The Chinese hard-tech sector found itself basking in a wave of celebration midway through 2026. Figures from the Chinese Institute of Electronics showed that domestic humanoid robot shipments had surpassed 40,000 units in the first half of the year, capturing a stunning 97% of the global market share. Dozens of leading embodied intelligence firms, clutching order intentions worth billions of yuan, lined up to knock on the doors of the secondary market for their IPOs.

From robotic hands creasing clothes with precision in showrooms to steel frames flipping mid-air at launch events, the message to the world was unmistakable: Chinese robotics was triggering a cyber-era "industrial revolution" with unstoppable force. Yet this upbeat atmosphere was shattered by a social media exposé that struck like thunder.

A public post from Shao Tianlan, a Tsinghua University graduate and founder of the well-known robotics company Mech-Mind, shredded the industry's polished facade. In his statement, Shao revealed that the much-touted prosperity in the humanoid robotics sector was, in large part, a fabricated loop of fake orders, subsidies, and inflated data. Behind the billions in contracts sat a suspicious cycle where companies established shell entities, bought their own machines, and padded their own revenue figures, a form of tech alchemy. How did these so-called hard-tech stars end up ensnared in a vortex where refusing to fake data means being crushed by competition, yet embracing it locks them into a spiral of deeper trouble?

Mastering the Shell Game: The Financial Alchemy of Self-Dealing

One might imagine robot manufacturing as workers tightening screws on a production line or machines hauling battery packs inside Tesla plants. For some embodied intelligence companies, however, actual "manufacturing" resembles a highly precise accounting dance. Based on Shao's allegations and industry realities, this self-dealing pipeline, aptly termed "produce-and-sell-to-self," operates with unnerving logic. First, a parent robot company sets up or co-founds a "shell firm" in regions offering generous subsidies for embodied intelligence, often disguised as an AI micro-research institute or innovation application center. This shell then seeks tens of millions of yuan in technology application subsidies from local governments, claiming it will build intelligent computing hubs to support policy goals.

Once the subsidy is secured, the shell firm makes a grand move, purchasing dozens or even hundreds of humanoid robots from the parent company at highly inflated prices. Step one is complete: the parent's hardware sales revenue spikes instantly, and its financial reports look stellar. But the narrative doesn't end there, robots can't just sit in a warehouse, right? The parent company then rolls out the trendiest buzzword of the day: "Embodied AI depends heavily on multimodal visual-tactile data from the physical world." In a reverse move, it channels the funds back to the shell firm under the guise of "procurement of high-quality data services." The shell places the robots in its own office building, setting up an automated "data collection pipeline" where the machines endlessly grasp plastic cups, sweep floors, and pace back and forth.

The parent then converts the outgoing money into compliant "R&D expenses," which not only boosts its tech credentials but also unlocks additional tax breaks. It's akin to a Michelin-starred bakery using "food innovation subsidies" to open a shell shop next door, which buys up every loaf at a premium, then sells "chewing data reports" back to the bakery at a high price. On paper, thousands of loaves are sold (high revenue) and hefty R&D sums are spent (high tech). But in reality, no hungry passerby ever tasted a single piece of bread.

The Absurd Reality: 3.5 Billion in Orders and a "15-Second Miracle"

This paper-thin boom of self-purchase has produced the most absurd statistical contrast in the 2026 hard-tech world. Publicly disclosed framework agreements and planned procurement amounts have collectively exceeded 3.5 billion yuan, yet IDC delivery reports deliver a stinging rebuke: the number of robots truly making it to factories for real industrial deployment barely reaches a few thousand units. Firms achieving genuine profitability from core humanoid robotics operations are virtually nonexistent.

Inside those "fully automated, AI-driven" cyber test sites, engineers are wearing what might be called "pain masks" daily. To film flawless PR videos for investors and local officials, test engineers clock in 16-hour days. The physical world is brutally unforgiving: a 10-centimeter shift in indoor lighting or a single speck of dust on the floor can send a multi-million-dollar machine tumbling to the ground. These high-caliber AI engineers are reduced to well-paid "cyber babysitters," whose main tasks involve sweating over fallen robots and picking them back up. The 15-second clip of "perfect grasping, smooth walking" that goes viral on social media often masks 499 prior crashes that smashed sensors and frantic off-camera efforts to steady the machines.

What's more bizarre: approach those robots wowing crowds at exhibitions with folding clothes or driving screws for visiting officials, and you might find, just a few meters behind a curtain, an engineer in a motion-capture suit and VR headset remotely puppeteering the bot in real time. The financial spectacle is even stranger. One leading humanoid robotics firm burns hundreds of millions of yuan annually in R&D while failing to sell a single machine; yet its most profitable division is finance. Why? The finance team takes billions in cash from a Series B round and parks it in conservative bank wealth-management products, with yearly interest covering nearly half the R&D staff's salaries. A hard-tech company has effectively morphed into a "financial-product-driven enterprise."

Echoes of History: An Industry's Cyber Pains

Zoom out and place these doubts within the broader arc of industrial evolution, and the current self-dealing dilemma starts to look like another rhyme in history's poem of bubble-popping in hard tech. Look back at China's new-energy vehicle sector around 2014-2016, which endured a strikingly similar "subsidy fraud frenzy." Generous government EV subsidies spawned a wave of startups. To milk the system and inflate revenues, firms manufactured crude, low-range "golf-cart-grade" EVs and set up their own ride-hailing or rental companies to "sell" vehicles to themselves. Streets were dotted with rusting, subsidy-cheating cars, a grand spectacle of self-dealing.

But when the subsidy tide receded and capital regained its senses, those relying on financial alchemy went bankrupt. In the aftermath, giants like BYD and NIO emerged as globally competitive champions, along with other automakers. Today's embodied intelligence sector is reliving that script. From 2021 to 2023, the "mechanical frenzy," the competition was all about hardware: who had better reducers, who could run, jump, and backflip, that won them hundreds of millions in funding. From 2024 to 2025, the "big model possession" phase, the embodied intelligence (VLA) concept exploded as firms boasted brains, claiming robots could understand speech and drive screws. Now, in 2026, the "financial report patch-up" phase, capital's patience is wearing thin and IPO windows loom. When technological progress stumbles against quarterly reporting demands, the industry's slide toward exploiting data-pipeline loopholes becomes almost inevitable.

Capital markets are accustomed to the fast cadence of SaaS and internet software, scaling in three years and going public in five, yet they overlook the brutal grind of marrying hard tech with the physical world. When investors whip too aggressively and technology hits a physical wall, companies see little choice but to manufacture miracles in their spreadsheets. Shao's exposé lands like an ice cube on the feverish embodied intelligence track: it's cold and stark, yet sobering. The leaders sitting on billions in liquid wealth-management funds keep churning through cash in endless loops of prototyping, bid-rigging, and self-dealing. Everyone knows it's a poison they're drinking to curb thirst, but with an IPO sword dangling overhead, who dares to halt this cyber statue game first? As the false heat fades, a shakeout is coming. When will those robots, stuck in office buildings "buying from themselves" to gather data, truly walk out of the shell companies, step into factories, and tighten that first real bolt?