WAIC 2026 Public Day: AI Moves from Showy to Scalable

July 18, 2026 — For years, the narrative around humanoid robots was dominated by flashy stunts: backflips, dance routines, and carefully choreographed performances designed to go viral. But at this year's WAIC, the script has been flipped entirely. The robots aren't here to perform — they're here to work.

On the conference's first public day, the question being asked across the expo floor wasn't "what can this robot do?" but "what does this robot actually do?" The answer, increasingly, is: real work, in real settings, for real customers [citation:1][citation:11].

60 Robots Clock In: The 'Deployment Era' Has Begun

On the morning of July 17, before the conference officially opened to the public, 60 humanoid robots had already "clocked in" for their shift [citation:1].

Deployed across the four venues — the Shanghai World Expo Exhibition and Convention Center, Expo Center, Zhangjiang Science Hall, and the West Bund — these robots performed real, unscripted public service tasks: greeting attendees in both Chinese and English, guiding visitors to their destinations, answering questions, and navigating through crowds [citation:5].

"Before, I thought robots were just exhibits. I didn't expect them to be 'colleagues' this year," one international exhibitor was quoted as saying [citation:5].

This marks the first time that full-size humanoid robots have been deployed at scale in a public-facing environment at a major international event — and it signals a shift that the industry has been waiting for [citation:5].

2026 is being called the "year of deployment" for embodied intelligence [citation:1]. At WAIC, that shift was on full display. The robots that showed up this year weren't just walking and dancing; they were working. In pharmaceutical warehouses, robots sorted medicines with 90-second turnaround times. On automotive production lines, robots performed sub-millimeter-precision assembly. In logistics hubs, they sorted packages in live demonstrations [citation:1][citation:10].

The Hands Have It: Why Fingers Matter More Than Feet

Perhaps the most significant shift at WAIC 2026 was the industry's focus on one specific component: the hand [citation:11].

At the conference, dexterous hands — the robot's ability to interact with the physical world — were arguably the hottest category, with companies showcasing increasingly sophisticated capabilities [citation:1][citation:11].

One company demonstrated two robotic hands working together to twist a balloon into a dog shape — a task that requires real-time perception and millisecond-level adjustments in force. Balloons are soft and deform unpredictably; applying too much force pops them, while too little fails to shape them [citation:1].

An industry observer at the conference put it this way: "A robot that can run and jump can perform. A robot that can pick up an egg can work. Running fast won't get it into a factory. Jumping high won't get it into a home. The last mile from the showroom to the workshop is not legs — it's hands" [citation:11].

The data backs this up. The Chinese dexterous-hand market reached approximately 357 million yuan ($49 million) in 2025 — 19,200 units — representing 236.84% year-over-year growth. This year, sales are projected to hit 70,200 units, a further 265.63% increase. Hand joints account for 23% of the total cost of a robot, a clear signal of where the industry is placing its bets [citation:11].

OPC: The One-Person Company Revolution

If the robots represented AI's transformation of physical labor, the OPC (One-Person Company) zone represented AI's transformation of the workforce itself.

WAIC 2026 featured an OPC exhibition area for the first time, showcasing startups where individuals — often young, often in their 20s — are using AI to run entire businesses solo [citation:2][citation:3].

Selected from 1,200 innovation proposals, the exhibition featured 158 startups and 22 OPC projects, with many founders born after 2000 [citation:2].

One exhibitor, Lin Xiuchun, a "post-00s" founder, explained that their company solves a fundamental problem: trust in AI. Their product enables encrypted data to be computed without being decrypted, keeping user information secure while still delivering AI performance [citation:2].

Another team, the youngest humanoid robotics group at WAIC, with their youngest member born in 2007, has already raised nearly $100 million in funding. "We want to open-source humanoid robotics technology so the entire industry has fewer barriers, and every developer can build a robot from scratch and deploy algorithms," said co-founder Lu Junjie, born in 2001 [citation:2].

AI has fundamentally shifted the economics of entrepreneurship. According to the analysis at WAIC, technology startups used to require teams, office space, and significant capital. In the AI era, "one person becoming a company" is possible — with compute power and open-source tools, a single individual can take on projects that once required large teams [citation:2].

A new interpretation of AI was coined at the conference: it's not just artificial intelligence, but "AI as a visible individual" — a tool that makes solo creators visible to investors, customers, and the world [citation:3].

The Brain Behind the Body: From Posture to Perception

While hardware improvements in dexterity grabbed headlines, the real battle is being waged in software and the "brains" that power these machines [citation:1][citation:4].

At WAIC, companies like Ant Lingbo demonstrated a "one-brain-multiple-bodies" approach. Three robots of different brands and configurations worked together under a single "general-purpose brain" to complete a full logistics workflow — receiving orders, picking, and delivering — all without predetermined programming [citation:1].

Their LingBot-VLA 2.0 model has already been deployed in real pharmacies in Shanghai, where a robot assists pharmacists with overnight drug sorting [citation:1].

Tencent also unveiled its full-stack embodied intelligence play, from foundation models to native AI agents to development platforms, signaling that China's tech giants are betting heavily on physical AI [citation:4].

Ant Lingbo's approach represents a fundamental philosophical shift. Rather than taking the dominant industry approach — fine-tuning video generation models for robot control — they built their own model from scratch using a strict causal autoregressive architecture. As one executive put it: "It's like someone who's watched a thousand hours of swimming videos being thrown into the pool. Watching, no matter how much, doesn't replace actually struggling in the water." The model, which can run at 150Hz on consumer-grade GPUs, enables the robot to complete 150 "predict-act" cycles per second — enough to support continuous, fluid closed-loop operation [citation:1].

Fueling the Fire: How Data Drives Physical AI

One of the most significant shifts highlighted at WAIC was the emergence of a new data infrastructure for embodied intelligence [citation:8].

For years, data scarcity was the single biggest bottleneck for humanoid robot development. But a new paradigm is emerging: "human data." Rather than relying solely on expensive, time-consuming robot teleoperation, companies are now equipping humans with lightweight data-collection devices — head-mounted cameras, VR equipment, smart glasses — and having them perform tasks. These recordings are then used to train robots [citation:8].

"Previously, a company might have had only a few hundred hours of human data. Now we're seeing 500,000 hours, and some believe we can reach 10 million hours," one data infrastructure executive noted [citation:8].

Another company, Qiongche Intelligence, demonstrated a "RoboPocket" system — a wearable data-collection solution that allows ordinary workers to capture real-world manipulation data without touching a robot. In a single day, the company can process 9,000 to 10,000 data points, and has accumulated 13,500-15,000 hours of real-world operation data per month [citation:10].

But the field still faces "long-tail" challenges. As one industry observer noted, embodied intelligence is similar to autonomous driving — it's easy to reach 60%, 70%, 80% capability, but the final 20% takes a disproportionate amount of time and effort. The data flywheel hasn't fully started turning yet, and it may take another two to three years before it does [citation:8].

Key Takeaways