US Chip Export Controls Explained: The Complete 2026 Guide
One policy decision cost Nvidia $4.5 billion in a single quarter. Here is everything you need to understand about how chip export controls work, why they matter, and where they are heading next.
In April 2025, the US government sent Nvidia a letter. It contained no fine, no indictment, no court order. Just a notification that a license would now be required to export its H20 chip to China. Within days, Nvidia disclosed a $4.5 billion charge against earnings. The stock moved billions in market cap in 48 hours. That is what chip export controls look like in 2026: a single bureaucratic decision with the destructive radius of a major earnings miss.
If you work in enterprise technology, semiconductor supply chains, policy compliance, or investment, chip export controls are no longer a niche regulatory topic you can delegate to legal. They are now a first-order business risk. This guide explains how the system works, what it has achieved, where it is failing, and what the evidence tells us about where it is going.
What Are Chip Export Controls?
Chip export controls are US government regulations that restrict which advanced semiconductors, chip-making equipment, and related technology can be sold to specific countries, companies, or individuals. They are administered by the Bureau of Industry and Security (BIS) inside the Department of Commerce, operating under the Export Administration Regulations (EAR) with legal authority rooted in the Export Control Reform Act of 2018 (ECRA).
The policy logic is straightforward: the most powerful AI chips in the world require billions of dollars in research, highly specialized equipment, and years of manufacturing refinement. The US and a small number of allied companies control each of those inputs. By restricting exports, Washington aims to deny China’s military and intelligence services access to the computing power needed to train frontier AI models, design advanced weapons systems, and run large-scale surveillance.
Since October 2022, controls have primarily targeted China. But the system’s reach extends far beyond direct US-to-China sales. Through a legal mechanism called the Foreign Direct Product Rule, the US has effectively placed itself as the licensing authority for nearly every advanced chip manufactured anywhere on earth.
Chip export controls restrict advanced semiconductor sales to specific countries and end users. Since October 2022, the primary target has been China’s access to AI chips above defined performance thresholds, with Nvidia’s H100 and equivalent GPUs as the central focus.
The Four Enforcement Mechanisms
Understanding chip export controls means understanding four distinct tools that BIS uses in combination. Each one does a different job, and each one has a different set of vulnerabilities.
1. The Commerce Control List and Performance Thresholds
The Commerce Control List assigns Export Control Classification Numbers (ECCNs) to specific chip types. Chips meeting certain performance thresholds require a BIS export license before they can be sold to restricted destinations. Under the original October 2022 rules, chips capable of more than 300 tera operations per second or an interconnect speed of more than 600 gigabytes per second were restricted.
Nvidia immediately responded by designing the H800 and A800, stripped-down versions of its most powerful chips that fell just below the thresholds. BIS closed that gap in October 2023 by switching from raw compute metrics to performance density calculations, eliminating the loophole.
2. The Entity List
The Entity List is a register of foreign companies, organizations, and individuals that BIS has determined pose national security or foreign policy risks. Any export to an Entity Listed company requires a specific BIS license, and the default presumption is denial. Being listed effectively severs a company from the US technology supply chain.
Huawei was added in 2019. SMIC subsidiaries followed. By mid-2026, hundreds of Chinese semiconductor and AI companies are listed, including most of the major players in Huawei’s supply chain ecosystem. In March 2025 alone, the Trump administration added 42 additional Chinese entities.
3. The Foreign Direct Product Rule
This is the most powerful tool in BIS’s arsenal. The Foreign Direct Product Rule extends US jurisdiction to products made abroad using US equipment, software, or technology. Because virtually every advanced chip manufactured anywhere in the world is produced on machines containing US-origin technology, this means TSMC in Taiwan, Samsung in South Korea, and every other major fab must comply with BIS rules when selling to restricted end users.
In practical terms, it makes the US the licensing authority for the global semiconductor industry. The FDPR was expanded in 2024 to cover high-bandwidth memory and, in January 2025, was extended to AI model weights, taking US jurisdiction from hardware into software for the first time.
