Unitree Raises $9B IPO Backed by DeepSeek and Tencent

Unitree Robotics priced its Shanghai IPO on August 6, 2026, at a valuation of RMB 60.99 billion, roughly $9 billion. That number is striking not because of its size, but because of what sits behind it: a humanoid robotics company that is already profitable, already shipping more than five thousand units a year, and now backed by DeepSeek and Tencent under a three-year strategic lock-up. While Western humanoid robot makers race to justify nine-figure funding rounds on zero revenue, Unitree just handed the public markets a P&L. The comparison is uncomfortable for almost everyone outside China.

What Actually Happened

Unitree Robotics priced its STAR Market IPO at RMB 150.8 per share, issuing 40.45 million new shares to raise RMB 6.1 billion, well above its original fundraising target of RMB 4.2 billion. The IPO subscription window opens August 10. According to TrendForce, the company was listed at a valuation of RMB 60.99 billion, nearly $9 billion at current exchange rates. The raise was 45% higher than the original target, suggesting demand from strategic and institutional investors materially exceeded expectations before the public subscription window even opened.

The investor lineup reads less like a venture cap table and more like a Chinese industrial policy balance sheet. DeepSeek invested RMB 141 million under a three-year lock-up and will co-develop AI models and embodied AI technologies with Unitree. Tencent Holdings' subsidiary holds a 2.23% stake. Other confirmed strategic investors include China National Petroleum Corp, China Southern Power Grid, China Telecom, Citic Securities, and the National Council for Social Security Fund. According to Caixin Global, the breadth of state-linked investors is deliberate: Unitree is being positioned as national industrial infrastructure, not a consumer electronics venture. That positioning matters for how regulators, banks, and downstream customers will treat the company over the next decade.

Unitree's financial profile is the detail that most clearly differentiates it from the global humanoid robotics field. The company generated RMB 1.7 billion in revenue and RMB 591 million in adjusted net profit in 2025, making it one of the only profitable humanoid robotics companies anywhere in the world. Unitree shipped more than 5,500 humanoid units globally in 2025, capturing approximately 32.4% of global unit shipments by volume, according to South China Morning Post. The company is targeting annual capacity of 30,000 units by 2027, a nearly sixfold increase from 2025 output, and has already stated that 2026 shipments will at minimum double the 2025 figure.

Why This Matters More Than People Think

The Unitree IPO is the first major public market test of whether humanoid robotics as an industry category can trade at growth-tech multiples with liquid equity. At $9 billion on RMB 1.7 billion in 2025 revenue, the implied price-to-sales multiple sits around 5x on actual reported financials. That is not a speculative premium: it is a standard growth-stage technology multiple for a company doubling unit output annually. The market is saying this category is real and this company is viable. That verdict matters enormously for the entire humanoid robotics sector, because it creates a public-market pricing anchor that private-round investors and competitors can now reference. This is the IPO that sets the exchange rate between robotics ambition and actual capital.

The DeepSeek partnership is strategically more consequential than the investment size suggests. DeepSeek's core capability is building highly efficient large-scale AI models at a fraction of the compute cost of Western equivalents, as it demonstrated with DeepSeek-V3 and its successor models. Unitree's current constraint is not its hardware platform, which is already cost-competitive and mechanically capable. The constraint is the intelligence layer: making a robot that can navigate genuinely unstructured environments, adapt to novel tasks, and operate without continuous human supervision. A three-year co-development agreement on embodied AI models is a direct bet that the next step-change in humanoid capability will come from the AI stack, not from another actuator revision.

The geopolitical reading of the investor base is the part of this story most Western coverage skips. Unitree's capital table now includes China's national petroleum company, the national power grid, the national social security fund, the country's largest telecommunications provider, and its two dominant consumer technology conglomerates. This is not a startup raising venture capital: it is a national industrial asset being structured for public markets. The implications for how Western policymakers and businesses should think about competing with Chinese humanoid robotics carry real weight for policymakers and businesses. Critics argue that state-backed investors distort normal market signals and could sustain Unitree through losses that a purely commercial entity could not. That is a legitimate concern. However, the bear case ignores a more immediate fact: Unitree is already profitable, so the state capital is additive to a commercially viable operation, not a substitute for one.

The Competitive Landscape

The comparison to Figure AI is instructive and, for Figure's investors, uncomfortable. Figure AI was last valued at $39 billion in its 2025 funding round, roughly 4.3x Unitree's IPO valuation. Figure has a commercially deployed fleet of 40 Figure 03 units at BMW's largest assembly plant, billing at approximately $25 per robot-operating-hour. Figure has no reported revenue at a fraction of Unitree's unit output. The implicit claim in the 4.3x premium is that the Western humanoid market commands fundamentally higher long-term margins, or that Figure's technology is so superior that unit economics do not yet apply. Both assumptions remain unproven in publicly available data, and Unitree's IPO now gives skeptics a quantitative benchmark to push back with.

AgiBot, Unitree's closest Chinese domestic rival, is simultaneously pursuing a Hong Kong Stock Exchange listing at a valuation of HK$40-50 billion. AgiBot shipped 5,100 humanoid units in 2025 and has reached cumulative deliveries of 15,000 embodied intelligence robots across logistics, automotive pilot programs, and research deployments. The concurrent IPO processes at both Unitree and AgiBot mark a structural transition in Chinese humanoid robotics: the sector is moving from private-round capital formation to liquid public equity markets, which imposes transparency, quarterly reporting, and analyst scrutiny on companies that previously operated without external accountability. For Western investors trying to assess the Chinese humanoid field, the IPO filings will be the first reliable source of unit economics data at scale. According to WION News, the AgiBot listing could follow as early as this year.

