On June 1, Nvidia unveiled its first commercial humanoid robot — built on a Chinese body and powered by an American brain — and Beijing approved the robot maker’s IPO the very next day. The sequence wasn’t subtle. It signaled how physical AI has become the next front in the U.S.–China technology race, and why Nvidia and Tesla are now pursuing two radically different paths to dominate it.

Jensen Huang took the stage in Taipei ahead of Computex to announce that the Isaac GR00T Reference Humanoid — a nearly six‑foot machine built around Nvidia’s Blackwell GPU — would use a robot body from China’s Unitree Robotics and hands from Singapore‑based Sharpa. Researchers at Stanford, ETH Zurich, and UC San Diego will be among the first to work with it. The following day, Unitree received IPO approval from the Shanghai Stock Exchange in a record 73 days, a pace China’s state media framed as a signal of strategic intent rather than routine capital markets activity.

China Daily called the Nvidia-Unitree deal “a compelling example of how the respective industries of China and the US can leverage their unique strengths.”

The unspoken message: a major win for China in what it sees as one of the defining technology competitions of the coming decades.

What Physical AI Really Means

American technologists call it physical AI. Chinese policymakers call it embodied intelligence (具身智能, jùshēn zhìnéng). The terminology differs, but the ambition is identical: giving artificial intelligence a body so it can perceive, reason about and act in the physical world.

The concept appeared in China’s Government Work Report for the first time in 2025 and is a centerpiece of the 15th Five-Year Plan. At GTC Taipei , Huang said “today, agentic and useful AI has arrived” — and that the Isaac GR00T platform, the Jetson Thor onboard chip and the Omniverse simulation environment form the infrastructure stack that will carry it into the physical world.

Digital intelligence running inside a data center, however powerful, remains constrained to the digital world. Physical AI that can operate in factories, hospitals, warehouses and homes is a labor force. And labor is the one input that has constrained economic growth and the power of empires since the beginning of recorded history.

The demographic mathematics are unforgiving. South Korea’s fertility rate rebounded slightly to 0.80 in 2025 but remains the world’s lowest — putting it on track to halve in population over the next 60 years. The Bank of Korea projects the country’s economy could begin to contract as early as 2041. Japan, Germany and China face versions of the same cliff, just at slightly different angles. No immigration policy credibly solves a problem of that scale. Humanoid robots, if the technology delivers, could soften the erosion of GDP and living standards.

Huang estimates the total addressable market for physical AI at $40 trillion . That number is a vibe as much as an estimate. As I noted in my read of the SpaceX S-1, large numbers are easier to type than to earn. But the directional point is serious. This technology plays a critical role at the intersection of labor economics, supply chain control and national security, rivaling that of generative AI.

Huang’s Strategy: Become The Arms Dealer

Nvidia’s approach to physical AI mirrors the strategy that made it the world’s most valuable semiconductor company: build the platform, let others build on top of it and collect the compute and software tax on the entire ecosystem.

The Unitree partnership reflects that philosophy. Nvidia did not build a robot. It chose the best robot body available, regardless of where manufacturers built it, installed its brain and sold the resulting system to the world’s leading research institutions. The arms dealer doesn’t need to win the war. He needs to be indispensable to every army in it.

The choice of a Chinese body wasn’t incidental. Eight of every 10 humanoid robots produced worldwide come from China. Unitree’s revenue grew 335% year over year in 2025 . The company is, by a wide margin, the most commercially proven humanoid manufacturer on the planet.

Nvidia has also signaled that Unitree isn’t its only hardware partner. The company pursues similar collaborations with robot makers in the United States, Europe and South Korea, though it hasn’t named them. It also built Blackwell’s security architecture — secure boot, confidential computing and software verification — directly into the Unitree platform, addressing at least some congressional concerns about Chinese-manufactured robots in federally funded labs. The message: we are the neutral infrastructure layer, and we are making it trustworthy.

This is a delicate position to hold. Some U.S. lawmakers have called for banning Unitree products from federally funded research entirely, citing alleged ties to Chinese government programs. Security researchers found serious vulnerabilities in Unitree products in 2025, including a wormable Bluetooth flaw .

Nvidia’s arguments on security are debatable, but its commercial logic is clear: the arms dealer who refuses to sell to the largest army in the market isn’t neutral — he’s irrelevant.

Musk’s Strategy: Build The Whole Stack

Elon Musk’s approach is the philosophical inverse. Tesla has described itself as “a physical AI company” in its SEC filings, and the Optimus robot embodies that claim — literally. Where Huang partners horizontally across the globe, Musk seeks to build vertically within one organization.

