Pantera Partner’s Insights on Embodied AI in China: Hardware Supply Chain Holds an Advantage, Scaling “Data Farms” Cannot Resolve Model Shortcomings
Author: Cosmo Jiang, Erik Hopyan (Pantera Capital)
Edited by: TechFlow
TechFlow Summary: As liquidity tightens in the primary crypto market, leading crypto VCs have significantly extended their reach into the AI and robotics sectors.
Pantera Capital is one of the earliest and most well-known institutional-level crypto/blockchain investment firms in the U.S. Founded in 2003, it launched America’s first institutional Bitcoin fund in 2013. For years, it focused on digital assets and blockchain, before gradually expanding into broader technological frontiers such as AI and robotics.
Recently, partners of this fund, Cosmo Jiang, and investor Cody Poh, took a special trip to Shenzhen to conduct field research on China's robotics supply chain. They observed every link in the chain, from humanoid robot OEMs and model teams to dexterous hand developers and data acquisition equipment vendors, and released a comprehensive summary report.
Their core findings boil down to a direct conclusion: China leads in hardware, the U.S. leads in software, and in the intermediate data segment, quality matters far more than quantity.
More interestingly, the piece offers insights and comparisons on the U.S.-China capital ecosystems from a VC perspective:
In China's venture capital landscape, projects are generally expected to show revenue within 2-3 years or even push for an IPO. Coupled with relatively weak intellectual property protection, this forces entrepreneurs to prioritize "quick cash" businesses like hardware sales and data outsourcing, making them hesitant to bet heavily on long-cycle underlying software R&D.
Furthermore, low domestic labor costs mean humanoid robots currently remain largely confined to showroom reception or performances (dancing, martial arts displays). In contrast, high U.S. labor costs and a shortage of blue-collar workers drive industrial and logistics automation with tangible commercial orders.
We have edited the full text as follows.
At 2 a.m., after a six-hour delay in Singapore, our Cathay Pacific flight finally landed in Hong Kong. Having slept less than three hours, we hopped into a car straight to Shenzhen. This is China's hardware capital and the headquarters hub for tech giants like Huawei, BYD, and DJI.
"Genchi Genbutsu": Go and see for yourself. Having previously visited numerous domestic U.S. robotics companies and selectively invested in several of them, these experiences taught us how crucial China is to the global robotics supply chain. But if we merely watch from afar, can we truly grasp the reality?
For the past two years, the robotics sector has been our research focus. Breakthroughs in embodied AI have brought step-function changes to the industry's growth potential. Today, models can already perceive vision, plan logically, and act in the physical world, while the hardware required to drive these models becomes increasingly powerful and affordable.
Under the most optimistic projections, robots will disrupt the broader global manual labor market. Currently, there are 2.2 billion manual workers worldwide, generating an annual payroll pool exceeding $20 trillion. Even a mere 1% penetration rate would generate revenue comparable to today's top global tech giants.
China is indispensable to the embodied AI industry. Global equipment manufacturers rely heavily on China's supply chain, with Chinese makers accounting for over 90% of the global humanoid robot production volume in 2025. Meanwhile, recent restriction measures implemented by the U.S. Federal Communications Commission (FCC) on advanced foreign-manufactured robotics equipment highlight how geopolitics is profoundly reshaping the industry.
As fundamental investors, we must understand and navigate this rapidly shifting macro landscape. The best way to achieve this is to go deep to the front lines and witness the most authentic industry conditions firsthand.
Thus, our Shenzhen journey officially began. Here, we conducted a first-hand, in-depth observation of China's rapidly evolving robotics ecosystem. We visited a wide array of enterprises covering the entire robotics value chain, from vertically integrated humanoid robot OEMs and frontier model developers to data acquisition equipment suppliers and specialized dexterous hand researchers.
During this period, we held in-depth conversations with startup founders, executives overseeing sales, manufacturing, and R&D, and had candid exchanges with fresh university graduates entering the workforce. Our inspection goals were crystal clear: determine China's true position and core competitive advantages in the global robotics race, and more importantly, identify current market gaps and opportunities worthy of investment.
Hardware: China Leads the Manufacturing Race
Throughout our trip, one thing was unequivocal: China holds an absolute lead in the robotics hardware race, backed by unparalleled manufacturing scale and supply chain integration capabilities. This advantage is most acutely felt in Shenzhen, which serves as the beating heart of China's manufacturing engine.
