Liang Wenfeng's Unitree Bet Is Technology News, but the 700 Million Yuan Gain Is Not Real Yet

Liang Wenfeng is tied to a Unitree Robotics investment that prompted a 700 million yuan profit claim across Chinese technology news feeds. The verified transaction is more modest and more interesting. DeepSeek's operating company received 933,400 Unitree shares through the robot maker's initial public offering.

That allocation cost about 141 million yuan and carries a 36-month lockup. The viral gain is therefore neither cash nor a freely tradable paper profit. It depends on Unitree reaching a future market value that has not been established through regular trading.

The distinction matters because the investor is DeepSeek's corporate entity, not Liang in his personal brokerage account. It also matters because Unitree and DeepSeek signed a strategic cooperation memorandum. Their arrangement connects large AI models, computing infrastructure, and physical robots through a formal commercial relationship.

This is not simply a famous founder winning an IPO lottery. It is a long-duration bet that two prominent Chinese technology companies can make embodied AI commercially useful before demanding valuations outrun operating results.

What DeepSeek Actually Bought

DeepSeek received a strategic allocation with business obligations and a long lockup, not an ordinary short-term IPO position.

On August 6, 2026, Unitree disclosed the investors participating in its strategic placement. The group included state-backed capital, Tencent-related entities, and Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Company.

That legal entity operates DeepSeek and is associated with founder Liang Wenfeng. Public coverage often shortens this relationship into “Liang bought Unitree.” That phrasing is memorable, but it obscures who owns the shares.

DeepSeek's company received 933,400 shares, representing about 2.31 percent of Unitree's IPO issuance. The allocation required approximately 141 million yuan and cannot be sold for 36 months.

Those figures come from Unitree's issuance disclosures, as summarized in a detailed report on the strategic allocation. The same disclosures show that strategic investors collectively committed to about 8.09 million shares, or 20 percent of the offering.

Unitree planned to issue approximately 40.45 million new shares. That represents 10 percent of its enlarged share capital. The company expected to raise about 6.10 billion yuan through the offering.

Demand during institutional price discovery was heavy. Valid indications covered more than 73 billion shares, vastly exceeding the number offered. Strong demand helped create the conditions for dramatic profit projections before public trading established a dependable market price.

The allocation followed regulatory approval rather than an informal private deal. China's securities regulator approved the offering in July, after the Shanghai Stock Exchange accepted Unitree's application on March 20 and reviewed it during the following months.

The regulatory approval required Unitree to follow its submitted prospectus and underwriting plan. That framework separates strategic investors from ordinary online subscribers.

Strategic participants receive larger, negotiated allocations because they are expected to contribute more than capital. In return, they accept restrictions that ordinary market traders do not face.

DeepSeek cannot simply sell the shares during an early trading surge. Its 36-month restriction extends far beyond the short window that normally produces viral IPO profit screenshots.

That makes the word “gain” unusually slippery. A headline can calculate a hypothetical increase from an assumed market price. DeepSeek still cannot convert that calculation into cash while the lockup remains active.

As of the August 13 reporting cutoff, the available exchange and issuance materials did not establish a completed sale by DeepSeek. They also did not support describing 700 million yuan as realized income.

The precise statement is narrower. DeepSeek owns a restricted Unitree allocation, and that position would show a large theoretical gain if Unitree traded near the valuation assumed by commentators.

That is a conditional calculation, not a completed financial event.

Why the 700 Million Yuan Figure Needs a Warning Label

A gain above 700 million yuan requires Unitree's shares to reach roughly six times the IPO allocation level and remain valuable through a three-year lockup.

The arithmetic behind the claim can be reconstructed without treating it as fact. DeepSeek controls 933,400 allocated shares and invested about 141 million yuan. For the position's appreciation to exceed 700 million yuan, its market value would need to rise above approximately 841 million yuan.

That represents close to a sixfold increase over the amount invested. It is not an impossible scenario for a highly sought-after technology IPO, but it is not a result contained in the allocation announcement.

The viral headline replaces several unresolved steps with one confident number. Unitree must begin regular market trading, investors must assign it the assumed valuation, and that valuation must survive volatility.

Even then, DeepSeek cannot freely sell its restricted position. The company must wait through a period in which Unitree's revenue mix, competition, operating expenses, and robotics market can change substantially.

