The Macro Shift · established evidence

The Export-Control War: Nvidia, Entity Lists, and the Weaponization of Chips

Last reviewed 2026-08-11. Written by Chandranshu Kumar, Founder, Raveneye Global. · 11 min read

Every information age has had a medium that mattered most, and in this one, that medium is the chip. Washington's response to China's rise in artificial intelligence was not a tariff or a treaty. It was an export license: control over the physical hardware that trains and runs artificial intelligence itself. That license became the central instrument of the US-China technology contest, and it produced effects far messier than either government intended. Nvidia, the company most exposed to the policy, booked a $5.5 billion charge in a single quarter in April 2025, tied to a chip it had engineered specifically to stay inside earlier restrictions. China still accounted for $17 billion, 13 percent, of Nvidia's total revenue that fiscal year, even as Washington moved to cut the flow further. The same restrictions meant to deny China frontier compute are now credited with hastening the rise of Huawei, the domestic rival forecast to take most of that market instead. Six years into the fight, export controls have become Washington's sharpest lever against a rival's technology base, and also its least predictable one.

The permission slip becomes policy

Washington's response to China's rise in artificial intelligence did not arrive as a tariff schedule or a treaty. It arrived as a list. The Bureau of Industry and Security, a division inside the Commerce Department that most people have never heard of, decides which foreign buyers a US company may sell certain technology to, and its central instrument is the Entity List, a roster of firms that any US supplier needs a special license to ship to. On May 15, 2019, Commerce added Huawei Technologies to that list, cutting the world's largest telecom-equipment maker off from routine access to US-origin chips, software, and design tools.

The 2019 action proved to be an opening move rather than a settlement. On August 17, 2020, Commerce tightened the restriction again, adding 38 Huawei affiliates to the Entity List and, more consequentially, expanding a rule called the Foreign-Produced Direct Product Rule. That rule reached past US soil entirely. Any chip made anywhere in the world, by any company, using US-origin technology or software, now needed a license before it could reach Huawei. A foundry in Taiwan running American design software could no longer sell Huawei a finished chip without Washington's approval, even though no American firm had touched the part.

Export controls on strategic technology were not new. Cold War-era regimes like the Coordinating Committee for Multilateral Export Controls restricted the flow of computers, machine tools, and other dual-use technology to the Soviet bloc for four decades after 1949. What changed after 2019 was the target and the instrument: a peacetime economic rival rather than a Cold War adversary, and a single American agency's licensing power, extended worldwide through the Foreign-Produced Direct Product Rule, doing work that once required multilateral treaties among allied governments.

What began as a Huawei-specific action metastasized into standing policy. Over the following decade the Entity List grew roughly nine-fold, from a narrow instrument to a roster of nearly 1,200 entities spanning firms as different as China's Huawei and Russia's Gazprom. Chip export licensing, once a compliance footnote, became one of the US government's primary levers for slowing a rival state's technological progress. It reached finished chips, and it reached the design software, the manufacturing tools, and eventually the cloud compute that make advanced semiconductors usable.

The choice of chips as the lever was not accidental. A modern AI system is trained and run on racks of specialized processors; without them, the software has nothing to think with. Controlling who can buy the processor is functionally close to controlling who gets to build the intelligence that runs on top of it, which is why a licensing decision inside one American agency came to carry the weight of a foreign policy.

Nvidia's China math

Nvidia built the H20 for exactly this environment. When Washington's October 2022 controls barred the company's top-tier AI chips from Chinese customers, Nvidia engineered a lower-power variant that stayed inside the technical thresholds the new rules allowed. It worked, for a while. Nvidia earned an estimated $12 billion to $15 billion in revenue from H20 sales to Chinese customers in 2024, evidence that a market judged too dangerous to arm with frontier compute could still be profitably supplied with a step below it.

That workaround ended on April 9, 2025, when the US government began requiring an export license for the H20 as well, closing the gap Nvidia had engineered around the 2022 rules. The company disclosed the cost six days later: a $5.5 billion quarterly charge tied to inventory and purchase commitments for a chip it could no longer reliably ship. The announcement sent Nvidia's stock down sharply the same day. The figure was not a projection. It was money the company had already spent building chips for a market it was, for the moment, locked out of.

The scale of what was at stake shows up in Nvidia's own filings. China accounted for $17 billion of the company's revenue in fiscal year 2025, 13 percent of the total, a single national market large enough that one licensing decision in Washington could move Nvidia's reported earnings by billions of dollars in a single quarter. Few companies of any size carry that much exposure to a foreign government's export policy toward them, let alone their own government's.

