Unitree Robotics IPO 2026: What Its Financials Reveal About the Economics of Humanoid Robots

The short answer

Unitree Robotics is preparing to issue approximately 40.45 million shares on Shanghai’s STAR Market, equal to 10% of its post-offering share capital. Preliminary price inquiries are scheduled for August 5, the offer price is expected to be determined on August 6, final terms are due on August 7, and subscriptions are scheduled to open on August 10.

The company is seeking to raise approximately RMB4.202 billion, implying a post-offering valuation of roughly RMB42 billion, or about US$6.2 billion, if the offering is completed at the fundraising level described in its prospectus. The final market valuation has not yet been determined.

Unitree is not the first publicly listed humanoid robotics company. UBTECH Robotics has traded on the Hong Kong Stock Exchange since December 2023. What makes Unitree different is that it may become the first publicly traded benchmark combining high humanoid shipment volume, positive operating cash flow, substantial reported profits and a mass-market hardware strategy.

That distinction matters.

For the first time, investors may be able to compare privately valued humanoid companies with a public manufacturer that discloses revenue, margins, shipment volume, geographic exposure and operating costs.

Unitree IPO: key verified figures

These numbers come primarily from Unitree’s prospectus, Shanghai Stock Exchange disclosures and the company’s updated offering documents. Differences between some published figures, such as RMB1.699 billion versus RMB1.708 billion in 2025 revenue, reflect different versions of the filing rather than fundamentally different business results.

What Unitree’s IPO actually changes

For most of the recent humanoid robotics boom, valuations have been established through private financing rounds.

Private-market prices are negotiated among a small number of investors. They may include preferred-share rights, liquidation preferences, anti-dilution provisions, strategic relationships and assumptions about future financing. A headline post-money valuation therefore does not always represent the price that ordinary shares would command in a liquid public market.

A public listing introduces a different mechanism.

Unitree’s share price will eventually reflect changing expectations about revenue growth, margins, competition, government policy, product reliability and commercial deployment. Those expectations will be repriced continuously rather than only when the company raises another private round.

However, public-market pricing should not be confused with an objective measurement of technological value.

A STAR Market valuation may also contain:

• a scarcity premium for a rare publicly traded humanoid company;
• a domestic technology-policy premium;
• expectations of government-supported adoption;
• unusually strong retail demand;
• restrictions on the tradable share supply;
• different required returns from those applied by U.S. venture investors.

Unitree will therefore become an important valuation reference, but not a mechanically perfect comparable for every humanoid robotics company.

Unitree is not the first public humanoid robot company

One of the most important corrections to the original thesis is that the humanoid robotics industry already has a public-market benchmark.

UBTECH Robotics listed on the Hong Kong Stock Exchange on December 29, 2023, under ticker 9880.HK. The company described itself as the first humanoid robot company listed on the exchange.

Unitree’s significance is narrower but potentially more consequential.

It could become:

The first major public benchmark for a profitable, high-volume and comparatively low-cost humanoid robot manufacturer.

UBTECH’s listing established that a humanoid robotics company could trade publicly. Unitree’s listing may reveal how the market values a business that has already shipped thousands of humanoids and built a meaningful revenue base from both humanoid and quadruped platforms.

That is the more accurate—and more analytically useful—claim.

The financial result that separates Unitree from most competitors

Unitree reported approximately RMB1.7 billion in 2025 revenue, up roughly 335% from the prior year. Its profit attributable to shareholders excluding non-recurring items was approximately RMB590–600 million, while reported net income under the filing’s accounting presentation was materially lower at approximately RMB288 million.

The difference between reported net profit and profit excluding non-recurring items is important. Investors should not casually treat the larger figure as equivalent to conventional net income or free cash flow.

Nevertheless, the broader financial conclusion remains unusual for the sector:

• Unitree has substantial recognized revenue.
• It reports positive profits.
• It generated positive operating cash flow.
• Its core-business gross margin was approximately 60%.
• Its humanoid business achieved a gross margin of approximately 62.9% during the first nine months of 2025.

Most humanoid robot developers are still absorbing large research, manufacturing and deployment costs. Unitree appears to have reached a different stage: its early customers are already financing the next generation of its hardware and embodied-AI research.

