshou
On the hardware front, the industry faces a "performance-cost-reliability" trilemma. Core components like high-load joints, high-degree-of-freedom dexterous hands, and distributed tactile sensors either meet performance specs at prohibitive costs, or offer controllable costs but lack the lifespan and stability for commercial use. Add to this the widespread issues of short battery life and high failure rates in humanoid units, and the hardware barrier to mass deployment becomes significantly higher.
The "last mile" of engineering deployment is equally thorny. Production line layouts, workpiece specifications, and process flows vary widely across factories, just as home layouts, furniture, and item placement differ from household to household. This means every deployment project requires bespoke adaptation.
More importantly, embodied robots will not operate in isolation. In factories, they must interface with WMS and MES systems, collaborate with automated equipment like conveyor belts and machine tools, and work alongside human operators. Adapting, debugging, and optimizing this entire business chain places extreme demands on a company's industry know-how.
Beyond this, infrastructure gaps—such as the scarcity of high-quality real-world data, fragmented industry standards and ecosystems, and immature safety compliance systems for human-machine coexistence—are invisibly lengthening the deployment cycle for the entire industry.
Given these challenges, the aforementioned insider offered a blunt assessment: pure large-model companies and pure hardware manufacturers will both face an extremely difficult path forward.
"As competition deepens, all players must shore up their weaknesses: pure model companies must ramp up hardware investment; pure hardware players must supplement algorithm capabilities through partnerships." In other words, the endgame competition in embodied intelligence will inevitably hinge on deep software-hardware integration.
After all, the core of embodied intelligence is physical interaction. Without a real hardware carrier, algorithms cannot acquire real-world scenario data, making it difficult to form a positive "data-model-iteration" loop. Conversely, the core value is "intelligence," not the "body"—algorithmic capability is the long-term moat. Simply put, neither a pure software nor a pure hardware path can easily sustain a complete business model on its own.
"At this stage, China's manufacturing technology and supply chain capabilities are very mature, which can significantly lower hardware costs. Therefore, the barrier for software companies to enter hardware is relatively lower. In contrast, for pure hardware companies to pivot to large models at this point means they have already missed the window," the insider noted.
Ultimately, following the explosive emergence of ChatGPT over the past two years, many are wondering if embodied intelligence is poised for its own "ChatGPT moment."
Addressing this, GALAXEA partner Luo Tianqi stated at a recent WAIC forum that over 99% of embodied intelligence's future value will come from the productivity sector, with entertainment accounting for less than 1%. However, the penetration of the productivity sector will not create a sharp inflection point in public awareness like ChatGPT did.
"The penetration of embodied intelligence will progress gradually, like a child growing from age three to five, seven, ten, and eventually to adulthood. Each stage has corresponding tasks it can handle. Overall, it is a subtle process, starting first with warehouses and various production scenarios," Luo said. "Looking back in three or five years, we will find it is everywhere, and we won't even be able to pinpoint the exact year of the inflection point."
Looking further ahead, regarding the highly anticipated home scenario, Li Junlan, research manager at IDC China, also believes deployment will unfold in three stages.
"The first stage is the pilot phase, targeting high-net-worth families with high-priced products focused on technological novelty. The second stage is essential deployment, covering core family needs like elderly care and basic housework. The third stage is true entry into ordinary households, which we estimate will take five to ten years."
Conclusion
Looking back from the middle of 2026, there is no need to mythologize the role of funding, nor to uniformly sing the dirges of the bubble.
The laws of development for hard tech have always been thus: hot money aggregates first, followed by industrial deployment. The influx of capital is a necessary stage of industry growth and a catalyst for technological iteration.
The real divergence lies in who converted capital into technology and core moats, and who spent it on marketing and conceptual narratives.
The next six months will mark the first watershed for the embodied intelligence industry. Financing fever will likely persist, but capital patience will increasingly tilt toward tangible results. Companies that can produce real orders and validate profitable models in single scenarios will continue to ascend; those capable only of demos and storytelling will see their valuations correct faster than imagined.
This means the industry should focus on three core anchors in the coming period:
First, whether real deliveries from top OEMs can break the 10,000-unit threshold and if repurchase rates in single scenarios meet targets;
Second, the domestic substitution rate and cost reduction magnitude for core components like joints and dexterous hands;
Third, whether the first batch of projects that burned through their initial funding without validating a business model begin to see valuation corrections.
The boom phase tests fundraising ability; the ebb phase tests execution capability. This long industrial marathon of embodied intelligence has barely cleared the starting line.
