Signals Inbox·August 22, 2026·Robotics

Is Dexmal really worth $3B today?

Dexmal’s $3B target looks aggressive but plausible today: the peer valuations are real, Atomix gives it revenue and deployment data, and the missing proof is whether Dexmal’s own AI layer can grow into the price.

Q1Did Dexmal actually raise money at a $3B valuation?

The $3 billion Dexmal figure currently refers to a valuation being discussed in an ongoing fundraising round, so investors have not yet validated that price in a completed deal.

Bloomberg’s latest reporting says founder Tang Wenbin is seeking roughly RMB20 billion for the company. The talks are still underway. That is an important distinction when judging the headline: a founder can ask for RMB20 billion, but the valuation becomes much stronger evidence once outside investors actually wire money at that price.

Dexmal does have recent completed financings behind it. The company raised close to RMB1 billion across its A and A+ rounds in late 2025, with NIO Capital leading the A round and Alibaba investing alone in the A+ round. It later combined with logistics-robotics company Atomix and closed another strategic financing backed by Zhipu AI, StepFun, SenseTime, Alibaba, Huaqin and SAIC Hengxu.

For now, $3 billion is Dexmal’s current market ask, not its last unquestionably completed valuation. If the present round closes near that level, the debate becomes much cleaner.

Q2How can a company founded last year already be asking for $3B?

Dexmal reached its current valuation range unusually fast because the company we see today contains years of older robotics work, an acquired commercial operation and a much broader product stack than the startup had at launch.

Dexmal itself was created in March 2025, but founder Tang Wenbin had already spent more than a decade building computer-vision and robotics systems. He co-founded Megvii in 2011 and later worked extensively on warehouse automation. Several other Dexmal executives followed the same path.

Then came the Atomix combination. Atomix grew out of Megvii’s logistics-robotics activity and already had working systems inside customer warehouses. By absorbing that business, Dexmal suddenly gained customers, engineers, deployed robots and existing revenue alongside its newer foundation-model work.

The valuation trajectory still looks wild. Financing databases put Dexmal around RMB4.5 billion after its A+ round and around RMB6 billion near the Atomix transaction. The current fundraising target is more than four times the earlier estimate in less than a year.

Yet the asset changed heavily during that period. Investors are now looking at DM0.5, DexOS, MaaS, Apex, Ferrata, the Atomix deployment base and an experienced Megvii team under one company. That helps explain the speed, even if it does not automatically justify the final price.

Q3Is Dexmal actually a RMB1B revenue company?

Dexmal can reasonably claim a combined business approaching RMB1 billion in annual revenue, although the identifiable sales currently come overwhelmingly from Atomix rather than Dexmal’s newer embodied-AI products.

When the two companies combined, 36Kr reported that Atomix was approaching RMB1 billion of annual revenue. That gives Dexmal something many young embodied-AI companies still lack: a real commercial base.

The gap appears when we look for revenue from DM0.5, DexOS, MaaS, Apex or DexDev. Dexmal has disclosed pilots, research customers and early commercial work, but it has never published a standalone revenue figure for those businesses. Tang Wenbin’s team has also been quite open about the fact that large-scale deployment remains a work in progress.

We therefore use approximately RMB1 billion as the best available revenue anchor for the combined group, with medium confidence rather than treating it like audited consolidated revenue.

The quality of that revenue is still hard to judge. We do not know Dexmal’s consolidated gross margin, recurring-revenue share, customer concentration or how much revenue comes from software versus hardware, integration and project work.

Q4Is paying about 20x sales for Dexmal crazy today?

A roughly 20x revenue multiple is expensive, but today’s Chinese robotics market has already accepted similar multiples for several companies with real revenue, while Unitree’s latest public-market valuation has gone vastly further.

Using the reported combined revenue as our denominator puts Dexmal around 20x sales. That would look extreme for a conventional warehouse-automation company. For a company being priced as an embodied-AI platform, the comparison has changed dramatically.

UBTECH generated about RMB2 billion of revenue in 2025. Its Hong Kong market capitalization has recently been around HKD40 billion to HKD47 billion, which translates into a sales multiple in the low 20s after adjusting for currencies. Dobot sits in roughly the same range, with 2025 revenue of RMB492 million and a recent market capitalization near HKD12 billion.