Why Unitree’s gross margin is strategically important

A gross margin near 60% is unusual for a young hardware manufacturer, but it should not be interpreted as proof that Unitree has already solved general-purpose robotics.

The margin is better understood as evidence of four advantages.

1. Shared architecture across product families

Unitree spent years developing quadruped robots before entering humanoids. Motors, motor-control systems, encoders, power systems, motion-control software, mechanical design processes and supplier relationships could be reused across product generations.

That reduced the amount of technology that had to be created from zero when the H1 and G1 were introduced.

Unitree’s filing specifically identifies shared components and technical overlap between quadruped and humanoid platforms as contributors to its cost advantage.

2. Actuator-centered product development

Actuators are the mechanical and electrical joints that generate robot movement. In a humanoid robot, they combine a motor, transmission, encoder, driver electronics, bearings, thermal design and structural integration.

They are not simply components. They largely determine:

• torque;
• speed;
• backdrivability;
• energy consumption;
• thermal limits;
• mechanical precision;
• maintenance requirements;
• manufacturing cost.

Unitree’s architecture is frequently described as quasi-direct drive, or QDD: a relatively high-torque motor combined with a lower-ratio planetary transmission than the high-ratio strain-wave gearing commonly used in industrial robots.

The terminology needs care. There is no single reduction ratio used across the entire G1. In Robotopian’s teardown of one small G1 joint module, the two-stage planetary transmission produced an estimated total reduction ratio of approximately 20.58:1—not the 1:6 to 1:9 range sometimes repeated in secondary commentary.

3. Lower-cost manufacturing and faster iteration

A planetary transmission can generally be produced by a wider supplier base than a high-precision strain-wave reducer. Lower gearing can also improve impact tolerance and dynamic movement, although it places a greater torque and thermal burden on the motor.

This produces a fundamental engineering trade-off:

• high-ratio transmissions can generate greater static output from a smaller motor;
• QDD-style systems can be cheaper, more responsive and more backdrivable;
• lower-ratio systems may require larger motors, higher current and stronger thermal management.

SemiAnalysis argues that this architecture gives Unitree a faster actuator iteration cycle than competitors relying on multiple external motor and gearbox suppliers. That conclusion is an independent estimate rather than an audited company disclosure, but it is directionally consistent with Unitree’s rapid product cycle and declining prices.

4. A hybrid supply-chain model

Unitree should not be described as manufacturing every component internally.

Robotopian’s G1 teardown identified a hybrid architecture:

• self-developed or Unitree-designed joint electronics and motor architecture;
• externally supplied reducers and bearings;
• an Intel RealSense depth camera;
• a Livox 3D LiDAR;
• a Rockchip main processor;
• externally sourced memory and storage;
• several possible dexterous-hand suppliers.

The teardown did not identify an external motor brand on the examined joint, which is consistent with Unitree’s self-development claims. It is not, by itself, proof that every motor manufacturing process is vertically integrated.

The more defensible conclusion is that Unitree owns enough of the actuator design and system integration to control cost, performance and iteration speed while continuing to use China’s dense component ecosystem where external procurement is more efficient.

What teardown estimates reveal—and what they do not

Robotopian estimated the bill of materials for a base G1 at approximately RMB41,574, with joint modules accounting for RMB27,500 of that total. After estimated processing costs, the analysis calculated an operating product cost of approximately RMB44,600. These are analytical estimates, not audited cost disclosures.

SemiAnalysis separately estimated a bill of materials of approximately US$8,976 for a higher-end G1 configuration sold at an estimated pre-tax price of approximately US$27,300. Its configuration, timing and methodology differ from Robotopian’s base-model analysis, so the two estimates should not be treated as contradictory or directly interchangeable.

Both estimates point to the same strategic conclusion:

Unitree’s advantage is not simply cheap labor or high production volume. It is the ability to design a robot around a cost structure that competitors may find difficult to reproduce without changing their actuator architecture and supply chain.

That is similar to the logic that helped DJI scale consumer drones and BYD compound its advantage in electric vehicles. The analogy is useful, but it should remain an analogy—not a prediction that Unitree will necessarily achieve the same market dominance.