The H20 was not Nvidia's first attempt at a China-compliant workaround. Earlier versions, the A800 and H800, absorbed the October 2022 controls before Washington tightened the technical thresholds again in 2023, closing that gap too. Each new compliant chip bought Nvidia a shrinking window before Commerce redrew the line, a cycle of engineering around a rule and then losing the workaround that ran through nearly three years of the relationship.

The H20 episode captured something export-control advocates rarely state plainly: restricting a chip does not only cost the country being denied it. It costs the company that built its plans around selling there, and it costs the exporting government's own credibility with the industry it is trying to steer, since a rule reversed once can be reversed again.

The reversal that wasn't

Three months after the charge, Washington reversed itself. In July 2025 the Commerce Department began granting Nvidia export licenses for the H20, restoring the company's legal path to sell into China. On paper, the immediate crisis had passed.

Nvidia never resumed meaningful sales. Chinese authorities discouraged the country's domestic AI companies from buying the H20 on security grounds, effectively closing, from the Beijing side, the door Washington had just reopened. The result was a policy that failed on its own terms twice. It cost Nvidia billions of dollars to comply with a restriction, and once the restriction lifted, the customer it had been restricted from no longer wanted the product, at least not openly.

Beijing's discouragement was not simple retaliation. Chinese officials framed the H20 as a security risk for firms building sovereign AI capacity, a justification that also served an industrial goal already underway: pushing domestic buyers toward domestic chips, whatever the near-term performance cost of doing so. A restriction meant to deny China frontier compute ended up reinforcing a decision China's own government had reasons of its own to make.

The restriction and its reversal did not track a single administration's policy either. Chip export controls toward China have drawn support from both major US political parties since 2019, a rare point of continuity across a change in the White House, even as the specific licensing decisions on individual chips like the H20 swung back and forth.

For Nvidia, the lesson was uncomfortable. Being cleared by one government no longer guaranteed access to a market, because the second government now had its own reasons, independent of Washington's, to keep the chip out. Export control had stopped being a single lever operated from one capital. Both governments were now pulling on it.

The chip that built a rival

The clearest evidence that the export-control regime produced an outcome its architects did not fully intend sits in market-share forecasts for China's AI chip market. As of 2025, Nvidia and Huawei ran close, each holding something near 40 percent of China's AI chip market, according to one industry analysis. The same analysis forecasts Nvidia's share falling to about 8 percent as Huawei scales its domestic alternative, a swing that would hand most of the world's second-largest AI chip market to a company Washington had spent six years trying to isolate.

The mechanism is not mysterious. Denying a market its preferred supplier does not remove the demand, it reroutes it. Every added license requirement, every new name on the Entity List, gave Chinese buyers one more reason to qualify a domestic chip they might otherwise have kept buying around, and gave Chinese state investment one more argument for funding the fabs and design houses capable of building it. A policy built to slow China's AI progress by denying it the best chip may, over the medium term, have accelerated China's ability to build its own.

That reading needs a caveat. China's push to build a domestic semiconductor industry predates the Entity List by years, rooted in state industrial policy that set out to reduce dependence on foreign chips well before Huawei's 2019 listing. Export controls did not create Huawei's ambition to build AI silicon. What they plausibly did was compress the timeline and harden the political will behind it, turning a long-term industrial goal into an urgent one.

Beijing's own industrial policy supplies the timeline. The Made in China 2025 plan, unveiled in 2015, set explicit targets for domestic semiconductor self-sufficiency years before Huawei's Entity List listing in 2019. The chip restrictions did not launch that ambition. They handed it a more urgent argument and, for buyers weighing a foreign chip against a domestic one, a legal reason to choose the domestic option regardless of price or performance.

None of this means Huawei's chips match Nvidia's on every measure that matters to an AI lab. Advanced packaging, software stacks built over a decade, and manufacturing yield at the smallest transistor sizes are all areas where independent assessments still put Huawei behind. The forecast market-share swing describes what Chinese buyers may choose given the restrictions they face, not a claim that the underlying hardware has closed every gap.

Entangling the allies

The 2020 expansion of the Foreign-Produced Direct Product Rule did more than close a loophole for Huawei. It established a template Washington would reuse: US export law reaching into the supply chains of other countries whenever US-origin technology sat anywhere inside them. That template drew allied governments into a fight that had, until then, been framed as bilateral, a matter between Washington and Beijing alone.

The Netherlands became the clearest case. ASML, the Dutch company that builds the extreme ultraviolet lithography machines without which no advanced logic chip can be manufactured, found itself the subject of direct diplomatic pressure from Washington. Starting in 2023, the Dutch government tightened its own licensing on ASML's most advanced deep ultraviolet tools bound for China, a restriction the Netherlands adopted under its own law but in close coordination with US policy. A company headquartered nowhere near Washington became one of the most consequential actors in a dispute between two other countries, because its machines sat at a chokepoint no substitute could reach.