Decades of accumulation as a core global manufacturing hub have built an exceptionally deeply entrenched foundational ecosystem here: readily available raw materials, highly cost-effective logistics networks, mature large-scale mass production lines, and a vast army of skilled labor proficient in precision manufacturing. This forms a structural moat that other emerging manufacturing centers find extremely difficult to replicate.
This supply chain advantage translates not only directly into extreme cost competitiveness but also into astonishing innovation iteration speeds. In Shenzhen, a humanoid robot startup can come up with a new idea in the morning, notify upstream suppliers, and have newly custom-machined parts delivered straight to their office by noon.
This extreme responsiveness allows Chinese manufacturers to supply higher-quality frontier hardware at lower prices. The outdated stereotype that China can only produce low-quality knockoffs or generic goods is severely obsolete.
As frontier model labs continue to scale up training, their demand for customized data acquisition equipment is exploding. China holds a unique advantage in the low-cost, scalable manufacturing of such equipment.
Some top U.S. robotics companies we spoke with often loudly claim to independently develop and manufacture all core components. However, when communicating with local Shenzhen suppliers, they openly name names, revealing which Mag 7 stocks or multi-billion-dollar Silicon Valley unicorn startups they consistently supply. This applies to everything from dexterous hands and actuators to wide-angle cameras and headsets used for data acquisition.
Take Wuji Tech's dexterous hand, for example. Integrating over 1,800 components within a compact footprint close to human palm size, its performance surpasses any mechanical hand we've seen in the U.S. Predictably, they serve as behind-the-scenes core suppliers for many U.S. tech giants.
Each finger supports reverse bending, significantly enhancing durability (a fiercely contested battleground in current mechanical hand development) and impact resistance against accidental collisions. On-site, we watched the mechanical hand (via teleoperation) precisely extract a single card from a full deck. Trivial for humans, this poses a massive challenge for robotic manipulation, requiring micron-level precise control over force and friction.
Another shocking yet slightly unsettling moment came upon seeing UBTECH's humanoid robot with synthetic skin, which instantly evoked a sense of déjà vu reminiscent of Cylons from Battlestar Galactica.
Another defining characteristic of the Chinese market is intensely fierce involution-style competition, which constantly drives down product prices and compresses industry profit margins toward zero.
One startup spent a full hour detailing the technical breakthroughs of their next-generation actuators. However, when asked what ultimately helps them break through in market competition, they blurted out: "Obviously, price is everything." Companies holding this view are far from alone.
Over lunch, a chain retail store owner lamented that it's hard to make profits doing business domestically because competitors always sacrifice current-year margins to grab market share, hoping to earn money next year; then the following year brings another disruptor competing viciously at even lower prices.
Chinese culture harbors a deeply rooted spirit of enduring hardship. This resilience to willingly grind through difficulties for the long term undoubtedly suppresses individual companies' ability to capture excess commercial profits, making survival exceptionally tough for each firm. Yet, macroscopically, it grants the nation's entire supply chain industrial cluster an astonishing overall competitiveness.
Beyond cost and scale advantages, China's position at the very upstream of the robotics supply chain grants it another naturally gifted advantage often underestimated by outsiders: the ability to closely observe the evolutionary trends of global frontier technologies.
Major frontier robotics labs constantly experiment with novel data collection methods, ranging from tactile and force-feedback sensing gloves to visual capture headsets equipped with eight HD cameras for wider-angle recording.
As the almost sole global supplier capable of rapidly and cost-effectively delivering such frontier hardware, Chinese manufacturers can catch the subtle shifts in technical demands from top-tier labs earlier than anyone, sometimes predicting the next generation's technological direction 3 to 6 months ahead of the broader market.
However, can Chinese companies truly translate this manufacturing dominance and preemptive market insights into core proprietary technological advantages? That is a completely different proposition. As our journey progressed, doubts grew deeper.
Data: Collection Is the Industry Bottleneck, Not a Competitive Moat
Nearly every robotics company told us that China is implementing massive-scale data collection initiatives, broadly exporting this data to global clients.
Superficially, this seems logical: companies have workers don capture headsets to record first-person proprioceptive data, earning supplementary income. However, as we engaged with more startups within the Chinese ecosystem, we found the underlying mechanisms far more complex, usually driven by explicit local policy guidance.
Multiple industry insiders revealed to us that local governments across various levels in China are actually the core drivers of this wave of data collection. Following central policies supporting robotics development and employment, local authorities keenly realized that financially subsidizing the data collection phase is a shortcut to creating new robot-related jobs in the short term.