High margins do not mean the commercial problem is solved

Unitree shipped more than 5,500 humanoid robots in 2025, according to its prospectus. Reuters reported that humanoids represented approximately 51.5% of revenue during the first nine months of 2025, up from 27.6% in 2024.

Those figures establish real demand.

They do not establish that 5,500 robots are performing economically productive labor.

Many units have been purchased by:

• universities;
• research laboratories;
• embodied-AI developers;
• corporate R&D departments;
• system integrators;
• event and demonstration operators;
• early industrial-pilot programs.

These customers matter. Research platforms can create an ecosystem, generate training data, attract developers and finance product improvement. DJI’s early drone customers and NVIDIA’s early GPU customers similarly preceded much larger commercial markets.

But research shipment volume is not the same as industrial utilization.

The essential commercial questions are:

• How many units operate every day?
• What percentage of operating time is autonomous?
• How often do they require human intervention?
• What is the mean time between failures?
• How much maintenance is required?
• What is the productive throughput per hour?
• Can the robot produce a return after integration, software, safety and support costs?

Unitree’s IPO will provide better financial visibility, but it will not immediately answer all of these deployment questions.

The G1 is still primarily a development platform

The standard G1’s official maximum arm load is approximately 2 kilograms, while certain higher configurations are rated at approximately 3 kilograms. Unitree also lists a depth camera, 3D LiDAR, four-microphone array, Wi-Fi 6 and Bluetooth 5.2. Claims that the standard G1 contains six Intel RealSense cameras and integrated GPS are not supported by the official specification.

The payload specification places the G1 in a particular commercial category.

It may be useful for:

• AI and locomotion research;
• lightweight tote handling;
• teleoperation experiments;
• data collection;
• education;
• inspection;
• low-payload manipulation;
• entertainment and demonstrations.

It is not currently a direct replacement for a heavy-duty industrial arm or a production worker repeatedly handling large automotive components.

This does not make the G1 commercially irrelevant. It means its present advantage is accessibility and ecosystem development rather than maximum industrial payload.

Unitree versus Figure AI: a useful but imperfect comparison

Figure AI announced in September 2025 that it had secured more than US$1 billion in committed Series C capital at a US$39 billion post-money valuation. Figure does not publish the type of audited financial statements that Unitree is required to disclose for an IPO.

At Unitree’s prospectus-implied valuation of approximately RMB42 billion, or US$6.2 billion, Figure’s private valuation would be more than six times higher.

If Unitree eventually traded at RMB100 billion, equivalent to roughly US$14 billion at approximate recent exchange rates, Figure’s valuation would still be almost three times higher.

The gap is real, but several qualifications are necessary.

It is not a direct price-to-price comparison

Figure’s valuation reflects expectations around:

• its Helix embodied-AI system;
• software and data economics;
• potential household and commercial deployment;
• U.S. intellectual property;
• manufacturing scale;
• strategic partnerships;
• future general-purpose autonomy.

Unitree’s existing financial performance is driven more heavily by hardware platforms, research demand, supply-chain execution and current product sales.

Private preferred equity is not identical to public common stock

Private financing rounds may include contractual protections unavailable to public shareholders. A headline post-money valuation is therefore not always equivalent to a freely traded market capitalization.

Company valuation is not comparable with annual market revenue

Goldman Sachs estimated that the global humanoid robot market could reach US$38 billion in annual value by 2035, with a base-case shipment forecast of approximately 1.4 million units. A company can rationally be valued above one year of expected industry revenue if investors expect future growth, strong margins and many years of cash flow.

Therefore, the statement that Figure’s valuation “already exceeds the entire 2035 market” is rhetorically striking but financially misleading.

The better question is:

What share of future industry profits would Figure need to capture—and at what margin—to justify a US$39 billion valuation?

Unitree’s public trading multiples will make that question easier to examine.

What valuation would Unitree’s proposed offering imply?

Using approximately RMB1.708 billion in 2025 revenue:

• an implied RMB42 billion valuation equals approximately 24.6 times trailing revenue;
• an RMB100 billion valuation would equal approximately 58.5 times trailing revenue.

Using approximately RMB600 million in profit excluding non-recurring items:

• RMB42 billion equals approximately 70 times adjusted earnings;
• RMB100 billion equals approximately 167 times adjusted earnings.

