When a company builds humanoid robots and its revenue quadruples in a single year, the public markets take notice. Unitree Robotics (known formally as Yushu Technology) filed for a $904 million IPO last week, targeting a $9 billion valuation and becoming China’s first publicly traded humanoid robot maker. Alongside that, DeepSeek announced it is taking a 2.31% strategic stake in Unitree, investing 140.8 million yuan to co-develop AI models for physical robots.

Together these announcements mark a turning point: physical AI is no longer a research project. It is an investable asset class.

What Happened

Unitree has been building quadruped robots (the dog-shaped ones) and humanoid robots since 2016, but 2025 was when the numbers moved. Humanoid robot sales reached 867.8 million yuan in 2025, overtaking quadruped robots as the company’s largest business line. Total revenue hit 1.7 billion yuan, more than four times the prior year.

The IPO, listed on China’s mainland exchange, would give Unitree access to capital to scale production, expand its product lines, and invest in the core technology stack: embodied AI models, reinforcement learning, and the high-performance actuators that let robots move with enough precision to do real work.

The DeepSeek deal adds another dimension. DeepSeek, best known for the open-source AI models that rattled the AI industry earlier this year, is expanding its scope beyond language and reasoning. The partnership commits DeepSeek to developing AI models specifically for physical agents: models that let robots understand their environment, plan sequences of actions, and recover from failure. Unitree brings the hardware expertise. DeepSeek brings the brain.

Why Physical AI Is Accelerating Now

The timing is not accidental. Three things converged in 2025 and 2026 that made physical AI viable at scale:

Better foundation models. The same improvements in reasoning, planning, and multimodal understanding that power software agents also transfer to robotic control. A model that can navigate a complex conversation can, with the right training, navigate a complex physical environment.

Cheaper compute. AI inference costs have dropped dramatically. Running the AI models that control a robot in real time is now economically feasible in a way it was not three years ago.

Better actuators and sensors. Hardware has caught up. The mechanical components that let robots move, grip, and sense their environment have improved in reliability and dropped in cost, following the same price trajectory that personal computers followed in the 1980s.

Unitree’s IPO is a signal that the infrastructure layer of physical AI is mature enough to fund at scale through public markets. That is a meaningful threshold.

What This Means for Business

Most businesses are still working out how to deploy software agents. That is the right priority for 2026. But physical AI has a shorter runway than most business leaders currently assume.

The automation frontier is moving. Agentic software handles information work. Physical AI handles physical work. The combination, a software agent that coordinates with a physical agent in the same workflow, is what makes fully automated operations possible. We are not there yet. But the companies that will deploy this at scale in 2028 and 2029 are the ones making the infrastructure bets now.

DeepSeek’s involvement matters. Open-source AI accelerates physical AI development by giving smaller companies access to the same model quality that was previously only available to well-funded labs. If DeepSeek applies its approach to embodied AI models the way it did to language models (high-quality, open, and cheap to run) — it could compress the timeline for physical AI adoption by years.

Regulated industries will move slower, but should plan now. Healthcare, construction, and logistics are among the highest-potential deployment sectors for humanoid robots. Unitree is specifically targeting high-intensity and high-risk tasks: the kinds of environments that are difficult, dangerous, or physically demanding for humans. For businesses in these sectors, the question is shifting from “will this happen?” to “what do we need in place when it does?”

The workforce conversation is evolving. 2026 has been dominated by debates about agentic AI and white-collar work. Physical AI opens the same conversation for blue-collar and field-based roles. The businesses that handle this transition well will be the ones that have already built cultures of continuous upskilling and that understand their own operations well enough to identify where physical automation adds value without eroding the human expertise that still matters.

The Bigger Picture

Unitree going public is a milestone, not a prediction. Humanoid robots in every warehouse is still a 2028-2030 story, not a 2026 story. But the IPO, the DeepSeek partnership, and the quadrupling of revenue are all data points pointing the same direction: physical AI has crossed from experimentation into commercial deployment, and the pace is accelerating.

For Enterprise DNA’s audience of business owners, data professionals, and AI adopters: the agentic software wave is where to invest your energy right now. But keep one eye on what comes next. The window to understand these technologies before you have to deploy them is closing faster than the headlines suggest.

---

Enterprise DNA helps business leaders navigate AI adoption across strategy, operations, and workforce. Learn more about Omni by Enterprise DNA, AI-powered operations built for the realities of your business.