The closest historical parallel is the wave of Chinese EV manufacturers that listed on global exchanges between 2018 and 2022. NIO, Li Auto, and XPeng were unprofitable at the time of listing, trading on the promise of a market that China was clearly winning on unit economics and government support. Those companies have since delivered genuine scale, though with widely varying profitability outcomes. Unitree is attempting a harder version of the same move: listing as a profitable company and inviting the market to value the growth trajectory on top of a demonstrated earnings base. The historical precedent suggests that profitable Chinese hardware manufacturers with coordinated state-backed distribution can sustain higher public-market premiums than their Western competitors typically project. The question is whether the humanoid robot market grows fast enough to justify the implied multiple by 2028.

Hidden Insight: Unitree's Real Bet Is the Software Data Flywheel

The mainstream story about Unitree focuses on manufacturing cost leadership: the G1 humanoid retails at roughly $16,000, dramatically undercutting Western competitors whose comparable units list at $100,000 or more. That cost advantage is real and documented. But it understates the deeper strategic logic that the DeepSeek partnership reveals. Unitree's leadership appears to understand that cost leadership is a temporary moat, because manufacturing efficiency is replicable given sufficient capital. The durable competitive advantage in humanoid robotics is not building a cheap robot: it is training a robot that can perform tasks no competitor's robot handles reliably.

Embodied AI is the technical term for the challenge of combining large-scale AI models with physical robotic systems operating in unstructured real-world environments. The fundamental bottleneck is training data: you need millions of hours of robots performing real tasks in real environments to train the perception, planning, and manipulation models that make a humanoid genuinely useful. Unitree, with 5,500 deployed units in 2025 and a stated target of 30,000 units by 2027, will generate orders of magnitude more real-world embodied AI training data in the next 24 months than any Western competitor currently operating at comparable scale. Data advantages in AI compound nonlinearly over time, and Unitree is quietly constructing a data moat that will be extremely difficult to close from behind.

The risk is that data volume alone does not determine embodied AI quality. The diversity of tasks, the richness of the environments, and the rigor of the annotation and curation pipeline matter as much as raw data quantity. Unitree's current deployments are concentrated in relatively structured environments: logistics sorting, simple assembly tasks, and research demonstrations. Boston Dynamics' electric Atlas, with its more limited but more varied deployments at Hyundai and Google DeepMind, may generate more useful training signal per robot-hour even at a fraction of Unitree's unit count. The bear case for Unitree's data flywheel thesis is that it produces a very large dataset that is too homogeneous to train genuinely general-purpose behavior, while Western competitors with smaller but more diverse deployment environments develop models that generalize better to novel real-world conditions.

The DeepSeek angle is the sleeper variable that almost no one in Western robotics coverage is examining closely. DeepSeek built its competitive large language models at an estimated fraction of what OpenAI spent per training run, by applying mixture-of-experts architecture innovations and aggressive inference optimization. If DeepSeek applies similar efficiency principles to embodied AI model training, Unitree could access robot intelligence that is both cheaper to develop per unit of capability and faster to iterate than anything the computationally resource-intensive Western labs are building. The combination of Unitree's field deployment scale and DeepSeek's training efficiency is the scenario that produces a discontinuous capability jump, not a gradual improvement curve. That is a speculative scenario, but it is not an implausible one, and the $9 billion IPO valuation is partly reflecting the optionality around it.

What to Watch Next

The IPO subscription window opens August 10, and the oversubscription rate will be the first concrete data point. If the offering is 50x or more oversubscribed, it will accelerate AgiBot's Hong Kong listing timeline and signal that public market investors are treating humanoid robotics as a durable technology category rather than a venture theme. A subscription rate below 20x would suggest more skepticism than the private-round valuations implied, and would likely slow other Chinese robotics companies from pursuing public listings in the near term. Watch the subscription outcome within 48 hours of the window opening.

Over the 90-day window following the listing, watch for any updates on the DeepSeek co-development agreement. A concrete embodied AI model benchmark, a new commercial deployment enabled by AI model improvements, or any announcement of a joint product would signal that the partnership is operational rather than ceremonial. If Unitree's robots begin demonstrating visibly better task performance in unstructured environments within two quarters of the DeepSeek partnership disclosure, the market will reassess the company as a software-enabled hardware platform rather than a pure manufacturing story, which would support a large multiple expansion from the IPO price.

Over the 180-day window, watch whether any Western humanoid robotics company attempts a public listing in response. Figure AI's $39 billion private valuation creates obvious strategic pressure: either demonstrate comparable financial metrics in a public forum or reduce the gap by cutting burn and expanding revenue before the Unitree comparison becomes a standard investor question. The Unitree IPO subscription performance will function as a price discovery event for the entire global humanoid robotics sector, giving Figure, Agility Robotics, and Hyundai's Boston Dynamics a cleaner read on the public market's current willingness to pay, and creating urgency around anyone whose private valuation now looks optically high.

Unitree just proved that a profitable, state-backed humanoid robot maker can price a $9 billion IPO. The Western robotics race just got a public market scoreboard.

---