Using reported net income of approximately RMB288 million:

• RMB42 billion equals approximately 146 times reported earnings;
• RMB100 billion equals approximately 348 times reported earnings.

These are not conventional mature-hardware valuations.

Investors paying these multiples would be valuing Unitree as a platform expected to expand into embodied-AI software, industrial deployment, data generation and new product categories—not merely as a company selling today’s G1 and H1 units.

That expectation creates both the opportunity and the risk.

China’s industrial policy is supportive—but should not be misread

China’s Ministry of Industry and Information Technology and the State-owned Assets Supervision and Administration Commission launched a 2026 initiative intended to move humanoid robots and embodied intelligence into real industrial, service and specialized environments.

The official objective is to validate representative applications, develop more than 100 high-value scenarios and build “10,000-unit-scale deployment capability” by the end of 2026. It is not an unconditional government order requiring exactly 10,000 completed commercial deployments by December 31.

This policy is still highly relevant.

It can provide:

• access to factories and operating environments;
• structured pilot customers;
• training data;
• public-sector coordination;
• deployment partners;
• faster iteration between manufacturers and users.

However, government-supported pilots do not automatically prove unit economics or sustainable demand. Investors should distinguish between:

1. equipment delivered to a pilot;
2. equipment operating regularly;
3. equipment replacing or augmenting paid labor;
4. equipment generating repeat commercial orders without subsidies.

Unitree’s next phase will be judged increasingly on the fourth category.

The 2026 slowdown is the most important financial warning

Unitree’s 2025 growth rate was extraordinary, but its 2026 figures show that the business is already entering a more demanding phase.

In the first quarter of 2026:

• revenue reached approximately RMB422.8 million, up 68.5%;
• profit excluding non-recurring items fell to approximately RMB40.3 million, down 52.6%;
• research, sales and other operating expenses increased substantially.

For the first half of 2026, Unitree forecast revenue of approximately RMB1.052–1.128 billion, representing growth of 35.6%–45.4%. It forecast profit excluding non-recurring items of RMB236–283 million, a decline of approximately 6.4%–22.0%.

Revenue growth above 35% remains strong.

The change is nevertheless meaningful because it indicates that:

• the 2025 comparison base was unusually low;
• price competition is intensifying;
• product development is becoming more expensive;
• embodied-AI investment is rising;
• sales and support infrastructure must expand;
• the easiest research and demonstration demand may already have been captured.

The IPO is therefore arriving at the moment Unitree must prove that it can transform one year of explosive hardware growth into a durable commercial platform.

U.S. restrictions are now a material business risk

International sales represented more than 40% of Unitree’s revenue in each of the disclosed reporting periods. The United States alone represented approximately 18.4%, 19.5% and 13.3% across those periods.

On July 28, 2026, the Federal Communications Commission added foreign-produced advanced robotic devices to its Covered List framework.

The practical position as of August 3 is:

• the restrictions primarily affect new models manufactured outside the United States;
• models that already hold FCC authorization are not automatically removed from the market;
• manufacturers may pursue exemptions or conditional approval;
• the FCC retains authority that could affect previously authorized products in the future.

Unitree stated in its filing that existing humanoid and quadruped models have FCC certification, while acknowledging that future models could face barriers to U.S. market entry.

This is not a total prohibition on every Unitree product currently used in the United States.

It is, however, a structural constraint on future product launches and a source of uncertainty for distributors, universities, enterprises and integration partners.

Unitree was also added in June 2026 to the U.S. Department of Defense’s Section 1260H list of companies the Pentagon identifies as connected to China’s military-civil fusion system. The designation should not be described as a universal commercial ban, but it increases federal procurement, compliance and reputational risks.

Cybersecurity risk: what is documented and what is not

The original draft made a broad claim that all audio, video, GPS and motion data collected by every Unitree robot was routinely transmitted to Chinese servers.

The sources reviewed do not adequately support that statement, and the official G1 specification does not list GPS as a standard sensor.

It should not be published as fact.

There are, however, documented cybersecurity issues that enterprise buyers should consider.

CVE-2025-2894 describes an undocumented CloudSail remote-access function affecting Unitree’s Go1 firmware. According to the U.S. National Vulnerability Database, possession of the relevant API key could permit complete remote control of the affected device. The CVE specifically identifies Go1; it should not be generalized automatically to every Unitree model.

