In early June 2026, Unitree Robotics announced a partnership with NVIDIA to produce the H2 Plus, a humanoid robot aimed at research institutions. In Shen Yi's close dissection, this seemingly routine piece of business news exposes a deeper structural question: when a Chinese AI company carrying the hopes of a "pride of domestic industry" chooses to bind itself tightly to a chipmaker already weaponized by the U.S. government, is that a sound commercial choice — or a strategic gamble that exposes the company to its adversary?

The Structural Tension Behind the Partnership

In a ten-point analysis, Shen Yi laid bare the paradox at the heart of the partnership. NVIDIA is an American company; Jensen Huang is an American — in the event of a serious conflict between China and the United States, they would side with the U.S. government without hesitation. In February 2026, the U.S. Department of Defense had already placed Unitree Robotics on its "Chinese Military Companies List," and bills brewing in Congress would impose further sanctions. The implication is stark: the more successful Unitree becomes, the fiercer U.S. containment and pressure will grow.

Against that backdrop, partnering with NVIDIA is not a matter of finding a technology partner — it is exposing one's soft underbelly to the U.S. government, handing over the grip and the lever for a future stranglehold.

Two Diverging Paths

Shen Yi set Unitree's choice in sharp contrast with DeepSeek's: DeepSeek has been backing China's domestic AI chip ecosystem — Huawei included — with real actions; even where homegrown solutions temporarily lag in performance, it has chosen to grind through the kinks and build alongside them. Unitree, on the eve of its IPO, chose to tie itself to NVIDIA — at least in appearance, a step in the opposite direction.

In the long run, even if importing NVIDIA's less-cutting-edge chips solves part of the compute need in the short term, this approach is neither secure nor beneficial to the development of China's domestic AI-chip industry and ecosystem. Especially since the Trump administration has already signaled its intention to take a "cut" from NVIDIA's chip exports to China, continued dependence is no longer merely a business decision — it touches on national security and national dignity.

Questions About H2 Plus's Positioning

Shen Yi pointed out that Unitree disclosed the H2 Plus is positioned as a product for research institutions. But this positioning itself rests on a risky assumption: does serving research institutions mean limited damage if the United States severs the partnership? If the answer is no, then this partnership has placed the company squarely over a fire.

📝 The Layered Risks of Partnering with NVIDIA

Supply-chain risk: NVIDIA could be barred by the U.S. government from supplying Unitree at any time

Intellectual-property risk: whether the partnership gives the U.S. side access to Unitree's technical details

Market-access risk: binding itself to NVIDIA could close doors into parts of the Chinese market

Signaling risk: it feeds the narrative that "Chinese AI firms still depend on American technology"

From a Single Case to a Structure

The deeper significance of this news reaches beyond Unitree alone. It exposes a pervasive dilemma facing China's AI industry during its technology catch-up phase: while the domestic chip ecosystem is not yet fully mature, should firms opt to "get running on foreign silicon first"? At the individual level, the choice is rational — use the best chips to build the best products. But at the collective level, such choices delay the maturation of the homegrown ecosystem and deepen dependence.

There is no easy answer. But Shen Yi's analytical framework offers a useful reminder: technology self-reliance is not a destination — it is an ongoing process that every firm must practice through its own choices. As Huawei's experience attests, had it "assumed back then that the U.S. government could be more ruthless, and pushed its spare-tire contingency plan further," the hardships it later faced would have been smaller.

⚡ Structural Insight

This is a classic collective-action dilemma: for each firm, choosing "the best imported chips" is rational at the individual level — but if every firm chooses that way, the domestic chip ecosystem never receives enough usage feedback or iteration momentum. Technology sovereignty is not achieved by waiting for conditions to ripen; it is pushed forward collectively by the firms willing to make the choice before conditions are fully ripe.

