China is funding an awful lot of robot brains that haven't done very much yet.
That is the tension beneath the headlines from EAIDC 2026, the world's first embodied AI developer conference, which X Square Robot staged in Shenzhen last week. X Square is a 27-month-old startup backed by Alibaba, ByteDance, and Meituan — three of China's largest internet platforms, rarely found on the same cap table — and it used the occasion to announce that its WALL-A architecture, which integrates vision-language-action models with predictive world models, is the "robot brain" it is selling to the market. The conference drew 20 university teams including Tsinghua and Peking to complete real-robot tasks: ring placement, cable plugging, fruit sorting by voice command. X Square called it the industry's first real-robot deployment competition. The company has raised approximately $280 million to date, according to PR Newswire.
What the company did not say, and what makes the conference worth sitting in on, is that Beijing is starting to worry about exactly this. The National Development and Reform Commission, China's top economic planner, has publicly flagged structural risks from the crowded field of humanoid robot listings, according to analysts who track the regulator's public statements. The message, as it was read by the market: we have seen this movie before. The EV sector spent 2024 and 2025 burning through capital on factories producing vehicles that customers were not buying fast enough to absorb them. The NDRC does not want a repeat.
The financing data makes the warning harder to dismiss. Since the start of 2026, China has recorded 189 equity financing events in domestic embodied intelligence, averaging 2.6 per day, according to 36Kr. February and March both exceeded 10 billion yuan in total volume. That pace has not slowed since the NDRC's concerns became public. It has not even hiccuped.
AGIBOT, a Shanghai-based humanoid robot company, chose the same week as EAIDC to announce it had shipped its 10,000th robot, doubling its fleet from 5,000 units in roughly three months, Gizmochina reported. The milestone is real. Ten thousand robots operating in the world is a meaningful number in a sector where most companies are counting in dozens or hundreds. But the milestone also landed in the same news cycle as a conference where the hardest task was plugging a cable into a socket, which suggests something about what "deployment" still means in practice.
Wang Qian, X Square's founder and CEO, has a PhD in Robotics Learning from the University of Southern California and was one of the earliest researchers to introduce the attention mechanism in neural networks in the Chinese academic community, 36Kr reported. His pitch is that the embodied AI competition is "essentially a battle of foundation models built on data closed loops and their capacity for model evolution," The Robot Report noted — language that tracks directly with how every foundation model company in the world describes its own work. WALL-A is positioned as the robot equivalent of the large language models that reshaped software between 2022 and 2025. The comparison is apt in one direction: just as LLMs required enormous compute and data to reach capability thresholds, robot brains appear to need something similar. The gap is that a language model that hallucinates is an inconvenience. A robot arm that misjudges a grab is a different class of problem.
X Square is not vaporware. The company is generating early revenue from deployments in education, hospitality, and elder care, and is running a pilot household cleaning service in Shenzhen through a partnership with 58.com, the classifieds platform that operates in more than 200 cities and serves over 45 million families with a network of more than 4 million domestic workers, PR Newswire reported. The 58.com pitch for why it chose X Square is worth quoting at length: "Households are the most challenging environments for robots. Unpredictable and full of edge cases. If a robot can master the living room, it can handle almost any physical space," The AI Insider reported. That framing is honest. It is also a description of a test environment, not a product.
ManipArena, the benchmark X Square launched at CVPR 2026, covers 20 real-world robotic manipulation tasks with over 10,000 teleoperation trajectories and a 56-dimensional multimodal dataset, VIR.com reported. It runs on X Square's proprietary Quanta XI dual-arm robotic platform and is accessible via cloud-based remote testing. The benchmark is a legitimate contribution to the field. It is also advertising for the platform.
What the NDRC warning actually signals, reading across the regulatory tea leaves, is that Beijing sees the embodied AI sector through the same lens it applies to EVs: a strategic priority that is also generating the conditions for a painful correction. The EV parallel is imperfect — robots are not cars, and the end markets are different — but the structural pattern is the one that caught regulators' attention: rapid capital concentration, a crowded field of similar products, and no clear answer to the question of what happens when the financing runs out before the deployment problem is solved.
X Square Robot has Alibaba, ByteDance, and Meituan on its cap table. That is a meaningful signal of commercial confidence. It is also a signal that three of China's most aggressive internet platforms believe the robot brain race will produce something worth owning — and that they are willing to find out whether that belief survives contact with a Shenzhen apartment and a toddler and a cooking pot.
The conference was real. The funding is real. The deployment gap is real too, and Beijing is watching it close.
— Reporting from Shenzhen