Driven by quantifiable assessment metrics, local officials tend to pursue easily measurable short-term results, a pursuit that sometimes comes at the expense of higher-dimensional, long-term industrial strategies.
Millions of dollars continue to be thrown into training data of questionable value. Across China, countless so-called "data farms" have emerged, where workers mechanically repeat identical movements day after day, stacking up tens of thousands of hours of "robotic training data."
However, once the initial batch is collected, datasets of this highly repetitive and homogenous nature offer virtually no substantial incremental value for improving model generalization capabilities.
Despite concerns over data quality, it continues to sell endlessly to overseas clients. This underscores a deeper, harsher industry reality: data scarcity remains the biggest bottleneck in current robotic model R&D. Unlike LLMs that can scrape massive existing text corpora from the internet, the robotics field simply lacks corresponding ready-made data pools.
Furthermore, truly valuable robotic training data takes highly varied forms. The types of data needed to unlock the next leap in model performance do not even exist at a usable scale in the real world today.
Therefore, the true core moat has never been data collection itself, but rather the ability to transform raw data into high-value training signals.
Encouragingly, growing frontier indicators suggest that scaling laws also apply to robotic models, meaning model performance deterministically improves as training data scales. This implies sustained, voracious market demand for high-quality, diverse heterogeneous robotic training data. Such data must accurately match the evolving standards across modalities, collection methods, and novel training architectures.
In an industry environment where the definition of "high-quality data" dynamically shifts, the most defensible competitive moat lies in becoming an irreplaceable, highly trusted scientific research partner to top-tier frontier labs globally.
Winning this industry reputation hinges on possessing exceptional professional discernment in data annotation, cleaning, filtering, and model evaluation, rather than relying solely on brute-force collection capacity.
The ultimate winners are likely not the teams collecting the largest total volume of data, but those with deep embodied AI research backgrounds capable of synchronizing with frontier robotics companies' technological evolution at the fastest pace. Just as Scale AI and Mercor deeply align with top global LLM labs through premier research understanding, this logic holds equally true in the embodied domain.
Software: The Elephant in the Room
Robotics automation technology has actually existed in the industrial sector for decades. For instance, since acquiring Kiva in 2012, Amazon has massively deployed automated robots in its warehousing and logistics systems, surpassing 1 million units in deployment by 2025 while continuously introducing novel form factors and functionalities. Thus, manufacturing increasingly advanced hardware is merely a natural extension of industrial evolution.
The true paradigm shift in robotics lies in embodied intelligence. It thoroughly breaks the limitation of narrow-predicated programming for specific tasks, fundamentally expanding robots' generalized cognition and execution boundaries.
During all our visits in Shenzhen, AI software remained an "elephant in the room" that no one wanted to openly address. Most humanoid robots and devices we saw on-site possessed extremely limited autonomous decision-making and generalization capabilities, heavily relying on manual teleoperation or preset fixed-program actions.
Multiple executives frankly admitted that their robots currently lack advanced AI capabilities, with underlying software algorithms lagging at least 1 to 2 years behind top overseas competitors. Among grassroots young employees, you often hear less official pleasantries and more brutally honest realities. They jokingly confess that their software systems cannot operate properly in complex scenarios right now, and previous attempts fall far short of going all-in.
Part of this stems from our focus on Shenzhen's hardware-heavy ecosystem rather than software-algorithm-centric cities like Hangzhou or Shanghai. Yet, beneath this lies a more universal underlying industry reality.
China's dominance in hardware manufacturing and data collection has not naturally translated into leadership in software algorithms. Many Chinese humanoid robot OEMs do not develop their own base models from the ground up; instead, they continue to rely on open-source models released by U.S. frontier labs as the cornerstone of their software capabilities, particularly the open-source ecosystems from OpenAI, Nvidia, and Physical Intelligence.
Exchanges with industry researchers indicate that many Chinese robotics companies seem to prefer avoiding massive heavy-capital investments in base model R&D, focusing instead on post-training fine-tuning of existing open-source models.
For U.S. frontier model labs, this creates a strategic wedge: open source can evolve into an incredibly powerful market distribution strategy, driving global developers to adopt and build ecological networks around their models. As these base models' general-purpose performance grows stronger, the incentive for Chinese manufacturers to independently develop alternative underlying models may further diminish.