Tesla’s unique asset is its AI stack. The FSD neural network that has now runs unsupervised robotaxi service in Dallas and Houston is the same end-to-end perception and control architecture Tesla trains into Optimus. No other humanoid manufacturer has anything comparable — a proven, real-world autonomous intelligence system running at scale.

The challenge is the hardware economics. A McKinsey analysis found that building Optimus without Chinese suppliers would cost roughly three times as much — a bill of materials rising from roughly $46,000 to $131,000.

The Gen 3 Optimus is designed for mass production, with a one-million-unit-per-year line planned at Fremont , but Tesla acknowledged in Q1 2026 earnings that Optimus was not yet in use “in a material way” in its own factories. The distance between Musk’s production ambitions and the current reality is a gap investors must evaluate carefully.

China vs. The United States: Who Leads Where

The simplest way to frame the competitive landscape: China dominates the body; America leads on the brain.

China’s EV supply chain — motors, power electronics, battery systems, precision actuators — maps almost directly onto humanoid robot components. Chinese manufacturers have cut their bill-of-materials costs about 40% year over year. The country installed 295,000 industrial robots in 2024 — more than the rest of the world combined — creating the factory floor training data humanoid developers depend on. And Beijing has committed a $138 billion state venture capital fund to AI and robotics, alongside national policy frameworks that make embodied intelligence a strategic priority through 2030.

America’s advantage is in the intelligence layer. U.S. foundation models, simulation environments and reinforcement learning research remain unmatched. Tight integration between AI platforms and robotics — Tesla-FSD, Nvidia-Isaac, Boston Dynamics with Google DeepMind — gives American robots a reasoning and generalization capability Chinese hardware hasn’t yet replicated.

The question is whether the chip constraint materially limits China’s physical AI trajectory or simply slows it. The Brookings Institution testified to Congress in April 2026 that China’s full-stack approach to physical AI represents a strategic challenge comparable to its dominance of solar panels and EVs.

This is a race between two genuinely capable industrial systems pursuing different strategies. China is deploying massive numbers of units into factories, spinning up the data flywheel and iterating at volume. The United States bets that the intelligence layer is the durable moat, and that an AI-first robot will ultimately outperform a hardware-first one.

Beneath all of this runs a corporate restructuring story that could reframe the entire competitive picture. SpaceX is set to begin trading on Nasdaq as “SPCX” on June 12 , targeting a $1.75 trillion valuation in what may be the largest IPO in history. As I wrote in my deep dive into the S-1 , the filing rewards careful reading. Buried in the risk factors is a sentence that has set Wall Street buzzing: SpaceX “may issue a significant amount of equity in connection with future transactions.”

CNBC has reported that Elon has spoken with colleagues about potentially combining Tesla and SpaceX into a single company. Wedbush’s Dan Ives puts the odds at 80 to 90 percent by early 2027 . The crowd on Polymarket is more skeptical — currently pricing the merger at roughly 43% by year-end .

The strategic logic, viewed through a physical AI lens, is compelling. A combined SpaceX-Tesla would house xAI’s Grok models, Tesla’s FSD stack and Dojo supercomputer, the Optimus robot program, Starlink’s global communications infrastructure, and Terafab — the chip manufacturing joint venture with Tesla and Intel being developed in East Texas — under a single roof. That would be the most vertically integrated physical AI company ever assembled: one entity controlling the intelligence, the robot body, the chip, and the network.

If Elon can assemble the empire he envisions, it is a genuine answer to Jensen’s ecosystem strategy. A single organism that makes the platform irrelevant for at least one very large customer. If the merger stalls, the production timelines slip again, or Terafab proves harder to execute than to announce — then Jensen’s arms dealer model wins by default.

It is far too early to call a winner in the race to dominate physical AI. It’s quite possible we end up with a segmented market, not unlike the Apple-Android dichotomy that defined the mobile device world for decades — the arms dealer and the emperor each carving out dominant positions in different segments.

What I will say is this: the societies that need this technology to work are not abstractions. South Korea, Japan, Germany, China, and eventually the United States all face versions of the same arithmetic — too few workers, too many retirees, too little time to solve it through conventional means. A humanoid robot that can perform factory, logistics, or care work reliably is not a luxury product. It is, potentially, the economic infrastructure of the next half-century.

The robots are coming. The only real question is whose name will be on the box.