cd /news/artificial-intelligence/in-the-embodied-ai-race-china-can-op… · home topics artificial-intelligence article
[ARTICLE · art-105313] src=scmp.com ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

In the embodied AI race, China can opt to look beyond bigger models

China is pioneering a more cost-efficient pathway in the embodied AI race, which could lower innovation barriers for developing countries, according to Yu Chao, founder and CEO of Lumos Robotics. The global pool of compliant, high-quality physical interaction data is only about 500,000 hours by early 2026, far short of the tens of millions of hours needed for general-purpose embodied AI models. The US pursues a scale-driven approach with larger models and massive infrastructure, while Europe and Japan focus on regulation and hardware, but China's strategy may offer an alternative.

read2 min views9 publishedAug 21, 2026
In the embodied AI race, China can opt to look beyond bigger models
Image: Scmp (auto-discovered)

Advertisement

Opinion

China is pioneering a more cost-efficient pathway which could inspire tech communities in the developing world by lowering the barriers to innovation

3-MIN READ3-MIN Yu Chao is the Founder and CEO of Lumos Robotics, an embodied intelligence company dedicated to bringing intelligence into industry.

The world has seen tremendous progress in the capabilities of robots in the past year. Robots can now walk, dance, run and perform

increasingly sophisticated movements, yet impressive demonstrations do not equal commercial value.Beyond the hype surrounding China’s red-hot embodied

robotics sector, one challenge confronts all players: data.Robotics has long faced the challenge of achieving

widespread adoption. To do so, robots must be for general purpose, reliable and economically viable – all at the same time. So far, no company has fully solved all three facets of this problem.The biggest bottleneck is access to

high-quality physical interaction data, the missing fuel required to power the next generation of intelligent robots. Building general-purposeembodied artificial intelligence(AI) models could require tens of millions of hours of real-world interaction data. However, by early 2026, the globally available pool of compliant, high-quality physical interaction data was only around 500,000 hours.The world is searching for solutions, but different regions are pursuing their own strategies. In the United States, tech giants follow a scale-driven approach of larger models, more compute and massive AI infrastructure. This strategy naturally favours companies with

enormous capital reservesand computational resources. Europe and Japan focusmore on regulation, safety and hardware advantages but have yet to establish leadership in AI algorithms and large-scale data ecosystems.Meanwhile, competition in advanced technology is expanding beyond chips, AI and communications into robotics.

Recent US restrictionson Chinese-made humanoid robots highlight how hardware access is increasingly becoming intertwined with geopolitical competition.Advertisement

Select Voice

Select Speed

1.00x

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @lumos robotics 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/in-the-embodied-ai-r…] indexed:0 read:2min 2026-08-21 ·