AgiBot Shipped More Robots Than Tesla, Figure, and Apptronik Combined. You Have Probably Never Heard of Them.

There is a company in Shanghai that has shipped more humanoid robots than Tesla, Figure AI, and Apptronik combined. Not marginally more. Not by a slim margin that analysts could argue away. Nearly seven times more.

AgiBot has shipped approximately 5,200 humanoid robot units as of early 2026. Tesla has shipped 500. Figure AI has shipped 200. Apptronik has shipped 50. Add the three most-discussed American humanoid programs together and you get 750 units. AgiBot, alone, delivered almost seven times that number.

You have probably never heard of them.

That gap is not an accident. It is the product of a specific set of decisions made by a specific person, executed within a specific industrial ecosystem, and almost completely ignored by the English-language technology press. Understanding how it happened matters far more than the number itself.

The man behind the machines

The AgiBot story starts with Peng Zhihui, and Peng Zhihui is not who Western tech media would cast as the protagonist of a humanoid robot revolution.

Before founding AgiBot, Peng was a hardware engineer at Huawei. That alone would be unremarkable. China has hundreds of thousands of hardware engineers at Huawei. What made Peng different was what he did in his spare time. Starting around 2020, he began posting videos on Bilibili, China’s equivalent of YouTube, showing himself building increasingly complex robotic systems from scratch at home. A self-balancing bicycle. A robotic arm that could pour coffee. An Iron Man-style helmet with a functioning heads-up display.

The videos went viral. Peng accumulated over 10 million followers on Bilibili and became one of the most recognized engineers in China. His appeal was rooted in something specific: he was not a talking head explaining concepts. He was a practitioner showing his work. Every video featured real hardware, real code, real problems solved in real time. In a country where tech celebrities tend to be founders and executives, Peng became famous for being an engineer who could actually build things.

This matters because it shaped what AgiBot would become. Peng did not come from the startup ecosystem. He did not start with a pitch deck and a vision for disruption. He came from Huawei’s hardware culture, where execution and manufacturing discipline are valued above all else. When he left Huawei in 2023 to found AgiBot (officially Shanghai Zhiyuan Juren Technology), he brought that engineering-first mindset with him.

From founding to factory in under a year

AgiBot was founded in 2023. Within its first year, the company raised over $140 million from a group of investors that tells you everything about the company’s strategic positioning: BYD, SAIC Motor, Sequoia China, and the Shanghai AI Industry Investment Fund.

Notice who is on that list. BYD and SAIC are not passive financial investors. They are two of the largest automotive manufacturers in the world. BYD alone sold over 3.6 million vehicles in 2024. SAIC is the parent company of MG and a major partner with both Volkswagen and General Motors in China. These companies invested in AgiBot because they wanted humanoid robots on their factory floors. The investment came with built-in customers.

The funding timeline was aggressive, but what Peng did with the money was more aggressive still. Rather than spending years on research and development before thinking about manufacturing, AgiBot built its factory first. The company established a vertically integrated production facility in Shanghai’s Lingang Special Area, the same industrial zone that houses Tesla’s Gigafactory Shanghai. The factory was designed from the outset for mass production of humanoid robots, not for prototyping.

This was a fundamentally different approach from what American competitors were doing. Figure AI, which raised $675 million in its Series B alone, spent its first two years perfecting its Figure 01 and 02 prototypes before establishing its BotQ manufacturing facility. Tesla announced Optimus in 2021 but only began mass production from converted Model S/X lines at Fremont in January 2026. Apptronik, despite being founded in 2016, was still running pilot programs.

Peng built the factory before the product was perfect because he understood something that Silicon Valley’s venture-backed approach tends to obscure: in hardware, you learn more from manufacturing at scale than from prototyping in the lab. The assembly line is the teacher. Every unit that rolls off the line generates data about tolerances, failure modes, and cost reduction opportunities that no amount of simulation can replicate.

What AgiBot actually builds

AgiBot produces two primary humanoid robot models: the G1 and the X2. More recently, the company launched the A2 series in Standard, Max, and Ultra variants, powered by a proprietary multimodal AI system called WorkGPT running on NVIDIA Jetson Orin hardware.

The A2 Ultra, the top-end model, stands 170 cm tall, weighs 75 kg, carries up to 20 kg, and can operate for approximately 4 hours on a single charge. Its 42 degrees of freedom allow it to perform complex manipulation tasks in manufacturing and logistics environments. The unit price sits around $100,000 for the A2 Ultra, with the Standard variant priced significantly lower for simpler applications.