The Netherlands was not the only ally drawn in. Japan, home to companies that make critical chipmaking equipment and materials, imposed its own export restrictions on advanced semiconductor manufacturing tools bound for China in 2023, timed closely with the Dutch measures and coordinated with Washington through diplomatic channels rather than a single joint treaty. Two of the world's most important toolmaking nations moved in step with a policy neither had designed on its own terms.

That is the shape export controls take once a technology becomes strategic enough. A US decision about who could buy a US company's chips became a standing obligation for allied governments to police their own firms' sales, on pain of losing access to the US technology their own industries depend on. Control of the chip did not stay a lever between two capitals. It became the organizing logic of an entire alliance's trade posture toward one country, binding partners who had their own commercial interests in the China market to a policy set largely in Washington.

The cost to those allied firms was real and rarely discussed in the same breath as Nvidia's. A lithography maker losing a national customer of China's scale absorbs its own version of the charge Nvidia booked, just without a quarterly earnings call to disclose it in. Export control, once it crosses a border, stops being a cost only the targeted country and the exporting country share.

The limits of this reading

Two of the figures in this account carry more uncertainty than the rest. The forecast that Nvidia's China AI chip share will fall from roughly 40 percent to about 8 percent comes from a single industry analysis published in 2026, built on Huawei's announced production scaling rather than measured shipment data. It describes a plausible scenario, not a locked-in outcome, and Huawei's manufacturing yield at the advanced nodes AI chips require remains a genuinely disputed figure among industry analysts.

The second open question is enforcement itself. By August 2026, a US enforcement body was reviewing how Chinese AI firms access Nvidia chips by renting GPU compute from cloud providers located outside China, a route current US export law does not classify as a chip export at all, because no physical hardware crosses a border. Whether that gap gets closed, and how much compute has already moved through it, was, at the time of this reporting, still under review rather than settled.

The 8 percent figure, in particular, should be read as one scenario among several that industry analysts have floated, not a projection anyone can bank on. Export control history over the past six years suggests the range of plausible outcomes remains wide, and that the next policy reversal, from either government, could move it again.

Held together, the record supports a narrower claim than either side of the export-control debate tends to make. The restrictions did impose real, measured costs on Nvidia, and they did slow China's access to frontier compute for a period. They did not stop China's determination to build an independent chip industry, and they created new enforcement problems, offshore compute rental among them, that did not exist before the controls did.

What the last six years demonstrate is not that export controls failed or succeeded outright. It is that the chip became, in a short span, the medium every other layer of the AI economy now runs on top of, and that controlling a medium this consequential is a harder thing to hold onto than adding a name to a list. The same pattern that once played out over printing presses and telegraph cables is playing out again, over a wafer of silicon a few centimeters wide.

The evidence

Key findings, with their sources

The US Commerce Department added Huawei Technologies to its Entity List on May 15, 2019, cutting the company off from routine access to US-origin chips, software, and design tools.

established Belfer Center for Science and International Affairs, citing Commerce Department records (2019 to 2020).
On August 17, 2020, Commerce added 38 Huawei affiliates to the Entity List and expanded the Foreign-Produced Direct Product Rule to cover chips made anywhere in the world using US-origin technology.

established Hogan Lovells, International Trade Insights (2020).
The Entity List has grown roughly nine-fold over the past decade to nearly 1,200 entities, spanning firms from China's Huawei to Russia's Gazprom.

established Belfer Center for Science and International Affairs (2025).
Nvidia earned an estimated $12 billion to $15 billion in revenue from H20 chip sales to Chinese customers in 2024, before the US required export licenses for the H20 starting April 9, 2025.

established Brookings Institution / IFP policy analysis (2025).
Nvidia disclosed a $5.5 billion quarterly charge in April 2025 tied to H20 export restrictions, sending its stock down sharply the same day.

established CNBC (2025).
China accounted for $17 billion, 13 percent, of Nvidia's total revenue in fiscal year 2025.

established Nvidia Corporation, Form ARS, fiscal year 2025 (SEC filing).
In July 2025 Commerce reversed course and granted Nvidia export licenses for the H20, but Nvidia never resumed sales after Chinese authorities discouraged domestic AI firms from buying the chip on security grounds.

established NPR (2025).
Nvidia's China AI chip market share, roughly 40 percent as of 2025 and running close to Huawei's, is forecast to fall to about 8 percent as Huawei scales domestic production.

emerging MarketScale / Tech Buzz China analysis (2026).
By August 2026, a US enforcement body was reviewing how Chinese AI firms access Nvidia chips through offshore cloud-compute rentals, a route not classified as a chip export under current US law.

emerging Bloomberg, reported by TechTimes (2026).