Unitree has pushed the ceiling much higher. The company entered the public market with 2025 revenue of RMB1.7 billion and an IPO sales multiple around 36x. Its shares then exploded on their first day of trading, leaving Unitree worth roughly RMB342 billion at the close. Against last year’s revenue, that is about 200x sales.

Unitree’s first trading days are a terrible fair-value benchmark for Dexmal. They do show how aggressively investors are currently pricing scarce exposure to Chinese robotics. Against that backdrop, Dexmal at around 20x looks expensive rather than bizarre.

Dexmal against current robotics sales multiples

Q5Are investors already paying $3B for similar Chinese robotics startups?

Yes, and this is probably the strongest argument that Dexmal’s current target can clear: RMB20 billion has become a real private-market valuation tier for leading Chinese embodied-AI companies.

X Square Robot gives us the cleanest comparison. The company recently completed four consecutive rounds through Series C and confirmed a post-money valuation above RMB20 billion, or roughly $2.8 billion. Those rounds actually closed. Its investor list includes Meituan, Alibaba, ByteDance and Xiaomi as lead investors at different stages.

Galbot had already reached around $3 billion and has continued raising heavily. Its commercial story also includes thousands of industrial robot orders involving companies such as CATL, Bosch, Toyota, BAIC and SAIC, plus deployments in retail and logistics.

AgiBot sits higher again. It has started the process for a Hong Kong IPO, with previous reporting pointing to prospective valuations around $5 billion to $6.4 billion. The company generated RMB1.05 billion of revenue in 2025 and has scaled production into the thousands of robots.

Dexmal is therefore asking investors to place it in an existing group rather than create a completely new valuation category. The harder comparison is operational: some members of that group have already shown clearer shipment, order or organic revenue growth than Dexmal.

Private-market peers around Dexmal’s valuation tier

Q6Is Dexmal actually growing fast enough to deserve that valuation?

We cannot currently prove that Dexmal’s core embodied-AI business is growing fast enough to deserve a $3 billion valuation, and that is the biggest hole in the bull case.

There is plenty of visible activity. Dexmal has moved from DM0 to DM0.5, launched DexOS, MaaS and Apex, merged with Atomix, built Ferrata for warehouse deployments and kept raising capital. Product velocity is obvious.

Financial velocity is much harder to establish. Dexmal does not publish year-over-year revenue growth, AI-software ARR, Apex shipments, paying developers, net retention or revenue from MaaS. Those are exactly the figures that would tell us whether the young AI business is compounding anywhere near the speed of the valuation.

The contrast with AgiBot is particularly useful. AgiBot says revenue went from roughly RMB60 million in 2024 to RMB1.05 billion in 2025. Unitree reported 2025 revenue growth of more than 300% in its IPO disclosures and was already profitable. Those numbers give investors something tangible to extrapolate.

Dexmal currently asks investors to extrapolate from a different set of facts: fast product releases, deep technical talent, access to industrial environments and an established logistics operation.

That can work in private markets. It simply leaves us with lower confidence than we would have if the company disclosed a year of clean organic growth from its new products.

Q7Why are investors so interested in Dexmal’s warehouse business?

Atomix gives Dexmal something unusually valuable in embodied AI today: places where robots can work every day, fail in messy ways and generate data while solving tasks customers already pay for.

Atomix had completed more than 500 logistics projects before joining Dexmal. Its customers have included Uniqlo, CATL and Mixue, and its operations extend across more than 20 countries.

More recently, Dexmal said its Ferrata system had been tested inside a major customer warehouse covering more than 100,000 SKUs. Ferrata mixes conventional automation with embodied-AI robots and human intervention. Easy tasks stay with cheaper machines, harder objects move to the AI system, and unusual failures can be handed to people rather than stopping the whole operation.

That setup is more interesting than a standalone robot demo. It creates a practical place for immature models to operate while engineers collect exactly the cases that the model still handles badly.

As seen above, Atomix also supplies the bulk of the group’s current commercial base. Investors therefore get two things from the same asset: existing customers today and a possible source of training data for the next generation of models.

Q8Can warehouse data really give Dexmal an AI moat?

Dexmal has a credible path to a data advantage, although we would wait for large-scale deployment before calling it a moat.

