NVIDIA Wins. America Loses

On the eve of the Trump-Xi summit in Beijing, the United States quietly cleared roughly 10 Chinese firms, including Alibaba, Tencent, ByteDance, and JD.com, to buy NVIDIA’s H200 accelerators, with Lenovo and Foxconn approved as distributors. NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) shareholders, sitting on a $5.33 trillion market cap and a stock up 19% in the past month, got exactly what Jensen Huang flew to Beijing to secure. America got something worse than a bad trade. It got a strategic concession dressed up as a commercial win.

The Visible Win: Huang Beats the Hawks

Huang was initially left off the summit invite list. Trump overrode his China-hawk advisors at the last minute to bring him along. That reversal matters more than the guest list suggests. Polymarket traders now assign a 73% implied probability that Trump announces AI export restrictions relief for China by May 22, with the AI-relief market the most actively traded contract on the summit board.

NVIDIA just closed fiscal 2026 with $215.94 billion in revenue, up 65%, and Data Center revenue of $62.31 billion in Q4 alone, up 75% year over year, while guiding Q1 FY2027 to roughly $78 billion in revenue explicitly excluding any China Data Center compute. Reopening that channel layers a multi-billion-dollar tailwind onto a business already running at 75% non-GAAP gross margins. Wall Street’s 57 Buy ratings against one Sell tell you how the Street reads it.

The Hidden Cost: Selling the Lead We Spent a Decade Building

Treasury Secretary Scott Bessent’s framing is that the U.S. can afford to discuss AI with China “because we are in the lead.” That sentence describes a moat. Approving H200 sales to Alibaba, Tencent, and ByteDance drains it.

H200 is the workhorse silicon behind frontier training and inference clusters. Huang himself describes Grace Blackwell as “the king of inference today, delivering an order-of-magnitude lower cost per token,” and H200 sits one rung below that on the same ladder. Selling it to the Chinese hyperscalers most capable of standing up domestic frontier models hands Beijing the one input it cannot manufacture at scale: leading-edge compute. The export-control regime that produced the $4.5 billion H20 charge in Q1 FY2026 and an estimated $8 billion in lost Q2 revenue was painful precisely because it was working.

Two caveats keep this from being settled. No deliveries have occurred yet, and Beijing’s own scrutiny is stalling deals. Reports also suggest a U.S./China deal could include China easing pressure on its firms to avoid NVIDIA purchases. That is a managed transaction in which both governments decide how fast the American compute lead erodes.

Who Pays

NVIDIA shareholders pocket the windfall. American workers, the national security establishment, and any future administration trying to hold an AI sovereignty line absorb the cost. The asymmetry is the point: a one-quarter revenue bump for one company against a generational shift in which country trains the next frontier model on whose silicon.

Watch three signals over the next two quarters. First, whether actual H200 shipments to the named Chinese buyers clear customs, or whether Beijing’s stalling tactics hold. Second, NVIDIA’s Q1 FY2027 earnings on May 20, where any reintroduction of China Data Center revenue into guidance confirms the policy reversal is real. Third, whether the Commerce Department publishes a formal licensing framework or keeps the approvals case by case. The first locks in NVIDIA’s win. The third decides whether America’s loss is recoverable.