Model distillation has long been an open secret in Chinese corporate circles, with some teams clearly betting on replicating this path in robotics. However, unlike LLMs which can achieve efficient model distillation via synthetic text, robotic models face far more complex engineering challenges.
Distilling robotic models requires not only base algorithmic frameworks but also vast amounts of real-world physical interaction data, co-optimized hardware infrastructure, and rigorous field testing. Consequently, even enterprises with formidable hardware manufacturing capabilities find the difficulty of purely replicating cutting-edge overseas model capabilities via distillation increases geometrically.
Deep Divides in Business Rules
In a country that lifted itself out of poverty within a single generation of contemporary executives, the profound imprint of material memory is clearly visible. Chinese society permeates a strong culture of striving and catching up, highly common in regions where new wealth rapidly emerges: people crave quickly securing deterministic commercial profits, fearing a return to the poverty-stricken years still vividly remembered in the rearview mirror.
While it sounds reasonable, what truly stunned us was how profoundly this mindset shapes the entire entrepreneurial and VC ecosystem.
While visiting Chinese startups, we repeatedly heard an envious sigh: "U.S. startups don't even have to worry about making money short-term." This directly points to the harsh commercialization pressure from investors borne by Chinese teams.
This divergence stems from the generally shorter payback period expectations in China's VC ecosystem. Many startups are expected to generate substantial business revenue within 2 to 3 years of founding; once showing early signs of traction, investors impatiently push them toward an IPO.
We hypothesize that Chinese VC firms' lack of patience for long-term investment cycles stems mainly from two factors: first, the absence of high-return alternative assets (stock markets oscillating narrowly over the past decade, residential rental yields hovering around just 1% in tier-1 cities, compounded by capital controls); second, a shortage of long-horizon institutional LPs akin to overseas university endowments capable of weathering ultra-long cycles.
This forces startup teams to severely pivot toward hardware ODM and data collection outsourcing services, as these businesses yield faster receivables and shorter capital turnover cycles; meanwhile, they shy away from underlying software R&D, whose commercialization pathways are lengthy and uncertain.
Additionally, attitudes toward intellectual property protection differ starkly between the two regions. Although we frequently saw corridors lined with proudly displayed patent certificates during our visits, substantively weak IP protection remains an undisputed fact within the industry.
This compels companies to commercialize every minor innovation at maximum speed, otherwise they risk being swiftly copied by peers at lower costs who then seize market share first.
This ultimately causes structural misalignment in incentives: while China has built extraordinarily strong barriers at the physical robotics layer, the current capital environment and assessment orientations objectively suppress the ultra-long-term patient capital required to elevate these physical advantages into software leadership.
Pantera's Core Thesis on the Robotics Industry
Upon leaving Shenzhen, we gained a far clearer and more multidimensional understanding of the global robotics industry's developmental stage, along with the critical gaps and breakthrough points embedded in its future evolution.
The Robotics Industry Remains in Its Nascent Stages, with China Playing an Irreplaceable Pillar Role
Our field research in Shenzhen and dialogues with Chinese robotics executives once again validated our judgment: the embodied AI industry remains in its infancy. Hardware performance continues optimizing, but at least within the consensus of Chinese teams, software algorithms are still far from reaching fully mature commercial deployment.
China's weight in the global robotics supply chain has become more pivotal than ever. Chinese manufacturing enterprises are not only core producers of data collection equipment but also primary suppliers of key robotic components.
Data remains the core bottleneck for unlocking model generalization capabilities; as long as scaling laws hold, collecting more diverse, high-quality data is the inevitable path to performance enhancement. China's manufacturing ecosystem possesses a unique industrial position in the mass production of data collection hardware and the stable global output of core components like dexterous hands and actuators.
How China chooses to deeply participate in the global robotics ecosystem moving forward will directly dictate the iteration speed of worldwide industry development. China's hardware-level dominance is highly likely to persist long-term, continuing to serve as the core supplier for global robotics firms (even if the FCC imposes potential bans on complete machines, these restrictions do not cover individual components).
If Chinese teams eventually catch up at the algorithmic model level, enterprises are highly likely to wield significant leverage over the global supply chain by raising core component prices or restricting hardware supplies. While this remains a low-probability scenario today, it represents a critical latent risk that must be heavily weighed during investment risk assessments for domestic U.S. robotics companies.