Tang Wenbin has repeatedly argued that useful robotics data comes from robots doing real work, especially when they fail or require human intervention. Dexmal’s target is unusually concrete: get 1,000 robots operating continuously for 1,000 hours in a logistics setting.

If it reaches that level, the company would collect data from roughly one million robot-hours of operation. The valuable part will come from diversity inside those hours. Repeating the same easy movement a million times teaches less than encountering strange packages, obstructed cameras, awkward grips, unusual objects, damaged goods and recovery attempts.

Dexmal has designed Ferrata around exactly those long-tail cases. A failed task can be retried, safely backed out or handed to a human, while the episode is retained for later training.

Recent capital flows suggest other investors now see robotics data as a bottleneck too. Chinese embodied-AI data infrastructure companies have raised large rounds this year, including businesses dedicated specifically to physical-AI data collection and evaluation.

Dexmal’s advantage is that it can potentially collect such data inside commercial operations rather than paying to manufacture all of it in dedicated training centers.

The proof will come when we see whether models trained through this loop improve faster across different tasks and machines. Until then, the data story deserves a premium, but a measured one.

Q9Are Dexmal’s AI models actually ahead right now?

Dexmal currently has strong benchmark results and an unusually fast release cadence, but there is still too little independent real-world evidence to call its models the clear leader in physical AI.

The latest DM0.5 release reports a 99.0% average success rate on LIBERO and 93.5% on RoboTwin 2.0. Dexmal has published model weights, training code, fine-tuning workflows and evaluation tools through OpenDM, which makes the work easier for outsiders to inspect and use.

The company has kept shipping after DM0.5. Its open-source work now includes DW05, a world model designed to predict future video, robot actions and value estimates. Dexmal has also released additional checkpoints and fine-tuning support for smaller robot platforms.

That pace is impressive for a company founded so recently.

Benchmarks still have limits. LIBERO and RoboTwin tell us something about model quality under defined evaluation conditions. A warehouse operating ten hours a day with thousands of different items asks a harder question: how often does the system fail, what happens after failure, how expensive is human intervention and does the customer still save money?

Dexmal itself seems to understand that distinction. Tang Wenbin has argued publicly that leaderboard position matters less than the speed at which the system improves once it meets real tasks.

That is the right distinction. DM0.5 is good enough to take Dexmal seriously. The valuation needs the next step: sustained production performance.

Q10Can Dexmal ever earn software-like margins from robotics?

Dexmal has built products that could improve its margins over time, but today we have too little financial disclosure to assume the company deserves software economics.

A large part of the existing business involves warehouses, robots, integration and physical deployments. Those activities can become profitable businesses, but they usually carry more hardware, installation and support costs than pure software.

Dexmal is clearly trying to move upward in the stack. DexOS is designed to separate models from robot hardware. MaaS offers model inference and deployment services with usage-based pricing. DexDev gives developers a common toolkit. If customers begin paying repeatedly for those layers across many robot types, revenue can scale without Dexmal manufacturing every machine itself.

For context, UBTECH reached a 37.7% gross margin in 2025 as higher-margin humanoid products and services became a larger part of its mix. That gives us an idea of what a successful robotics company can achieve while remaining far below classic SaaS gross margins.

Dexmal’s current valuation becomes easier to defend if software, models and recurring services start contributing a meaningful share of revenue. If most sales remain project-based warehouse automation, 20x sales becomes much harder to sustain.

This is one of the numbers we would most want from the company next: gross margin by business line.

Q11Is China’s robotics boom helping Dexmal or inflating its valuation?

Both forces are operating at once: China is building a huge embodied-AI industry, and capital is currently running ahead of proven commercial demand.

The growth story is real. A recent Chinese industry report estimated the country’s embodied-intelligence market at roughly RMB1.09 trillion this year after compounding around 22% to 23% annually since 2018. Manufacturers, technology companies and local governments are all spending heavily on automation and robotics.

Private valuations have climbed just as quickly. X Square crossed RMB20 billion after four completed rounds. Several other Chinese embodied-AI companies have entered the same valuation tier. Funding is concentrating around perceived leaders.

