Nvidia CEO Jensen Huang this week unveiled a humanoid robot blueprint that pairs Chinese hardware with American silicon: a 6-foot, 150-pound H2 Plus body from Unitree, an Nvidia Thor T5000 chip, a dexterous hand from Singapore's Sharpa, and a new software stack to train the machine. The package is aimed at researchers and US academic labs that want to assemble a frontier humanoid without building one from scratch. Unitree already sells a base G1 humanoid for around $15,000, versus rival robots that can run several hundreds of thousands of dollars apiece — a price gap that has made Chinese hardware the path of least resistance for Western labs.
The split of labor is straightforward. The Thor T5000 runs the AI models that let the robot perceive its environment and plan motion, while Unitree contributes the motors, actuators, sensors, and chassis. Sharpa's hand handles the part of robotics that has stubbornly refused to scale — fingertip dexterity good enough to peel an apple or run a card trick. Nvidia is packaging all of it under one reference design so a lab can wire its own models into a humanoid in weeks rather than years.
Nvidia's pitch is that it wants to be the brain inside as many humanoids as possible, not just Unitree's. Spencer Huang, Nvidia's director of product for robotics and Jensen Huang's son, told Wired the company sees the same Thor-plus-software stack going into other Chinese makers, including conventional industrial arms that could become more capable with a modern AI controller on board.
Key facts
- 01Nvidia's H2 Plus humanoid blueprint pairs a 6-foot, 150-pound Unitree body with a Thor T5000 chip and a Sharpa-built dexterous hand.
- 02Unitree's base G1 humanoid sells for roughly $15,000, versus competitors that can run several hundreds of thousands of dollars per unit.
- 03Jensen Huang frames humanoids as a multitrillion-dollar economic opportunity for physical AI across the world's largest industries.
- 04The US government loosened chip export restrictions to China late last year, allowing Nvidia to sell more advanced silicon there.
- 05Ghost Robotics CEO Gavin Kenneally warned the US risks ceding the commercial humanoid market to Chinese makers without a national robotics strategy.
That positioning matters because humanoids are turning into the next platform war, and Nvidia is trying to repeat its data center playbook: own the compute layer, stay neutral on who builds the body. If the H2 Plus blueprint catches on, every humanoid shipped — regardless of brand — pulls through a Thor chip and Nvidia's training tools.
The geopolitics is the awkward part. Robotics has become a front line in US-China technology competition, and some US politicians have floated banning Chinese humanoids outright. Last year, security researchers said Unitree robots could capture and transmit data, raising concerns that have not gone away. Nvidia says the H2 Plus blueprint ships with security features designed to reassure users that their models and data stay protected, an implicit acknowledgment of where the political pressure sits.
Scott Singer, a fellow at the Carnegie Endowment for International Peace who studies AI governance and China, called the team-up a fascinating development and noted the underlying logic: the US has the best AI chips, China has the supply chain edge in robot hardware, and the H2 Plus blueprint is what happens when both sides decide cooperation beats decoupling on this particular product.
“Both sides have key parts of the supply chain that they might be able to weaponize, but here they are working together.”— Scott Singer, Fellow, Carnegie Endowment for International Peace
The export-control backdrop has shifted in Nvidia's favor. The US government still bars Nvidia from selling its most capable chips to China, but late last year it loosened restrictions to allow more advanced chips into the country. That gave Nvidia room to court Chinese robot makers as customers rather than write them off.
Not everyone in the US robotics industry is comfortable with the arrangement. Gavin Kenneally, cofounder and CEO of Ghost Robotics, which builds legged robots for defense and security customers, said Unitree's technology has drawn extensively on innovations from Western labs and warned that the US is on track to repeat the drone-market loss to DJI if it does not respond with policy.
Unitree, for its part, has already built genuine mindshare. Its bots show up in social-media clips doing parkour and kung fu, and they appear in published research from many Western labs precisely because they are programmable and cheap. The G1's $15,000 entry price is the most disruptive number in the category — it is the difference between a humanoid being a single-unit research curiosity and being a fleet purchase.
Jensen Huang framed the broader prize in his usual terms, saying humanoid robots will bring physical AI to the world's largest industries and open a multitrillion-dollar economic opportunity. Whether that number ever materializes, the addressable market story is what justifies Nvidia treating robotics as a first-class platform alongside data center GPUs and automotive.
The H2 Plus blueprint is a market-share bet dressed as a research tool. By making the cheapest credible humanoid body even more capable — and by stamping every unit with a Thor chip — Nvidia raises the floor for what a lab-grade humanoid looks like and pulls the rest of the industry toward its software stack. The US policy question Kenneally raises is real, but it does not change the near-term incentive: any company trying to build a humanoid business in 2026 now has a faster, cheaper reference design to clone, and Nvidia gets paid either way.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
