Jensen Huang ran two negotiations at once through 2026134. Washington wanted a cut of every China sale. Gulf sovereign-wealth funds wanted gigawatt AI campuses finished before their neighbors finished theirs. Both talks moved fast. Then Beijing told its own companies to stop buying the chips Washington had just cleared for sale. Tens of thousands of GPUs went to Riyadh and Abu Dhabi instead of Shenzhen.
A Rule Rescinded, a Revenue Share Invented
The Trump administration rescinded the Biden-era AI diffusion rule in May 20251. That rule had sorted every country on earth into three tiers of chip access. A replacement arrived the following August, and it was a deal rather than a rule. Nvidia and AMD agreed to pay the U.S. government 15 percent of revenue from certain China sales as the price of an export license. Nvidia's share attached to the H20, a chip built to sit just under the limits Washington had set. Export control became a fee Washington collected on every sale that cleared it.
Nvidia and AMD agreed to pay Washington 15 percent of revenue from certain China chip sales
Share of revenue from certain China chip sales paid to the U.S. government for an export license, in percent
The full circle is all revenue from those licensed China sales, and the magenta arc is the 15 percent share the U.S. government collects.
Source: Wikipedia [1]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value | Note |
|---|---|---|
| Paid to the U.S. government | 15% | Agreed in August 2025; Nvidia's share attached to the H20 |
| Whole | 100% |
Revenue sharing solved one problem and made another. Chinese buyers watched a foreign government take a fixed cut of each purchase1. They had reason to shop at home. Beijing's regulators gave them a push.
China Steps Back
H20 production reportedly stalled in August 2025, after Chinese government directives discouraged domestic purchases1. September brought fresh limits, aimed this time at Nvidia's RTX Pro 6000D line. Count the moves. Two license approvals from Washington inside one summer, and two discouragements from the buying side, left Nvidia holding permission it had few customers to use.
Nvidia's own numbers show the standoff held into late 2026. Fiscal second-quarter results, covering the period ended July 26, 2026, logged revenue of $96.2 billion, up 106 percent year over year2. Data Center revenue climbed 117 percent to $89.0 billion. The H20 story sits in the reconciliation tables. A $4.5 billion inventory charge landed in the year-ago quarter. It vanished the next quarter, then came back as a $180 million partial release. Chips written off, a fraction recovered.
Nvidia recovered $180 million of a $4.5 billion H20 inventory charge while China stayed shut
Nvidia's H20 inventory charge and its later partial release, in billions of dollars
The blue bar is the H20 inventory charge Nvidia took in the year-ago quarter, and the magenta bar is the partial release it booked in the quarter ended July 26, 2026.
Source: Nvidia Newsroom [2]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value | Note |
|---|---|---|
| H20 charge, year-ago quarter | $4.50B | Landed in the year-ago quarter and vanished the next quarter |
| Partial release, latest quarter | $0.18B | A $180 million partial reversal of the earlier charge |
Nvidia made the position plain in its own guidance. "NVIDIA is not assuming any Data Center compute revenue from China in its outlook."2 That is a forecast built on a closed market, whatever license Washington grants.
The Gulf Buys What China Declined
Saudi Arabia moved first. The kingdom's Public Investment Fund founded Humain in May 2025 under Mohammed bin Salman, and named Tareq Amin chief executive that same month4. Nvidia announced it would ship 18,000 AI chips to the kingdom through the new venture. Humain added partnerships with AMD and Qualcomm the same year. Hardware was one part. The company also built ALLaM, an Arabic-first multimodal model trained on more than 500 billion Arabic tokens, and has stated plans for several gigawatts of data-center capacity.
The UAE answered with scale. G42, led by Group CEO Peng Xiao and chaired by Tahnoun bin Zayed Al Nahyan, announced Stargate UAE on May 22, 20253. Partners included OpenAI, Oracle, SoftBank Group, Cisco and Nvidia, and the target was Abu Dhabi compute campuses running in 2026. Washington cleared the hardware that November, authorizing export of semiconductors equal to roughly 35,000 Nvidia Blackwell chips for G42's use. July 2026 went further. Commerce moved the UAE into Country Group A:5 and named G42 an authorized receiving entity, easing limits that had held up earlier shipments.
Board Seats as Diplomacy
Corporate governance did quiet work alongside the formal licensing. G42's board includes Microsoft President Brad Smith3. That seat gives one of Washington's most experienced technology-policy figures a direct view of Abu Dhabi's AI buildout, and of the compliance questions every Gulf chip shipment raises. Mubadala CEO Khaldoon Khalifa Al Mubarak and Silver Lake co-CEO Egon Durban hold seats as well. Sovereign capital, hyperscaler diplomacy and private equity, on one board, assembled to withstand regulators on three continents.
Read the composition as insurance. A Gulf AI venture holding a Microsoft executive, a major American private-equity partner and Emirati sovereign leadership carries political cover a purely regional buyer would struggle to assemble. That cover likely helped Commerce approve the November 2025 authorization, and the further easing that followed in July 2026.