Economic Drivers for Adopting Robots Locally in China Have Not Fully Materialized
Although humanoid robot shipments in the current Chinese market look impressive, almost none are deployed into production scenarios yielding substantial economic output. Presently, Chinese humanoid robots are mainly utilized in low value-added display and entertainment contexts, such as providing basic guidance to visitors in showrooms or expos, or performing dances and martial arts routines. Overwhelmingly, quadruped and even bipedal robots are still sold primarily to academic research labs or geek enthusiasts.
This stands in stark contrast to the U.S. market. In the U.S., robots are accelerating penetration into high-value industrial manufacturing and logistics warehousing scenarios, with some enterprises securing multi-million-dollar long-term procurement contracts.
The key driver behind this fundamental divergence lies in the massive gap in labor costs and supply structures: Western high labor costs combined with younger generations' aversion to blue-collar physical labor create powerful economic drivers for automation-driven cost reduction.
Even so, technological progress often far precedes widespread commercial adoption. One Chinese entrepreneur shared a vivid analogy with us: today's humanoid robots resemble early automobiles in the horse-drawn era. People widely doubted the practical utility of cars back then because carriages perfectly satisfied daily transport needs; however, no one could foresee that automobiles would eventually revolutionize the entire mobility landscape and spawn entirely new, massive industries.
We firmly believe this disruptive evolutionary principle applies equally to humanoid robots. The public and enterprises struggle to understand today why general-purpose humanoid robots are necessary, simply because they haven't witnessed the full potential unleashed by generalized embodied AI. Moreover, lacking the urgent economic imperative to reduce costs by replacing manual labor domestically, the endogenous motivation to pioneer unprecedented commercial application scenarios remains relatively insufficient, which to some extent makes the local entrepreneurial ecosystem appear more conservative.
General-Purpose Humanoid Robots Will Ultimately Become the Lowest-Cost Labor Form of the Future
The industry fiercely debates a core question: given that specialized robots handle specific tasks more efficiently and stably, why does the world need general-purpose humanoid robots?
Our field research in China corroborates this reality: at the current stage, dedicated robots' actual operational capabilities truly outperform general-purpose ones. By focusing on well-defined, quantifiable micro-tasks, specialized robots can achieve faster commercial closed-loop cycles.
However, in terms of horizontal scaling, specialized robots hit inherent ceilings. A company cannot procure a dedicated robot set for every distinct workflow in daily operations; likewise, a typical household would never simultaneously buy a dishwasher robot, table-cleaning robot, laundry robot, and clothes-folding robot.
While specialized robots undoubtedly excel in singular processes, once multiple machines must be combined to complete a sequence of daily chores, the overall economics of procurement and maintenance drop significantly.
We believe general-purpose humanoid robots will ultimately become the most cost-effective labor form. Their core advantage does not lie in executing a single specific task with 99.9% extreme precision, but in competently handling hundreds of entirely different daily tasks with a solid 99% accuracy simultaneously.
Humanoid robots don't need to rank first globally in every single chore like dishwashing, ironing, or laundry; they simply need to perform adequately across these tasks and more. As model generalization improves, a single universal hardware chassis can unlock infinitely expandable skill libraries, eliminating the need for enterprises or households to purchase additional specialized equipment.
Ultimately, the final economic moat of general-purpose humanoid robots lies not in the extreme optimization of single-point operational capability, but in maximizing the number of task categories a single device can cover. This generalized capability, coupled with improved operational stability and continually declining hardware manufacturing costs, will fundamentally disrupt the entire economic model of physical manual labor.
Huge Market Scale for Data Collection, with Core Value Consolidating Toward "Data Intelligence"
With more conclusive evidence of scaling laws successfully extending into the physical AI domain, robotic data collection is bound to grow into an immensely vast market. Figure AI's recently released Helix 2.5 model provided encouraging empirical proof: as training data volumes multiply, enhancements in robotic model capabilities demonstrate high predictability.
Just as larger datasets and computing power drove capability leaps in LLMs, the robotics field is replicating the exact same evolutionary trajectory. This means that as frontier labs accelerate their race for generalized embodied AI, rigid demand for robotic training data will experience sustained explosive growth.
However, merely increasing data volume does not equate to obtaining higher-quality data. Our field research in China strongly reinforced this insight. Although mega data farms can batch-generate tens of thousands of hours of operational footage, this highly homogenized data offers negligible marginal incremental value for boosting model generalization performance.
As the industry enters deep waters, the bottleneck will swiftly shift from "collecting more hours of data" to "precisely filtering out truly valuable core data fragments."
