1. What happened to Nvidia's China AI chip market share?
Nvidia's share of China's AI accelerator market fell to 34% in the first half of 2026, a 44% decline from a year earlier, while Huawei climbed to 28%, according to market figures published in mid-October 2026. Nvidia shipped 678,000 units and Huawei 572,000 as total shipments reached roughly 2 million, a result cited by Rakesh Kumar, a University of Illinois Urbana-Champaign engineering professor, to argue that US export controls backfired.
The numbers matter because they describe a market that grew even as the American leader lost ground. Nvidia moved a very large number of chips, yet that volume amounted to a smaller slice of a bigger pie. For a company that built much of its modern business on selling accelerators to every large buyer of computing capacity, losing a decisive share in one of the world's largest national markets is a notable change in position.
The shift also reflects who is supplying the hardware. Chinese buyers are increasingly replacing foreign products with domestic ones, and Huawei is the clearest beneficiary. Its 572,000 shipments in the period put it within striking distance of the incumbent, a gap that would have been hard to imagine when the export restrictions were first introduced.
Market share figures of this kind are easy to misread, so it helps to separate three things: absolute volume, relative share and market growth. Nvidia's absolute volume remained large at 678,000 units. Its relative share dropped to 34%. And the market as a whole expanded to roughly 2 million units. A seller can therefore ship a great many chips and still lose standing if rivals grow faster, which is exactly the pattern described in the first-half data.
2. Who is Rakesh Kumar and what is his argument?
Rakesh Kumar is a professor of electrical and computer engineering at the University of Illinois Urbana-Champaign. For roughly four years he has been among the most persistent academic critics in Washington of the export controls designed to keep China's semiconductor industry from advancing. He has said he would not claim vindication outright, but the first-half data fits the case he has been making.
His central point is that restrictions can produce effects their authors did not intend. When a large customer is cut off from a leading supplier, the customer does not necessarily stop building; it may redirect money toward local alternatives. Those alternatives, in turn, receive revenue, engineering experience and a protected domestic market, all of which help them improve. In his reading, the policy designed to slow Chinese progress has helped create the conditions for faster domestic growth.
This is a claim about second-order consequences, not about whether the rules had any bite at all. The distinction is important, because the same data can be read in more than one way, and Kumar's own position acknowledges that the controls did achieve something.
3. Were the export controls ineffective?
No, and the argument does not depend on saying so. Restrictions imposed under the Biden administration, together with allied countries such as the Netherlands, limited China's ability to manufacture the most advanced chips. Those measures targeted the equipment that leading-edge production requires rather than only the finished products.
The most important piece of that equipment is extreme ultraviolet lithography, usually shortened to EUV. EUV machines use a very short wavelength of light to print extremely fine circuit patterns onto silicon, and they are essential for manufacturing chips at the leading edge. Because only a handful of firms in the world can supply this technology, it works as a chokepoint: control access to the machines and you cap how advanced a country's domestic chips can become.
China has invested tens of billions of dollars in its semiconductor supply chain over the past decade. Even so, it still depends on foreign equipment for advanced manufacturing nodes. Money alone has not delivered access to the tools at the top of the technology ladder, which is why the controls on lithography have been described as the part of the policy with real effect.
4. Why is the shift toward non-GPU chips significant?
The first-half figures point to a move toward accelerators that are not graphics processing units. GPUs are the core product category for Nvidia, so any growth in alternative designs bears directly on its business rather than on the market in general.
A GPU is a general-purpose parallel processor that became the standard engine for training and running AI models. Other accelerator designs are built specifically for machine-learning workloads and can be tailored to the needs of a particular buyer. When Chinese companies cannot rely on a steady supply of the leading foreign GPUs, they have a strong reason to invest in these alternatives and to build the software that makes them usable.
The result is a market in which domestic suppliers and non-GPU designs serve a growing share of demand. The total of about 2 million units in six months shows how large that demand is. It also shows that restricting one supplier did not shrink the overall appetite for AI computing hardware; it changed who fills the orders.
In practical terms, this means the contest is no longer only about who can build the fastest single processor. It is also about who can supply enough hardware, in a usable form, to buyers that need capacity now and cannot wait for restrictions to ease.
5. How has US policy changed in 2026?
Export controls on advanced computing moved during 2026 toward case-by-case review rather than blanket denial. The change includes case-by-case licensing and tariff adjustments under the Trump administration, which has renewed attention on the issue. The article reporting Kumar's views does not give a precise date for these changes beyond the year.
That approach gives regulators more flexibility. A case-by-case system allows officials to approve some sales and refuse others, depending on the product, the buyer and the circumstances. For manufacturers, though, it also means less predictability, since the rules governing a given shipment can differ from one decision to the next and can change again with a new administration.
The combination of tightening, then loosening, then adjusting policy has a cost that is easy to overlook. Chinese buyers who watch the rules change learn that foreign supply can be interrupted at short notice, and that lesson reinforces the case for domestic sourcing regardless of what the next round of rules says.
6. What does it mean for investors in US chip companies?
For investors in American chip companies, the first-half data indicates that revenue from China cannot be treated as stable. Policy can tighten or loosen with each administration, and the market share Nvidia holds there has already fallen sharply over a single year.
The growth to about 2 million units points the other way, showing that Chinese demand for AI hardware remains strong. The open question is how much of that demand American suppliers will be allowed or able to serve. A growing portion is being met by local suppliers and non-GPU designs, which suggests that even a more permissive licensing environment would not automatically restore the earlier share.
Analysts weighing exposure to the sector therefore face two separate variables: the size of the Chinese market, which is large and growing, and the proportion that is realistically open to foreign companies, which is shrinking. The first-half figures show them moving in opposite directions.
7. What lesson does Kumar draw for policymakers and India?
For policymakers, Kumar's point concerns second-order effects. Controls can protect a lead at the cutting edge while indirectly funding domestic rivals in lower-end segments. A restriction that keeps a rival from the most advanced technology may still channel resources into building everything beneath it.
In a recent interview he also addressed India, saying China offers both a model and a warning. Heavy state investment can build manufacturing capacity, he said, but money alone did not give China access to the most advanced tools, and reliance on foreign equipment remains a vulnerability. For a country weighing its own semiconductor ambitions, that is a reminder that funding is necessary but not sufficient.
The broader implication is that technology policy rarely produces a single clean outcome. The same measure can slow a rival at the frontier and speed it up elsewhere, and the balance between those effects only becomes visible after several years of data.
Seen from Beijing's side, the logic is straightforward. A supply that can be switched off by a foreign government is a strategic risk, and the sensible response is to build an alternative even at higher cost or lower performance in the short run. Seen from Washington's side, the aim of preserving a lead at the frontier remains valid, but the evidence that restrictions also accelerate domestic substitution adds complexity to the calculation. Neither reading is settled, and the figures reported for the first half of 2026 are a single data point rather than a final verdict.
8. What happens next?
The next set of shipment figures will show whether the trends in the first half continue: Nvidia's share, Huawei's gain and the pace of non-GPU growth. No specific release date or policy decision has been announced in the material reviewed for this report.
Further changes to the licensing approach under the current administration are possible, and any shift would be closely watched by chipmakers and their investors. Kumar's argument will likely be tested again as new data arrives, with supporters of the controls pointing to the limits on advanced manufacturing and critics pointing to the rise of Chinese suppliers.