August Embodied Robot Financing: Total 56 Rounds, 5 Firms Raised 14 Billion Yuan, Nearly Half Not Yet at Series A

Gasgoo Munich-Fervor in the embodied intelligence sector shows no sign of cooling.

According to Gasgoo Embodied Intelligence, the domestic embodied robot and core components sector saw 56 funding rounds disclosed in August.

The largest single round came from XPENG Robotics, exceeding $900 million—a record for a private equity round in China’s embodied intelligence sector. Next was Sharpa, which revealed its funding in August with a cumulative total exceeding 4.5 billion yuan and a post-money valuation of roughly 22 billion yuan.

Additionally, PaXini Tech, Five Ages‌, and INFIFORCE all secured funding in the 1-billion-yuan range.

That means these five firms alone absorbed roughly 14 billion yuan, intensifying the winner-take-all dynamic.

Yet even as the top heavyweights dominate, massive capital is pouring into early-stage ventures: nearly half of the 56 deals remain at the Seed, Angel, or Pre-A stages. Growth-stage projects around Series B are relatively scarce, painting a funding landscape that is "hot at the ends, cool in the middle."

Beneath the surface noise, however, lies a critical shift in where the smart money is going: this wave of financing is pivoting from "building the body" to "building the brain, feeding data, and racing to mass production."

Capital’s Pivot: From "Body" to "Brain"

Gasgoo data shows nearly half of August’s 56 deals flowed to companies focused on the "brain."

These startups fall into two camps: pure-play "brain" developers like Moushen Intelligence, LatentVerse, Light Robotics, and INFIFORCE, many of which secured nine-figure rounds; and integrated players combining hardware with proprietary "brains," such as XPENG, Sharpa, GAC-spawned Huilun Technology, and Five Ages‌, all treating the "brain" as their core moat.

The takeaway is clear: whether for incumbent hardware makers or new entrants, the "brain" has become essential battleground territory.

The logic is straightforward.

Two years ago, the industry’s core question was primitive: can we build a robot that walks, runs, and jumps without falling?

It was an era defined by mechanical prowess and athletic ability, where a simple backflip could go viral.

But by 2026, the focus has shifted. Walking and running are no longer novel; the real competition lies in whether a robot can "see, react, and complete tasks" in unfamiliar, volatile real-world scenarios—essentially, its generalization and operational capabilities.

Currently, many robots excel in structured, choreographed settings, yet they often "fail" once deployed in open environments facing unseen objects or untrained tasks. The core constraint: a lack of generalization in the "brain."

If motion control and hardware set the floor for embodied intelligence, the embodied brain determines its commercial ceiling.

In China, a mature supply chain has made hardware gaps relatively easy to close; the narrowing performance differences in motors, joints, and motion control across various machines prove this point.

The "brain," however, is different. It cannot be bought or rushed with brute-force parameters. It requires massive real-world data and continuous algorithmic iteration through complex interactions—a process that takes time.

For this reason, the embodied "brain" has become one of the hottest investment directions today.

Image Credit: Mifeng Technology

Beyond the "brain," the "data infrastructure" track is buzzing.

In August, over a dozen players, including Mifeng Technology, Jianzhi Robotics, Current Robotics, Kaiwang Data, Lingyu Intelligent, IO Intelligence, and Yuandian Technology, secured fresh capital.

Their common thread: they aren't just doing data collection or labeling in isolation. Instead, they are integrating collection, labeling, training, evaluation, and simulation to build end-to-end data loops—acting as the "picks and shovels" sellers for the industry.

The reason is clear: the "brain's" potential is capped by data quality, and the sector faces a shortfall of millions of hours of data.

According to the China Academy of Information and Communications Technology (CAICT), embodied foundation models need at least 10 million hours of real-world data to hit their "ChatGPT moment." Currently, global high-quality real data sits at just 100,000 to 1 million hours—a gap of an entire order of magnitude.

Unlike the "brain," the data sector offers certainty: no matter which technical route wins, data is indispensable.

Driven by this demand, regions are racing to build embodied intelligence training grounds. CAICT data shows over 70 such sites were operational by the end of June 2026, with another 46 under construction or planned.

Yet beneath the heat, a chill is setting in.

In late August, Beijing’s first humanoid robot data training center was reported to have ceased operations. Located in Shougang Park and launched in March 2025 with 100 robots and a 5 billion yuan market target, it shut down in less than 18 months.

That means from unveiling to closure, it lasted less than a year and a half.

The cause was a strategic pivot by partner Realm Intelligence—from laboratory teleoperation to remote teleoperation in actual work environments. The obsolete model wasn't data collection itself, but the "centralized teleoperation 1.0" approach.

This contrast serves as a warning: the "infrastructure model" of simply stacking hardware and venues is unsustainable. Data value is shifting from "scale-oriented" to "quality and scenario-oriented." The real barrier isn't the size of the training ground, but the ability to penetrate real production across industries to acquire high-value, long-tail, generalizable interaction data.

As collection methods iterate, a shakeout in the data sector seems inevitable.

Dexterous Manipulation: A Key Focus

On the hardware front, capital is targeting components that enable "dexterous manipulation:" joint modules, dexterous hands, and force/tactile sensors—addressing three hurdles: moving steadily, gripping firmly, and sensing accurately.

Take joint modules. In August, funding flowed to EYOUbot , Linkhous, NOUS BOT, Joy-Moiton, Talls Intelligent, and Hengxuan Technology.

The bet on joints rests on three pillars.

First, the execution system holds the most value in a robot, accounting for 35% to 60% of total cost.

That means whoever scales production and lowers costs first will gain the upper hand in pricing.

