US Expands AI Chip Licenses: AMD Joins China Trade

The trickle of American AI accelerators into China is widening — and for the first time, it now includes AMD. According to documents reported on July 14, 2026, the U.S. government has granted new export licenses allowing a fresh set of Chinese companies to purchase advanced AI chips, extending a licensing regime that until recently was confined to a handful of the country’s largest cloud players. Shares of AMD rose about 6% on the news, a sign that Wall Street sees the Chinese market reopening as a genuine revenue opportunity rather than a policy footnote.

Who got approved

The new licenses cover two distinct tracks. On the Nvidia side, ZTE’s Kangxun Telecom subsidiary and a company identified as Maginfra were cleared to buy Nvidia’s H200 GPU — the export-approved Hopper-generation part that Washington has permitted for sale into China under carefully reviewed conditions. Those firms join a short list of Chinese giants — including Alibaba, Tencent, and ByteDance — that have secured access to the chip.

The more notable line item is AMD. A subsidiary of cloud-computing firm Kingsoft — reported as Zhuhai Hengqin Yunxiang Zhisheng Network Technology — received approval to purchase AMD data-center accelerators that compete directly with the H200, understood to be Instinct MI300X-class parts. It marks one of the first times AMD has been named alongside Nvidia in a China AI-chip licensing approval, opening a lane the company had largely been shut out of.

The approvals reflect what officials describe as the Trump administration’s evolving approach to AI hardware: permitting exports of certain high-performance chips under case-by-case licensing while keeping the most capable silicon — Nvidia’s newer Blackwell generation and its equivalents — firmly restricted. The H200 and its AMD counterpart are a deliberate middle tier: powerful enough to be commercially meaningful, a generation behind what American hyperscalers now run.

The other half of the standoff

This week’s licenses are the U.S. side of a two-part story. Just days earlier, reporting indicated that Beijing was preparing to relax its own restrictions — the mirror-image barrier in which Chinese regulators had discouraged domestic firms from buying American chips at all, part of a campaign to steer demand toward homegrown accelerators. We covered that reversal in detail in our look at China’s H200 approval for its leading AI firms.

Put together, the two moves are unwinding a peculiar standoff: for much of the past year the H200 was legal to sell from the American side but effectively blocked on the Chinese side. With Washington now issuing licenses to specific buyers and Beijing signaling conditional approval for its national champions, both halves of the blockade are loosening at once — and the chips can finally start to flow.

Emphasis on start. A U.S. trade official cautioned this week that “very few” H200 chips have actually been shipped to China so far, a reminder that a license is permission, not a purchase order. Licensing, logistics, and Beijing’s own conditions still sit between approval and revenue.

Why AMD’s stock moved

For AMD, the market reaction was about optionality. The company has spent the AI cycle as the clear number-two to Nvidia in data-center accelerators, and China represents one of the largest pools of AI compute demand outside the United States — a market it had been almost entirely locked out of. Even a partial reopening expands AMD’s addressable market at the margin, and margins are where the number-two player fights for share. For the competitive backdrop, see our comparison of AMD’s accelerators against Nvidia’s lineup.

The analyst response tracked the optimism. Bank of America raised its price target on AMD to $620 while maintaining a Buy rating, citing strong demand for the company’s server processors alongside the accelerator story. AMD closed July 14 up 6.05% at $566.73, outpacing the broader chip tape on a day when the sector was still trying to find its footing after a rough June.

The chips, and the ceiling

It is worth being precise about what these licenses do and do not permit. The H200 is a Hopper-generation chip — highly capable, but a step behind the Blackwell-class parts that power the newest American data centers. The AMD accelerators cleared for Kingsoft are positioned as rough equivalents. In other words, the U.S. is allowing China to buy last-generation frontier hardware while reserving the current generation for domestic and allied use.

That ceiling is the whole design of the policy. It lets American vendors capture Chinese revenue and keeps China’s AI developers partially dependent on the U.S. ecosystem, while preserving a compute advantage at the true frontier. It also creates a pressure valve for Chinese firms that have struggled to close the gap with domestic inference and training silicon — a way to “temporarily ease the frontier training bottleneck,” as analysts have described Beijing’s parallel easing.

What it means

The expansion of export licenses is a small policy step with an outsized signal. It confirms that the U.S. approach has settled into a durable pattern — tiered access, granted buyer-by-buyer, rather than a blanket ban or open trade — and that the pattern now extends beyond Nvidia to AMD.

Who wins. AMD is the clearest beneficiary, not because China will move the needle overnight but because it gains a foothold in a market it had been shut out of, at exactly the moment it is trying to prove it can take data-center share. Nvidia benefits too, adding named buyers to its H200 book. The approved Chinese firms — ZTE’s telecom arm, Maginfra, and Kingsoft’s cloud unit — get access to hardware a cut above what domestic chips reliably deliver.

Who watches nervously. China’s homegrown chipmakers face a subtler challenge: every H200 or MI300X-class part that lands in a Chinese data center is one their own accelerators did not fill. Beijing’s self-sufficiency drive and its willingness to buy American silicon are in genuine tension, and this week’s licenses tilt the balance — modestly — toward the incumbents.

What to watch next. Three things. First, shipment volume: with a trade official conceding that “very few” chips have actually moved, the gap between licenses granted and units delivered is the real measure of impact. Second, which vendors and buyers come next — whether the AMD approval is a one-off or the start of a broader widening. Third, the Blackwell line: the entire framework holds only so long as the newest generation stays restricted. Any move to loosen that ceiling would be a far bigger story than this one — and would reset the competitive math for the whole chip sector.

For now, the direction is clear even if the volume is not. The door to China’s AI-chip market is opening a little wider, and AMD just walked through it.