These are not the most technically sophisticated humanoid robots in the world. Figure AI’s 02, for example, likely has superior AI-driven task generalization. Tesla’s Optimus Gen 3, with its FSD-derived neural architecture, probably handles novel situations better. But AgiBot’s robots are good enough for the specific, repeatable tasks that dominate early humanoid deployment: moving parts along a manufacturing line, performing quality inspections, handling logistics in warehouses, and assisting in hospitality settings.

“Good enough” is not a criticism. It is a manufacturing strategy. AgiBot ships robots that can do the jobs customers need done today, then uses the revenue and field data from those deployments to improve the next generation. The perfect is the enemy of the shipped.

The BYD and SAIC partnership machine

The automotive partnerships deserve their own section because they explain something that Western observers consistently underestimate about the Chinese industrial model: the speed at which partnerships translate into deployed units.

When BYD invested in AgiBot, it was not writing a speculative check on the future of robotics. BYD operates some of the largest and most complex manufacturing facilities in the world. The company was actively looking for ways to automate tasks that were difficult for traditional industrial robots but too tedious or dangerous for human workers. Humanoid robots, with their ability to navigate factory floors designed for humans and manipulate objects in human-scale environments, were a natural fit.

Within months of the investment, AgiBot units were operating on BYD production lines. Not as demonstrations. Not as pilot programs with carefully controlled conditions. As working production equipment with performance metrics and uptime requirements. The same pattern repeated at SAIC Motor, where AgiBot robots began handling component logistics and quality inspection tasks.

This partnership model has a compounding effect. Each deployment provides AgiBot with real-world operational data. That data feeds back into product improvements. Better products attract more partners. More partners mean more deployments. More deployments mean more data. The cycle accelerates.

Compare this to Figure AI’s relationship with BMW. Figure’s 02 completed an 11-month trial at BMW’s Spartanburg plant, where it assisted in the production of over 30,000 BMW X3s and moved 90,000 components across 1,250 operating hours. That is genuine progress. But an 11-month trial is a very different thing from a full production deployment at scale, and Figure shipped approximately 200 total units during the entire period.

Apptronik’s partnership with Mercedes-Benz and Jabil remains in the pilot phase. Tesla’s Optimus units are deployed almost exclusively within Tesla’s own factories, which limits the diversity of operational data the company collects.

AgiBot, by contrast, has deployed robots across at least eight distinct commercial verticals: automotive manufacturing, electronics assembly, warehouse logistics, hospitality, retail service, healthcare support, education, and industrial inspection. Each vertical generates different types of operational data, different failure modes, and different performance requirements. The breadth of deployment is itself a competitive advantage.

The open-source chess move

In a move that surprised Western AI researchers, AgiBot released the AgiBot World open dataset for embodied AI research. The dataset includes real-world robotic manipulation data collected from AgiBot’s deployed fleet, covering grasping, navigation, object recognition, and task planning scenarios.

This was not charity. It was strategy.

By releasing high-quality embodied AI training data, AgiBot accomplished several things simultaneously. It positioned the company as a serious contributor to the global AI research community. It attracted talent from universities and research labs who wanted to work with real-world robotic data. It created an ecosystem of researchers and developers whose work ultimately improves AgiBot’s own products. And it established a standard that competing datasets would be measured against.

The open-source strategy also maps directly onto the Chinese government’s broader approach to AI development. Beijing wants Chinese companies to lead in foundational AI research, not just in applications. By sharing data publicly, AgiBot strengthens China’s overall position in embodied AI while losing very little competitive advantage. The data is valuable, but the ability to continuously generate new data from an ever-growing fleet is far more valuable. That ability cannot be open-sourced.

The Western media blind spot

Here is where the story becomes uncomfortable.

Search for “humanoid robot” in any major English-language technology publication from the past two years. Count the articles about Tesla Optimus. Count the articles about Figure AI. Count the articles about Apptronik, about 1X, about Sanctuary AI. Now count the articles about AgiBot.

The disparity is enormous. Tesla Optimus generates more media coverage from a single demo video than AgiBot generates from shipping thousands of units to paying customers. Figure AI’s funding rounds receive more attention than AgiBot’s actual manufacturing output. The company that is winning the humanoid robot shipment race by the widest margin is almost invisible in Western technology media.

There are several reasons for this, and none of them are good.