Second, high technical barriers and clear import substitution potential. High-precision reducers and servo drives were long dominated by foreign giants; as mass production nears, supply chain autonomy has shifted from "optional" to "mandatory." Domestic breakthroughs here mean tapping into a certain market dividend.

Third, strong strategic positioning and customer stickiness. Joints require co-definition with OEMs and long validation cycles, making them a classic "lock-in" component.

In this sense, if the "brain" funds future imagination, joints fund current delivery capabilities. Unlike the fragmented "brain" landscape, the joint module track offers higher certainty and a clearer path to volume.

Image Credit: Sharpa

Then there are dexterous hands. For two years, the industry prioritized legs over hands. Now that mobility is solved, dexterity is the new battlefield.

The logic is direct: only when hands are agile enough to handle eggs, thread needles, or use tools can robots truly replace human labor and close the commercial loop. The value of a robot is ultimately realized through a "useful pair of hands."

August’s standout case was Sharpa and Ruiyan Intelligent Control.

Sharpa is particularly representative. Its disclosed funding topped 4.5 billion yuan, backed by industrial giants like Alibaba, Meituan, Tencent, JD, and Transsion, plus top VCs like Sequoia China and Qiming.

A key investment thesis here is the founding team—Li Yifan, Xiang Shaoqing, and Sun Kai—the original "Iron Triangle" behind lidar leader Hesai Technology. This is a second act with proven mass-production and engineering capabilities.

Moreover, while Sharpa started with hands, its goal is general robots. It plans to enter homes from 2028 after mastering specific industrial scenarios.

Its 22 billion yuan valuation, therefore, prices not just a "hand," but a next-generation platform integrating "dexterous hand-body-operational brain."

For dexterous hands to work, force and tactile sensing is essential, and this niche heated up in August.

In force sensing, six-axis force sensor leader BluePoint Touch closed a D round worth hundreds of millions, led by GAC Capital.

In tactile sensing, Panini, Dimon Robotics, Moliang Technology, and Linggan Robotics all secured funding.

This shift from "single actuation" to "multidimensional perception" highlights the drive to fix the industry's "perception layer"weakness.

If joints set the motion floor, force and tactile sensing define the interaction ceiling. In complex, unstructured scenarios like home service or precision assembly, robots must "feel" force and texture through high-precision feedback to operate smoothly and safely.

While both are hardware trends, the logic differs: actuators are about "certain payoffs" betting on volume; hands and sensors are "high-growth positioning" betting on the transition from "moving" to "working."

Their simultaneous rise signals one shift: the industry’s main thrust has moved from "making robots move" to "making them work with their hands."

The Unitree Lesson: When the "First Stock" Gets Priced

No matter how good the private market story sounds, it must face the test of public capital. In August, a clear benchmark emerged.

On August 19, Unitree Technology listed on the STAR Market. Priced at 150.80 yuan, the IPO valued the company at roughly 61 billion yuan with a P/E ratio of 219.23.

On debut, shares opened at 1,100 yuan—a 629% surge pushing the market cap to 444.9 billion yuan, marking the all-time high.

Image Credit: Screenshot from Eastmoney

But the rally didn't last. By the close on September 8, the stock had retreated to 523.68 yuan, cutting the market cap to 211.8 billion yuan. Both price and valuation shed roughly 52% from the peak.

In just over a dozen trading days, Unitree traced a "peak at listing" curve. This wasn't a sudden fundamental deterioration, but a correction from emotional pricing back to fundamentals.

Three factors drove this.

First, high valuation compounded by a specialized trading structure amplified volatility.

The static P/E briefly topped 800x, while the initial float was only 7.44% of shares. Intense demand squeezed into a tiny supply, combined with no price limits for the first five days and sector euphoria, pushed prices to extremes. Once sentiment cooled, gravity took over.

Nomura’s "buy" rating with a 370 yuan target—a far cry from the debut hype—underscores that short-term prices are a voting machine, but the long term is a weighing machine.

Second, interim earnings signaled caution.

In the first half of 2026, Unitree swung to a net profit, but adjusted net profit fell 19.34% year-on-year to 244 million yuan. Meanwhile, R&D costs jumped 152% and sales expenses 250%.

This mix—fast revenue growth but "rising sales, falling adjusted profit, and soaring costs"—shows Unitree is still in a "growth via investment" phase. R&D and channel expansion are eating profits, and scale effects haven't fully materialized.

When earnings growth lags behind valuation expansion, caution is natural.

Third, internal and external references cooled expectations.

Internally, management dialed down the heat. At the 2026 World Robot Conference, Unitree Chairman Wang Xingxing said the biggest bottleneck for general robots entering daily life is insufficient generalization capability. The industry is still 2 to 10 years away from its "ChatGPT moment."

Image Credit: UBTECH

Externally, UBTECH, listed earlier in Hong Kong, remains loss-making with a significantly lower valuation. The Hong Kong market has already priced in humanoid assets, acting as a mirror that highlights the premium in Unitree’s debut.

When founders pour cold water on the hype, capital markets ease off the accelerator.

For the sector, if a leading OEM like Unitree faces such a revaluation, startups with valuations of 10 or 20 billion yuan in the private market must reassess their IPO exit space. Cautious pricing for future listings is almost certain.

Conclusion

In retrospect, August’s 56 funding rounds were a collective "roadmap vote" from the industry: brains and data make robots "smarter," while hands, joints, and sensors make them "useful."

Unitree’s half-month correction, meanwhile, marked the weight of "delivery" for all players.

For practitioners, this means three things:

First, instead of telling a "full-stack story," clarify which irreplaceable link you own.

Second, rather than racing for funding speed, focus on tangible progress in mass production, orders, and data assets.

Third, don’t take comfort in private market valuations—use the public market’s profitability metrics to perform a health check today.