This creates three different concepts that readers should not merge.

First, there is allocation value. That is the amount DeepSeek committed for its shares.

Second, there is a hypothetical paper gain. Commentators can derive this from an assumed future quotation, even before that quotation becomes dependable.

Third, there is realized profit. That requires an actual sale after restrictions expire, with transaction costs and other financial consequences considered.

Only the first concept is fully documented today. The second is a scenario. The third has not happened.

The headline also attributes corporate assets directly to Liang Wenfeng. Liang is DeepSeek's founder and an important owner, but a gain recorded by the operating company would not automatically become personal cash.

Corporate ownership, shareholder ownership, and founder wealth are related but separate. Treating them as interchangeable exaggerates the immediacy of the story.

The 36-month restriction deepens that problem. A restricted investor can watch a quoted value rise and fall without having the legal ability to capture it.

Unitree's management has already acknowledged concerns about speculative trading. During an August 7 investor session, founder Wang Xingxing said the company would provide detailed information through lawful and authoritative channels.

He also said he wanted investors to buy because they recognized the company's value, rather than for speculation. That response, covered alongside the DeepSeek partnership, directly addresses the tension behind the viral estimate.

High demand does not eliminate valuation risk. Unitree's offering implied valuation multiples well above cited industry averages, including a diluted earnings multiple above 200 based on the company's disclosed methodology.

Such comparisons are imperfect because Unitree combines robot hardware, components, software, and embodied AI research. They still show that investors are paying for substantial future expansion.

That future value must come from more than popular videos or limited research deployments. Unitree must sell more robots, preserve margins, and move humanoid systems into repeatable applications.

If those results arrive, today's allocation may prove attractive. If adoption disappoints, the lockup can transform an apparent windfall into a long exposure to declining expectations.

The 700 million yuan claim is best read as a measure of market enthusiasm. It is not evidence that Liang or DeepSeek has already earned that amount.

The Real Technology News Is the DeepSeek and Unitree Alliance

The strategic placement formalizes an AI-to-robotics partnership that could matter more than any first-day stock move.

Unitree and DeepSeek signed a memorandum covering three areas. The first is joint research and product development involving large AI models and embodied intelligence.

Embodied intelligence describes AI systems that perceive and act through physical machines. Instead of only generating text or images, these systems must connect sensory input, planning, movement, and real-world feedback.

The second area concerns robot procurement and technical solutions. DeepSeek agreed that, under equal conditions, it would prioritize Unitree when exploring embodied AI applications or purchasing high-performance general-purpose robots.

The third area covers model services and infrastructure. Unitree agreed that, under equal conditions, it would prioritize DeepSeek for model adaptation, training services, and related technical solutions.

DeepSeek may also support model architecture, computing clusters, and data center operations when required. These commitments are conditional, but they give both companies a reason to test their technologies together.

The pairing is strategically coherent. DeepSeek develops models and model-serving capabilities. Unitree builds the physical platforms that can collect data, execute actions, and expose model failures outside controlled software environments.

A humanoid robot needs more than fluent language. It must interpret camera and sensor data, maintain balance, plan movement, recover from errors, and operate safely around people or equipment.

Large models can help with task interpretation and higher-level planning. They do not remove the need for dependable control systems, mechanical design, actuators, batteries, and carefully collected training data.

Unitree contributes those hardware and control capabilities. DeepSeek contributes model research and computing expertise. Their memorandum creates a route for combining the layers without requiring either company to build the entire stack alone.

This cooperation also gives DeepSeek a physical testing environment. Many model benchmarks measure answers on static datasets. Robots reveal different problems, including latency, uncertain observations, and the cost of incorrect actions.

For Unitree, the relationship offers another possible model supplier. It may reduce dependence on external systems that are difficult to customize or expensive to operate at scale.

Yet the memorandum does not guarantee exclusive cooperation. Its repeated “under equal conditions” language preserves flexibility for both parties.

Unitree can still evaluate other model architectures. DeepSeek can still work with other robot companies. That flexibility matters because embodied AI has not settled around one technical stack.

The strategic placement therefore serves two functions. It aligns incentives for several years, and it signals that each company sees value in the other's technical assets.

DeepSeek's shares remain locked during the period when the partnership should produce measurable work. That alignment makes the investment different from a fund buying IPO shares for a quick market return.