Separate 2025 vulnerability disclosures affected shared firmware components used by the Go2, B2, G1 and H1. These included command-injection and authentication weaknesses accessible through local wireless interfaces in affected versions.

These disclosures do not prove that every current Unitree robot is compromised, nor do they establish that Unitree is currently exfiltrating customer data.

They do establish that mobile robots should be treated as networked industrial systems rather than ordinary mechanical equipment.

Enterprise deployments should include:

• network segmentation;
• outbound-traffic monitoring;
• firmware-version control;
• disabling unnecessary Bluetooth and cloud services;
• controlled administrator credentials;
• vulnerability scanning;
• software-bill-of-materials review;
• documented patch procedures;
• contractual data-handling requirements;
• physical emergency-stop and recovery processes.

The correct conclusion is not “every Unitree robot is unsafe.”

It is:

A connected humanoid robot should pass the same—or a higher—security review as an autonomous vehicle, industrial controller or networked camera system.

What buyers should learn from the IPO

For universities and research laboratories, Unitree’s financial strength could be positive.

A profitable manufacturer is generally better positioned to:

• maintain documentation;
• support replacement parts;
• continue firmware development;
• preserve an SDK ecosystem;
• introduce compatible product generations;
• remain in business throughout a multi-year research project.

For enterprise buyers, the analysis is more complicated.

Unitree’s attractive hardware price does not include the full cost of deployment. Total cost of ownership may also include:

• integration;
• autonomy software;
• teleoperation;
• end effectors;
• safety engineering;
• replacement actuators;
• batteries;
• spare parts;
• training;
• cybersecurity controls;
• local maintenance;
• downtime;
• regulatory compliance.

A US$20,000–30,000 robot that requires extensive engineering can be more expensive than a higher-priced system delivered with validated software, local service and a production warranty.

This is why the next phase of humanoid competition will not be decided by hardware price alone.

The winning platforms will combine:

1. affordable hardware;
2. reliable operation;
3. useful autonomy;
4. integration tools;
5. local service;
6. security controls;
7. measurable customer ROI.

What the next 18 months must prove

1. Can Unitree convert research shipments into repeat deployment orders?

The market needs evidence of customers buying second, third and fourth batches because the first batch produced measurable operational value.

2. Can humanoid gross margins remain high as prices decline?

Unitree’s average humanoid selling price fell substantially between 2023 and 2025. High margins must survive continued competition, lower product prices and rising support costs.

3. Can the G1 architecture support longer duty cycles?

Dynamic demonstrations prove control quality. Industrial customers care about thermal performance, uptime, fault recovery and maintenance intervals.

4. Can Unitree build a software and service layer?

Hardware scale alone may not capture the highest-value part of embodied intelligence. Unitree must show progress in models, data, fleet management, deployment tooling and developer support.

5. Can overseas revenue withstand regulatory fragmentation?

The United States, Europe, the Middle East and Asia may develop different cybersecurity, radio, privacy and procurement requirements. A globally successful platform will need local compliance and service structures.

6. Can public-market expectations remain connected to operating results?

At an implied minimum valuation near RMB42 billion, Unitree would already trade at a premium growth multiple. A much higher opening valuation would require exceptional confidence in future deployment economics.

Final assessment

Unitree has already proved something important:

Humanoid robots can be manufactured and sold at meaningful volume by a company that reports positive profits and strong gross margins.

That is not the same as proving that general-purpose humanoid labor has arrived.

Unitree’s current business is supported by a combination of research demand, developer adoption, early enterprise pilots, demonstrations, quadruped revenue, cost-efficient hardware and China’s unusually deep robotics supply chain.

Its IPO will therefore price two assets at the same time.

The first is visible:

• existing revenue;
• current margins;
• more than 5,500 humanoid shipments;
• actuator and manufacturing capability;
• a large research and developer ecosystem.

The second is still speculative:

• general-purpose embodied intelligence;
• recurring industrial deployments;
• scalable autonomy;
• a large software and service business;
• sustained global market access.

The public market will not settle the humanoid robotics debate in one day.

But it will force the debate to use better numbers.

And that may be the most important consequence of Unitree’s IPO.