Language barrier. AgiBot’s primary communications are in Mandarin. Peng Zhihui’s massive Bilibili following is entirely Chinese-language. The company’s press releases, technical documentation, and media appearances are overwhelmingly in Chinese. Western journalists who do not read Mandarin are structurally disadvantaged in covering the company.

Access. AgiBot’s factory is in Shanghai. Its partners are Chinese companies. Its deployments are primarily in Chinese facilities. Western journalists cannot easily visit, inspect, or verify claims. This creates a trust gap that feeds into broader skepticism about Chinese technology companies.

Narrative preference. Western tech media has a strong preference for stories about charismatic founders who raise billions of dollars, partner with household-name companies, and promise to change the world. Elon Musk, Brett Adcock, and the various founders of American robotics startups fit this template perfectly. Peng Zhihui, despite being one of the most famous engineers in China, does not have the same profile in English-language media.

Geopolitical framing. Coverage of Chinese technology companies is increasingly filtered through a geopolitical lens. Articles about Chinese robotics tend to focus on national security implications, intellectual property concerns, or government subsidies rather than on the companies’ actual products and capabilities. This framing makes it difficult for readers to form an accurate picture of what Chinese companies are actually achieving.

The result is a distorted information environment in which the English-speaking world has detailed knowledge of humanoid robots that barely exist in production quantities and almost no knowledge of the humanoid robots that are actually being deployed at scale.

The numbers in context

It is worth stepping back and examining what the shipment numbers actually mean in competitive terms.

AgiBot’s 5,200 units make it the second-highest shipping humanoid robot company in the world, behind only Unitree at 5,500. But the comparison is misleading in one important respect: Unitree’s numbers are dominated by the G1, a smaller, less capable, and far cheaper platform priced around $16,000. AgiBot’s units are primarily full-scale humanoid robots deployed in industrial settings at significantly higher price points.

In terms of revenue generated from humanoid robot sales, AgiBot is almost certainly the global leader. If we assume an average selling price of $50,000 across all variants (some units are the cheaper G1 model, some are the $100,000 A2 Ultra), AgiBot’s 5,200 shipped units represent roughly $260 million in humanoid robot revenue. Tesla’s 500 units at an estimated $100,000-$150,000 per unit might represent $50-75 million. Figure AI’s 200 units at a similar price point might be $20-30 million.

These are rough estimates, and AgiBot does not publicly disclose its financials. But the order of magnitude is clear. AgiBot is not just shipping more units. It is generating more revenue from humanoid robots than any company in the world.

The vertical integration advantage

AgiBot’s manufacturing approach deserves detailed examination because it explains how the company achieved costs and production rates that Western competitors have not matched.

Vertical integration, in AgiBot’s case, means the company controls the production of most critical components in-house or through tightly integrated Chinese supply chain partners. Actuators, the motors and gearboxes that power a humanoid robot’s joints, are one of the most expensive and technically demanding components. AgiBot sources these from Chinese manufacturers who also supply the electric vehicle industry, benefiting from the massive economies of scale that the EV boom created.

Battery cells come from the same Chinese suppliers that power BYD’s vehicles and CATL’s global battery business. Sensors are sourced from Shenzhen’s electronics manufacturing ecosystem, the densest concentration of sensor and semiconductor production in the world. The NVIDIA Jetson Orin modules that run AgiBot’s WorkGPT AI system are the primary exception, being designed by NVIDIA (an American company) but manufactured by TSMC in Taiwan.

The net effect is that AgiBot can produce a full-scale humanoid robot for a fraction of what it costs Figure AI or Apptronik to produce an equivalent unit. Lower production costs enable lower selling prices, which enable higher sales volumes, which enable further cost reductions through economies of scale. The flywheel spins.

The timeline tells the real story

When you lay out the milestones side by side, the pace of AgiBot’s execution becomes even more striking.

For comparison, consider the American timelines. Figure AI was founded in May 2022, more than a year before AgiBot, and has shipped 200 units. Tesla announced Optimus in August 2021, nearly two years before AgiBot’s founding, and has shipped 500 units. Apptronik was founded in 2016, seven years before AgiBot, and has shipped 50 units.

The execution speed differential is not marginal. It is structural. AgiBot moved faster because it prioritized manufacturing from day one, because its investors were also its customers, because the Chinese supply chain could deliver components at scale, and because the Chinese market was willing to deploy humanoid robots before the technology was fully mature.

What the West gets right (and wrong)

This article is not a victory lap for China or a eulogy for American robotics. The competitive landscape is more nuanced than the shipment numbers alone suggest.