A Record Quarter, a Closed Market
Huang's company posted the largest quarterly revenue in its history during the same stretch China stayed shut2. He framed the moment in one line. "AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue." Four short sentences, and the last carries the claim. Chips sold are revenue booked, wherever the buyer sits.
Huang then turned to demand. "And demand is accelerating. This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem and physical AI coming online — with strong momentum across the U.S. and around the world. The AI infrastructure buildout is at full steam. Vera Rubin, now in full production, was built to power exactly this moment."2 Count the customers he names. Labs, startups, open-model builders, physical AI, the U.S. and the wider world. China appears in that list zero times.
The buildout scaled past the point where one national market decides the outcome. Saudi and Emirati orders, counted in tens of thousands of chips, now stand in for a China channel closed by distrust on both sides of the transaction34.
Whether the substitution lasts depends on three things outside Nvidia's reach. Beijing could shift its domestic-chip strategy1. Gulf capital could move elsewhere34. A fresh Washington administration could revisit the 15 percent arrangement its predecessor invented. Huang kept building toward whichever market stayed open longest.
The Fee Defended, the Fee Challenged
Commerce Secretary Howard Lutnick made Washington's case for the arrangement in a CNBC interview, explaining why the administration kept a controlled Nvidia chip flowing to China at all. "You want to sell the Chinese enough that their developers get addicted to the American technology stack," Lutnick said6. Read the verb he reached for. Addicted describes a market Washington meant to keep, fee included, more than one it meant to cede to Huawei.
Raja Krishnamoorthi, the U.S. Representative serving as Ranking Member of the House Select Committee on the Strategic Competition Between the United States and the Chinese Communist Party, wrote directly to President Trump four days after the deal became public. "By requiring companies like NVIDIA and AMD to pay a percentage of their sales revenue to obtain export licenses, your Administration is flagrantly violating this clear legal prohibition," Krishnamoorthi wrote, pointing to the Export Control Reform Act's ban on license fees5. The letter grounds its objection in statute, the line Krishnamoorthi says Washington crossed for a fee Congress barred by name.
By the numbers
- Fifteen percent of certain China chip revenue: the cut Nvidia and AMD agreed to pay the U.S. government for export licenses, August 20251.
- Total revenue hit $96.2 billion in the quarter ended July 26, 2026, up 106 percent year over year2.
- Data Center revenue reached $89.0 billion that same quarter, up 117 percent year over year2.
- Roughly 35,000 Blackwell-equivalent chips gained export authorization to G42 in November 20253.
- Eighteen thousand AI chips make up Nvidia's committed shipment to Saudi Arabia's Humain, announced May 20254.
- A $4.5 billion H20 inventory charge from a year earlier saw partial reversal through a $180 million release this quarter2.
What to watch
Nvidia's next earnings call answers one question first. Does China Data Center revenue stay out of guidance a second straight quarter? That is the clearest read available on whether Beijing's pullback holds. G42's Country Group A:5 reclassification took effect in July 2026, and the deliveries against that 35,000-unit authorization deserve a separate check, since a license and a shipment are different events3. Humain's gigawatt ambitions give a concrete milestone to track through 2027, because the gap between announced capacity and running capacity widens fastest in sovereign-scale buildouts4. Watch, too, whether other Gulf or Southeast Asian buyers negotiate board seats for Western executives the way G42 secured Brad Smith.
Sources
- Wikipedia contributors, "Nvidia," Wikipedia, retrieved Sept. 4, 2026, https://en.wikipedia.org/wiki/Nvidia
- Nvidia Corporation, "NVIDIA Announces Financial Results for Second Quarter Fiscal 2027," Nvidia Newsroom, Aug. 27, 2026, https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-second-quarter-fiscal-2027
- Wikipedia contributors, "G42 (company)," Wikipedia, retrieved Sept. 4, 2026, https://en.wikipedia.org/wiki/G42_(company)
- Wikipedia contributors, "Humain," Wikipedia, retrieved Sept. 4, 2026, https://en.wikipedia.org/wiki/Humain
- Raja Krishnamoorthi, "Letter to President Trump on NVIDIA and AMD Export Control Revenue-Sharing," U.S. House Select Committee on the CCP (Democrats), Aug. 15, 2025, https://democrats-selectcommitteeontheccp.house.gov/sites/evo-subsites/democrats-selectcommitteeontheccp.house.gov/files/evo-media-document/2025.08.15-krishnamoorthi-letter-to-president-trump-export-controls_0.pdf
- Mariella Moon, "China Reportedly Discouraged Purchase of Nvidia AI Chips Due to 'Insulting' Lutnick Statements," Engadget, Aug. 21, 2025, https://www.engadget.com/ai/china-reportedly-discouraged-purchase-of-nvidia-ai-chips-due-to-insulting-lutnick-statements-123055120.html