Public markets have lately become even hotter. Unitree’s IPO attracted extraordinary demand and its shares multiplied on their first day, leaving the company worth tens of billions of dollars despite annual revenue measured in hundreds of millions.

There is a weaker side to the boom. Financial Times reporting has found that government-backed training centers are important buyers of humanoid robots and sometimes purchase machines partly to create training data that flows back into the ecosystem. That helps the industry develop, but it can make headline shipment numbers look more commercial than the underlying demand really is.

Dexmal is better insulated from that problem than many early humanoid companies because warehouses already spend money on automation for clear economic reasons. Still, its fundraising valuation is being negotiated during one of the most generous periods Chinese robotics has ever seen.

We should price some heat into the $3 billion figure.

Q12How much revenue would Dexmal need to grow into a $3B valuation?

Dexmal would need roughly RMB1.3 billion of revenue at a 15x multiple or RMB2 billion at 10x, so the valuation does not require science-fiction growth if the combined business can keep expanding.

At a 20x sales multiple, the current revenue estimate already gets us close to the target valuation. The problem is that maintaining 20x becomes much easier when revenue is growing rapidly and increasingly comes from high-margin AI products.

Suppose the market cools. At 15x sales, Dexmal needs roughly one-third more revenue than the current reported base. At 10x, it needs to double. Neither threshold looks impossible for a robotics business in a rapidly expanding market.

Getting there with better revenue mix is more important than simply getting there. RMB2 billion generated mainly from custom hardware projects would deserve a very different multiple from RMB2 billion containing meaningful recurring model, operating-system and service revenue.

Revenue needed to support a ~RMB20B valuation

Q13What has to happen for Dexmal to actually be worth $3B?

Dexmal can grow comfortably into $3 billion if its warehouse deployments turn into repeatable AI revenue rather than remaining mainly an interesting technology layer on top of an established automation business.

The first test is continuous deployment. Customers need to keep meaningful fleets running for months, with intervention rates low enough to produce an attractive payback period. Tang Wenbin has said customers quickly ask how many years the system takes to repay its cost and that projects needing more than five years are generally unattractive.

The second test is revenue mix. DexOS, MaaS, Apex and related AI products need to start showing up materially in the accounts. We would care much more about RMB500 million of new recurring or high-margin embodied-AI revenue than about another RMB500 million obtained mainly from integration projects.

The third is learning speed. Real warehouse failures should make the next model measurably better. If Dexmal can reuse those improvements across different customers and robot bodies, every new deployment becomes more valuable than the one before it.

There are also clear ways the valuation can go wrong. Models could commoditize as Nvidia, Chinese AI labs and other robotics startups converge on similar capabilities. Customer ROI could remain weak once installation and human fallback costs are included. Warehouse data could turn out to generalize poorly beyond logistics. And private robotics multiples could simply fall from today’s unusually high levels.

Deployments, AI revenue and gross margin now matter far more than another funding announcement.

Q14So, is Dexmal really worth $3B today?

Dexmal looks worth something close to $3 billion in today’s private robotics market, but the company has not yet produced enough standalone financial evidence for us to call that valuation comfortably justified.

The relative valuation case is stronger than it first appears. A completed RMB20 billion valuation already exists at X Square Robot. Galbot has raised around the same level. Listed robotics companies such as UBTECH and Dobot trade around low-20s sales multiples, while the Unitree debut has shown just how far current market enthusiasm can stretch.

Dexmal also has a more substantial commercial base than the typical young AI lab. Atomix gives it paying customers, operating environments and a route to collecting real failure data. DM0.5, DW05, DexOS, MaaS and Ferrata show that the company is building aggressively around that base.

The missing proof is financial. We still cannot see clean organic revenue growth from Dexmal’s new embodied-AI products, their margins or how much customers are paying repeatedly for the software layer. Some close peers already disclose much stronger shipment or revenue growth.

Our judgment: $3 billion is aggressive, but plausible.

If the current round closes near $3 billion, we would view the price as defensible under today’s robotics benchmarks, with investors paying early for Dexmal’s warehouse-data and AI-platform strategy. If the AI products fail to become a meaningful source of growing, higher-margin revenue, that same valuation will look stretched very quickly.

Right now, the market case for $3 billion is stronger than the fundamental case. Dexmal’s next stage has to reverse that.