The Escalating Congressional Hunt: From Sanctions Lists to Dedicated Legislation

Within hours of the official announcement, the U.S. Congress launched the corresponding blocking procedures at unusual speed. Two streams of legislative pressure converged on Unitree at once.

One came from the Senate: Cotton and Schumer jointly introduced the American Security Robotics Act, whose core provision bans the U.S. government from purchasing or using unmanned ground systems from "adversary nations" such as China, while also prohibiting the use of federal funds for such purposes. The stated rationale is blunt — "such robots could collect data or be remotely controlled."

The other came from the House: Moolenaar, Obernolte, and McClellan jointly introduced the GUARD Act, which requires a review of humanoid and quadruped robots made by "foreign adversaries" such as China; products deemed to pose national-security risks would be placed on the FCC Covered List, barring them from the U.S. market.

The two bills aim at different chokepoints — the first targets government procurement and federal funding, the second the commercial channel of market access. The combined effect, however, is clear: once both tracks become law, Unitree's U.S. business will face a full-spectrum blockade from public markets to commercial markets. Judging by the precedents of Huawei, TikTok, and DJI, the eventual passage of these bills is largely a matter of time.

“ Shen Yi's Judgment

"These moves show that the U.S. Congress has placed Chinese robots into a new round of technology controls — after Huawei, TikTok, DJI, and AI chips."

At the same time, there is another layer to the senators' concerns — NVIDIA officially announced that it would bring data-center-grade security mechanisms such as secure boot and confidential computing from its Blackwell chips into robotics. That means software updates on Unitree's robots must pass authenticity verification through NVIDIA chips. A Chinese company already on the military-companies list, handing cybersecurity verification over to an American firm — the arrangement itself closes the risk loop.

NVIDIA's "Riding the Wave" Strategy and an Asymmetric Bond

Shen Yi's follow-up analysis further exposes the inherent asymmetry of this partnership. The Isaac GR00T humanoid-robot reference design that NVIDIA officially unveiled is, in essence, a bundle — the Unitree H2 Plus body, the Sharpa five-finger dexterous hand, Jetson Thor onboard compute, and the Isaac GR00T software stack — packaged as an open platform for universities and research institutions. NVIDIA simultaneously emphasized that it would seek similar partners in the United States, Europe, and South Korea — it never intended a deep bond with Unitree, and keeps itself ready to let go at any time.

In other words, for NVIDIA this is a classic "stepping-stone" strategy: use the buzz around Unitree's hardware to cut into the embodied-AI research track and keep its valuation story alive in the capital markets. For Unitree, tying itself to NVIDIA means exposing critical interfaces — including software updates and cybersecurity certification — to an adversary.

That said, the partnership is not without short-term gains. Jetson Thor is built on the Blackwell GPU, with 128 GB of unified memory, up to 2,070 FP4 TFLOPS of AI performance, and a 40-to-130-watt power envelope. Edge-inference chips of this kind carry far less security sensitivity than data-center chips such as the B200 or GB200, and are compliant products under the U.S. export-control framework. Unitree thus gains a decent-quality edge-computing platform — and an entry point into the international embodied-AI research community.

STAR Market IPO Approval: A Parallel Acceleration in Capital Markets

On the very day the NVIDIA partnership was announced, Unitree's IPO on the STAR Market cleared its approval hearing — from acceptance by the Shanghai Stock Exchange to approval by the listing committee, just 73 days, the fastest on record for A-shares this year. The same day, "Chinese robotics firms race to list" became a headline across multiple outlets.

The timeline itself is a signal. Unitree's leading position in the humanoid-robot race (more than 5,500 pure humanoid robots shipped in 2025, first in the world, with a gross margin of 68.44%) and the Chinese capital markets' enthusiasm for the embodied-AI concept together form the foundation of its valuation. But the dual-track legislation hunting it down in Congress, the replaceability of the NVIDIA partnership, and the potential loss of overseas markets should China and the United States decouple — these constitute the greatest uncertainty in the valuation narrative.