FAIR plus 2026: Has Embodied AI Really "Made It"?

Last Friday (April 22nd), I went to the Shenzhen Convention and Exhibition Center to accompany a friend at the FAIR plus 2026 robotics exhibition.

Honestly, I didn’t have high expectations before going—after all, I’ve seen too many videos of “stunning robot debuts” these past two years, only to discover they could only take two steps in a lab afterward.

But this time, something felt different.

First, the Scale

15,000 square meters, over 500 upstream and downstream companies, more than 70 concurrent events. The organizers say they’re expecting 2,500 overseas professional visitors.

This isn’t about “displaying the future”—it’s about “displaying the present.”

The most intuitive feeling in the exhibition hall: No humanoid robots dancing, no one showing off difficult movements.

Everyone was talking about: How many units shipped? What’s the mass production cost? Which factories have already validated the technology?

From “Technology Race” to “Mass Production Battle”

The theme of this year’s conference was “A New Era of Embodied Intelligence Productivity.”

In plain terms: The technology exists, now everyone’s competing over who can sell it first.

Here’s an interesting data point: Zhiyuan Robotics shipped 5,100 units throughout 2025, compared to just 1,000 in early 2024. That’s 5x growth in one year.

UBTECH delivered 1,079 full-size humanoid robots for the full year.

These numbers are nothing special in the consumer electronics industry, but in the humanoid robotics industry—these are real batch deliveries, not PPT demonstrations.

What’s the Real Key to “Mass Production Year” Success?

I chatted with several robotics engineers at the exhibition, and the general consensus was: Before 2026, the core problem for embodied intelligence was “can we make this work?” Starting in 2026, the core problem became “can we make this work consistently at a controllable cost?”

There’s an interesting contradiction here:

Customers want “stability,” “reliability,” and “low cost”—but these three words are fundamentally at odds with current embodied intelligence technology.

Not stable enough, not reliable enough, costs still high.

But at this exhibition, I saw some approaches to solving this problem. For example, one company showcased their “spatial memory module”—enabling robots to accurately perceive and efficiently operate in complex environments. Their founding team came from HKU and landed in Qianhai.

Simply put: installing a “hippocampus” in robots.

This direction is very pragmatic. Not pursuing universal robots, but solving a specific problem in a specific scenario.

My Observations

This wave of embodied intelligence is somewhat similar to the drone wave of 2015-2016.

Back then, everyone was also asking: Can drones really transform from toys into tools? Can costs come down? Where are the application scenarios?

The later answer we all know—not drones themselves won, but industry application scenarios won. Plant protection, inspection, aerial photography…

Embodied intelligence might follow the same logic. Not replacing humans in any scenario, but forming real productivity in specific scenarios and specific tasks.

One direction I’m paying close attention to is industrial flexible production lines. Tesla’s Optimus is reportedly already “interning” at a certain factory. If this path validates, the market size for embodied intelligence will be on an entirely different scale.

Of course, I might be overly optimistic.