American companies have genuine technological advantages that should not be dismissed. Tesla’s FSD-derived neural architecture for robotics is probably the most sophisticated AI system ever deployed on a humanoid platform. Figure AI’s collaboration with OpenAI and its demonstrated performance at BMW show real progress in AI-driven task generalization. Apptronik’s deep NASA heritage and UT Austin research foundations represent decades of accumulated knowledge in bipedal locomotion.

The problem is not that American companies lack technology. The problem is that technology alone does not win hardware markets. Manufacturing capability, supply chain access, unit economics, and deployment speed matter just as much, often more, in the early stages of a new hardware category. China understood this lesson from its experience in solar panels, batteries, drones, and electric vehicles. The United States appears to be learning it again the hard way.

There are also legitimate questions about AgiBot’s long-term competitive position. Export restrictions on advanced AI chips could constrain the company’s ability to develop next-generation AI systems. Geopolitical tensions may limit AgiBot’s access to Western markets, particularly in the United States and European Union. The company’s dependence on Chinese domestic demand means it is vulnerable to changes in Chinese industrial policy or economic conditions.

But these are future risks. The present reality is that AgiBot is shipping more humanoid robots than anyone except Unitree, generating more humanoid robot revenue than anyone, and accumulating more real-world deployment data than any Western competitor. Dismissing these advantages because of possible future constraints is precisely the kind of complacent thinking that created the blind spot in the first place.

The $39 billion question

Figure AI is valued at $39 billion. AgiBot is valued at approximately $1.4 billion.

Sit with that for a moment.

The company shipping 200 humanoid robots is valued at 28 times more than the company shipping 5,200. Figure AI’s valuation per shipped unit works out to roughly $195 million. AgiBot’s valuation per shipped unit is about $269,000.

This disparity reflects different capital markets, different investor expectations, and genuinely different strategic bets. Figure AI’s investors are betting that the company’s AI capabilities will eventually enable a platform that captures enormous value per unit. AgiBot’s investors are betting on volume, manufacturing efficiency, and market share. Both bets could pay off. But only one has produced thousands of working robots in customers’ hands today.

The valuation gap also reflects the media blind spot discussed earlier. American investors read American media. American media covers American companies. The feedback loop creates a world in which Figure AI can raise $1.85 billion in total funding while AgiBot raises $140 million, despite AgiBot shipping 26 times more units. The information asymmetry is not just a curiosity. It is a pricing inefficiency with real capital allocation consequences.

What happens next

AgiBot’s trajectory points toward a 2026 production target of over 10,000 units, which would make it the first humanoid robot company to achieve five-figure annual production. The company is expanding its Shanghai factory and reportedly exploring additional manufacturing sites.

Tesla, with its massive balance sheet and Gigafactory infrastructure, remains the only Western company with a plausible path to matching AgiBot’s production volumes. Musk has stated targets of 50,000-100,000 Optimus units in 2026, though he acknowledged on the Q4 2025 earnings call that the robots remain “very much in the R&D phase.” Tesla has converted Model S/X production lines at Fremont for Optimus manufacturing and is building a dedicated facility at Gigafactory Texas. If Tesla hits even a fraction of its stated targets, the production race will look very different by year’s end.

Figure AI is scaling its BotQ manufacturing facility toward a target capacity of 12,000 units per year. The company’s $1.85 billion in funding gives it substantial runway. The Figure 03, introduced in October 2025 as a complete hardware and software redesign built for scale production, represents a serious commitment to manufacturing. But Figure has yet to demonstrate production at scale.

Apptronik, with 50 shipped units and ongoing pilot programs with Mercedes-Benz and Jabil, is operating at a fundamentally different scale from the other three companies. The company’s Apollo platform has real technical merits, particularly in modularity and configurability, but the gap in production volume is large and growing.

The lesson

AgiBot’s rise is not a story about one exceptional company. It is a case study in how industrial ecosystems produce outcomes that individual companies, no matter how well funded or technically sophisticated, cannot match on their own.

Peng Zhihui built AgiBot the way BYD built electric cars, the way DJI built drones, the way Huawei built telecom infrastructure. Prioritize manufacturing. Ship imperfect products. Iterate in the field. Let the supply chain do the heavy lifting on cost reduction. Partner with customers who will deploy your product before it is perfect.

This model has won before. It is winning again. The only question is whether the rest of the world will recognize it in time to respond, or whether the media blind spot will persist until the production gap is too large to close.

Five thousand two hundred robots are a lot harder to ignore than a demo video. But somehow, Western media has managed.