Calibration

What is proven, what is promising, what is unproven

Evidence tier	Tactics	What the evidence says
established	The documented history of Entity List actions against Huawei, the extraterritorial reach of the Foreign-Produced Direct Product Rule, and Nvidia's own disclosed financial figures: the $5.5 billion charge, the $17 billion, 13 percent China revenue share, and the July 2025 license reversal.	Corroborated by primary sources: Commerce Department actions, SEC filings, and company disclosures reported across multiple independent outlets.
emerging	The forecast that Nvidia's China AI chip share will fall from roughly 40 percent to about 8 percent, and the scale of the offshore compute-rental workaround under enforcement review.	Each comes from a single analysis or a single enforcement-review report published close to 2026, with real data behind it but not yet cross-verified by independent sources.
contested	How much of China's push toward domestic chip independence was caused by the export controls specifically, versus an industrial policy already underway before 2019, and whether the offshore compute loophole meaningfully undercuts the controls' intent.	China's chip self-sufficiency drive predates the Entity List; the controls plausibly accelerated it but did not create it, and the causal weight is not something any single source settles.

Reference

Glossary

Entity List: A US Commerce Department roster of foreign parties that US companies need a special license to sell certain technology to. Inclusion does not ban trade outright, it requires government approval for it.
Bureau of Industry and Security (BIS): The Commerce Department division that administers US export controls, including the Entity List and the license requirements placed on chips like Nvidia's H20.
Foreign-Produced Direct Product Rule (FDPR): A US rule that extends export licensing requirements to products made anywhere in the world, by any company, if they were made using US-origin technology, software, or equipment.
H20 chip: A lower-power Nvidia AI chip engineered to stay under the technical thresholds of the 2022 export controls, later itself brought under a license requirement in April 2025.
Offshore compute rental: Accessing chip capability by renting GPU time from a cloud provider located outside a buyer's home country, a workaround current export law does not classify as a chip export because no physical hardware crosses a border.

Straight answers

Frequently asked questions

Why did the US restrict chip exports to China?

The restrictions began narrowly, with Huawei added to the Commerce Department's Entity List in May 2019 over national-security concerns tied to its telecom equipment. The scope widened over the following years to cover the chips, software, and manufacturing tools needed for advanced AI, as Washington came to see compute itself as the resource most worth denying a strategic rival.

How much has the export-control fight cost Nvidia?

Nvidia disclosed a $5.5 billion charge in a single quarter in April 2025 tied to H20 chips it could no longer reliably sell to China. China still accounted for $17 billion, 13 percent, of Nvidia's total revenue in fiscal year 2025, so the stakes on both sides of the policy run into the billions.

Did Nvidia ever get to sell the H20 to China again?

Commerce granted Nvidia export licenses for the H20 in July 2025, reversing the April restriction. Nvidia did not resume meaningful sales, because Chinese authorities discouraged domestic AI firms from buying the chip on security grounds, closing the market from the Beijing side instead.

Is Huawei replacing Nvidia in China's AI chip market?

One industry analysis put Nvidia and Huawei near even, each around 40 percent of China's AI chip market, as of 2025, and forecast Nvidia's share falling to about 8 percent as Huawei's domestic production scales. That figure is a single-source projection, not a settled outcome, and independent assessments still rate Huawei behind Nvidia on some technical measures.

Can Chinese firms get around the chip restrictions?

One route under review by August 2026 is renting GPU compute from cloud providers located outside China, since current US export law does not classify a compute rental as a chip export. How large that workaround has grown was still under enforcement review, not publicly settled, as of the reporting available.

Provenance

Sources

Belfer Center for Science and International Affairs, "The United States Entity List Limits American Exports" (2025)belfercenter.org
Hogan Lovells, International Trade Insights, "U.S. Adds 38 New Huawei Affiliates to Entity List While Again Expanding Foreign-Produced Direct Product Rule" (2020)internationaltradeinsights.com
Brookings Institution, "Ball game's over: The US is out of the AI chip market in China" (2025)brookings.edu
CNBC, "Nvidia says it will record $5.5 billion quarterly charge tied to H20 processors exported to China" (2025)cnbc.com
Nvidia Corporation, Form ARS (Annual Report to Shareholders), fiscal year 2025sec.gov
NPR, reporting on Nvidia's H20 export license reversal and China's discouragement of domestic purchases (2025)npr.org
MarketScale, "Nvidia's China AI Chip Share Is Forecast to Collapse From 40% to 8% as Huawei Scales" (2026)marketscale.com
Bloomberg, "US Reviews China's Offshore Access to Nvidia Chips After AI Breakthroughs" (2026)bloomberg.com

Every figure above is attributed to a real, dated source and tagged with its evidence tier. Where a claim could not be verified to a primary source, it is not stated as fact.