The real test is whether the agreement produces deployed systems. Readers should look for named pilots, jointly developed models, repeat robot purchases, or infrastructure contracts.

Without those outputs, the memorandum remains a promising framework. With them, it becomes evidence that Chinese model developers and robot manufacturers are forming an integrated supply chain for embodied AI.

That is the durable technology news inside the financial headline.

Unitree's Growth Supports the Excitement, but Its Costs Are Rising

Unitree reached profitability and expanded rapidly, yet its latest quarterly figures show that growth does not automatically protect earnings.

Unitree reported approximately 1.70 billion yuan in 2025 revenue. That was up from about 393 million yuan in 2024 and 159 million yuan in 2023.

The two-year compound growth rate exceeded 226 percent. Unitree also reported approximately 278 million yuan in 2025 net income attributable to shareholders.

Profit excluding nonrecurring items was about 591 million yuan. The gap between those measures reflects accounting effects that readers should examine before applying a simple earnings multiple.

The company's core-business gross margin reached about 60 percent, according to figures highlighted by the exchange review. These results distinguish Unitree from several robot makers that remain deeply unprofitable.

Product mix changed quickly. Humanoid robot sales grew from a small portion of revenue into a major business line. Four-legged robots continued to serve commercial, educational, and industrial uses.

Unitree's prospectus says its humanoid products mainly served research, education, data collection, and emerging industry applications during the reporting period. That mix is important.

Selling robots to laboratories and developers proves demand for a platform. It does not prove that humanoids can perform general labor economically inside factories, warehouses, offices, or homes.

Research demand can also be uneven. Customers may buy a limited number of machines for experiments without moving to fleet deployments.

Unitree's first-quarter 2026 results illustrate the pressure. Revenue increased 68.49 percent from the prior-year period, reaching about 423 million yuan.

However, net income fell 47.69 percent. Profit excluding nonrecurring items fell 52.55 percent as research, sales, and other operating expenses increased.

Those figures appear in the updated IPO prospectus. They do not invalidate the growth story, but they show what Unitree must fund next.

Robot companies face several expensive tasks at once. They must improve hardware, train control and perception systems, expand production, support customers, and develop applications.

A successful demonstration can generate attention quickly. Reliable deployment requires longer testing, maintenance, spare parts, safety procedures, and integration with customer workflows.

Unitree also operates in a crowded market. UBTech has pursued industrial humanoid deployments and listed in Hong Kong in 2023. AgiBot has emphasized production volume and varied commercial scenarios.

DEEP Robotics focuses heavily on industrial quadrupeds. Figure AI and Tesla are developing humanoid systems around manufacturing and general labor ambitions.

These companies use different definitions for shipments, deliveries, deployments, and orders. Direct comparisons therefore require caution.

Unitree's advantage is not simply that it has an expressive humanoid robot. Its stronger case rests on existing robot shipments, internal component capabilities, manufacturing experience, and positive annual earnings.

Its risk is that public-market expectations now demand continued growth across all those dimensions. A company can lead a young category and still disappoint investors if adoption takes longer than expected.

The partnership with DeepSeek may help Unitree improve higher-level intelligence. It cannot by itself solve mechanical reliability, battery duration, safety, or customer return on investment.

Likewise, a strong robot platform gives DeepSeek valuable real-world data. It does not guarantee that a language model will become an effective robot controller.

The alliance offers a plausible mechanism for progress. The valuation assumes that mechanism will produce commercial results at unusual speed.

The Main Contest Is Strategic Value Versus Tradable Profit

The investment should be judged by the partnership's operating results, while the viral narrative judges it through an unavailable exit price.

Strategic value develops through shared technology, procurement, and product integration. Tradable profit develops when an investor can sell an asset for more than it paid.

DeepSeek's position belongs firmly in the first category today. The 36-month lockup was designed to keep the investor aligned with Unitree after the offering.

That structure places pressure on both companies. Unitree must show that DeepSeek improves its models or computing operations. DeepSeek must show that access to physical robots adds value beyond a financial investment.

A partnership announcement alone does not meet that standard. The companies need technical milestones that outsiders can evaluate.

One milestone would be a jointly identified model running on Unitree hardware. Useful disclosure would explain what the model controls, what data it uses, and which tasks improved.