Therefore, the true blue-ocean opportunity does not lie in extensive data collection, but in how to convert raw training durations into quantifiable model capability enhancements. This demands more refined data annotation, filtering, cleansing, and evaluation systems to extract precious effective training signals from massive streams of redundant, low-quality assembly-line data.
Amidst the rapid iteration of robotic models and their constant reconstruction of data standards, the professional capacity to continuously mine and deliver high-value, premium dataset collections will form a competitive moat far deeper than merely throwing bodies at data collection.
The Model Layer Will Gradually Face Commoditization, but the U.S. Will Sustain Algorithmic Leadership Over China
We believe competition at the underlying robotics model layer is intensifying sharply, carrying long-term risks of commoditization (commoditized).
Over the past few months, several top U.S. teams, including Physical Intelligence, Generalist, Dyna Robotics, and Figure AI, have independently announced major breakthroughs in model capabilities and generalization. The fact that multiple teams achieved substantive breakthroughs within a very short timeframe indicates that frontier embodied intelligence is highly unlikely to be monopolized by two giants like OpenAI and Anthropic, as seen in the LLM space.
Most blue-collar physical labor doesn't require superhuman IQ scores of 200; it merely needs moderate environmental comprehension paired with excellent post-training fine-tuning. As training methodologies gradually mature and data scales expand, the industry's competitive focus will shift comprehensively from developing universal base models to scenario-based deployment and targeted post-training. Under this landscape, any company solely betting on base models will find it increasingly difficult to maintain differentiated competitive moats.
In the foreseeable future, we believe the U.S. will sustain an algorithmic lead over China in software. This largely stems from Chinese robotics companies preferring to chase immediate cash flows from hardware sales and data outsourcing, lacking the patience for long-term heavy-CAPEX R&D investment. Continuous breakthroughs by U.S. frontier labs in underlying algorithms will likely deepen the dependency habit of Chinese manufacturers on overseas open-source ecosystems.
Unlike LLMs, which can be relatively easily reproduced via rapid textual distillation, the threshold for embodied AI model distillation is significantly higher. Model cognition must achieve deep closure through real interaction data spanning diverse physical environments and varying mechanical ontologies. Relying solely on single-point advantages in hardware manufacturing cannot easily overcome this comprehensive barrier. We infer that the generational gap between the U.S. and China in embodied software may prove more enduring than the disparities observed in LLM domains.
The Robotics Sector Possesses Significant "Late-Mover Advantages"
Currently, massive capital is being funneled into resolving fundamental disagreements in the robotics field (such as model architecture roadmaps, mechanical configuration designs, vertical integration vs. horizontal specialization, etc.), meaning many early-stage startups are destined to hit dead ends during this exploration phase.
New entrant teams today can absolutely stand on the shoulders of pioneers to absorb lessons learned, bypassing the massive capital drain of trial-and-error tuition fees, and directly build cutting-edge new architectures based on the industry's latest paradigms (meta).
Conclusion
At 8 p.m. on the final day of our itinerary, we boarded the same car, beginning our journey back to New York via Hong Kong International Airport. After conducting a comprehensive, in-depth examination of Shenzhen's core robotics industrial belt, we gained a refreshing and updated understanding of the global robotics industry's development panorama.
"Genchi Genbutsu." This field visit once again solidified our conviction: in this era of information overload, being on the front lines and seeing things with your own eyes remains the most robust cornerstone for fundamental investors building independent thought and deep analysis.
China will undoubtedly play a pivotal role in reshaping the future landscape of global robotics, and we will closely monitor the evolutionary processes across hardware, software, and every link bridging the two.
Decades of manufacturing accumulation have endowed China with enormously vast, even insurmountable, barriers in hardware production. However, flaws driven by short-term quantitative targets are surfacing in data collection, and capitalism's shortsightedness regarding immediate returns means domestic companies face greater constraints compared to American firms when confronting foundational innovations requiring long-cycle breakthroughs.
One point is clearer than ever: upon departing Shenzhen, we carry an even firmer confidence in the future prospects of the robotics track, firmly believing that now is the perfect moment to take a heavyweight position in this industry.
Before underlying software truly possesses sufficient generalization capabilities and endows humanoid robots with substantive productivity, the industry still faces enormous structural gaps in data collection, high-dimensional labeling, and data intelligence. We are constantly seeking outstanding entrepreneurs pioneering breakthrough solutions in robotic data and embodied AI. If you are deeply cultivating this space, our investment team eagerly looks forward to connecting with you.