Another would be a repeatable commercial deployment. A pilot involving one robot in a controlled demonstration offers little evidence about fleet economics.

A stronger signal would include multiple machines, sustained operating hours, intervention rates, task completion data, and a customer willing to discuss results.

DeepSeek's procurement behavior also matters. The memorandum gives Unitree preference under equal conditions, but it does not commit DeepSeek to a minimum order.

No disclosed minimum means the relationship can remain strategically friendly without generating meaningful hardware revenue.

The same applies to DeepSeek's model and infrastructure services. Unitree may evaluate those services without becoming a major paying customer.

These uncertainties should shape any assessment of the investment. The companies have aligned incentives, but they have not disclosed guaranteed revenue.

There is also governance risk in attributing every benefit to Liang. DeepSeek's operating company has its own obligations, shareholders, employees, and capital needs.

A future increase in Unitree's share value would strengthen DeepSeek's balance sheet. It would not automatically represent personal disposable wealth for its founder.

The market-price assumption presents another risk. Newly listed technology stocks can experience sharp early moves because few shares are available for trading.

Strategic allocations and other restrictions reduce the immediate float. Scarcity can amplify price movements without revealing a stable long-term valuation.

By the time DeepSeek's shares become transferable, more Unitree stock may be available, competitors may have launched new products, and industry expectations may have changed.

A quotation during the first week and an exit price three years later answer different questions. The first measures immediate demand. The second reflects years of execution and market change.

This is why the 700 million yuan figure is simultaneously possible and premature. It describes a scenario in which market enthusiasm reaches an extreme level.

It does not account for the lockup, future dilution, competitive pressure, or the possibility that Unitree's earnings growth slows.

The strongest bullish argument is operational. Unitree already generates meaningful revenue, has reported annual profitability, and can pair its machines with a notable model developer.

The strongest skeptical argument is also operational. Humanoid robot adoption remains early, Unitree's recent expenses have risen, and the partnership contains conditional preferences rather than binding purchase commitments.

Investors and technology readers should keep both points in view. The allocation is not meaningless hype, but neither is it a realized windfall.

What This Technology News Story Must Prove Next

Three observable signals will determine whether DeepSeek's Unitree allocation becomes a strategic success or merely an expensive headline.

The first signal is Unitree's formal listing and the quality of public price discovery. Readers should separate any opening move from sustained trading over several reporting periods.

A large early increase would strengthen the paper-gain narrative, but not the realized-profit claim. Weak or volatile trading would expose how much the viral estimate depended on speculation.

The second signal is a disclosed DeepSeek and Unitree deployment. The companies need to move beyond broad cooperation language and identify a model, robot, customer, or internal application.

A named pilot with measurable task performance would support the strategic thesis. Another memorandum or promotional demonstration would add little evidence.

The third signal is Unitree's post-IPO financial performance. Revenue growth must translate into durable earnings while the company spends on models, products, manufacturing, and international expansion.

Watch the relationship between humanoid revenue, research expenses, gross margin, and operating cash generation. Those measures will show whether Unitree is scaling a business or funding an extended development cycle.

The three signals must be read together. A rising stock without commercial progress would support the speculation critique.

A successful deployment without improving economics would prove technical value but leave valuation questions unresolved. Strong operations combined with repeat deployments would offer the clearest support for DeepSeek's investment.

The 36-month lockup gives this assessment an unusually clear timeline. DeepSeek will remain exposed while Unitree tries to convert early leadership into repeatable commercial demand.

That exposure is the story's core reversal. What looks like a quick IPO win is actually a multiyear commitment to one of AI's least settled markets.

For developers, the partnership may reveal how large models connect with real-time control systems. For enterprise buyers, it may clarify whether humanoid robots can move beyond research and demonstrations.

Knowledge workers should also care because embodied AI changes the consequences of model errors. A flawed text response is inconvenient. A flawed physical action can damage equipment or injure someone.

The next meaningful update will not be another theoretical wealth estimate. It will be evidence that DeepSeek's models make Unitree's robots more capable, reliable, or economical in a documented setting.

Until then, treat the 700 million yuan number as a valuation scenario. Treat the 933,400 restricted shares and the three-part cooperation agreement as the verified event.

That distinction keeps the financial excitement in proportion and preserves the more consequential question: can two leading Chinese AI companies turn a locked investment into working embodied intelligence?