4. The Validated End-User Program
The Validated End-User program runs in the other direction. It allows designated entities to receive certain dual-use items without requiring a BIS license for each transaction. In September 2025, BIS removed the named Chinese facilities of Samsung and SK Hynix from the VEU program, effective December 31, tightening controls on Korean-operated fabs operating inside China.
A Timeline of Key Actions: 2022 to 2026
The policy has moved faster than most compliance teams have tracked it. Here is the full sequence of major decisions.
BIS restricts export of advanced logic chips including Nvidia’s A100 and H100 GPUs to China. Semiconductor manufacturing equipment exports to advanced Chinese fabs also restricted. This is the foundational action that everything since has built on.
BIS tightens the original rules. Nvidia’s compliant H800 and A800 chips are swept in. Performance density thresholds replace raw compute metrics, blocking the design-around strategy.
BIS proposes a global three-tier licensing framework. Tier 1 includes 18 US allies exempt from licensing. Tier 2 covers most of the world with per-country compute caps. Tier 3 covers China, Russia, and arms-embargoed nations with presumption of denial.
Nvidia is informed that a license is required for H20 exports to China. The company discloses a $4.5 billion charge associated with H20 excess inventory and purchase obligations. Sales of H20 products were $4.6 billion in Q1 FY2026 before the requirement took effect.
The Biden-era AI Diffusion Framework is paused pending revision. BIS also assesses that Huawei developed its Ascend chips in violation of US controls and warns that using those chips risks violating export laws.
Yih-Shyan Liaw, co-founder of Super Micro Computer, is arrested along with two colleagues for conspiring to violate US export controls on AI chips. The case involves $2.5 billion in allegedly illicit chip transfers.
BIS changes the export license review policy for advanced computing chips to China and Macau. H200 chips shift from presumption of denial to case-by-case review. The White House announces a simultaneous 25% tariff on chips meeting the same performance thresholds.
Nvidia CEO Jensen Huang tells CNBC that Nvidia has “largely conceded” China’s AI chip market to Huawei. Huawei is reported heading toward $12 billion in annual AI chip revenue.
What Chips Are Actually Banned From Export to China?
| Chip | Manufacturer | Current Status (June 2026) | Key Ruling |
|---|---|---|---|
| A100 | Nvidia | Broadly restricted | Oct 2022 original rule |
| H100 | Nvidia | Broadly restricted | Oct 2022 original rule |
| H800 / A800 | Nvidia | Restricted | Oct 2023 tightening |
| H20 | Nvidia | License required, presumption of denial | Apr 2025 license requirement |
| H200 | Nvidia | Case-by-case review | Jan 2026 policy shift |
| Blackwell GPUs | Nvidia | Broadly restricted | Covered under advanced compute thresholds |
| MI300X | AMD | Broadly restricted | Meets performance density thresholds |
The technical cutoff is defined by a combination of total processing power (TPP) and interconnect bandwidth. Any chip meeting or exceeding those thresholds requires a license for export to China or Macau. The January 2026 rule change did not lift restrictions broadly. It shifted the H200 specifically from automatic denial to a case-by-case review process, which matters for multinational companies applying for specific customer approvals.
The Financial Reality: What Controls Are Costing Nvidia
For years, export control costs were theoretical. The April 2025 H20 decision turned them into SEC-reportable numbers.
Nvidia disclosed in its Q1 FY2026 SEC 8-K filing that the company incurred a $4.5 billion charge associated with H20 excess inventory and purchase obligations. Sales of H20 products were $4.6 billion in Q1 FY2026 before the new license requirements took effect. The company was unable to ship an additional $2.5 billion of H20 revenue in the first quarter. For Q2 FY2026, guidance reflected an additional $8.0 billion loss in H20 revenue due to export control limitations.
Put those numbers together: a single policy decision on a single chip SKU targeting a single country is tracking toward more than $30 billion in annualized revenue impact for one company.
“Huawei is very, very strong. They had a record year, they’ll likely have an extraordinary year coming up, and their local ecosystem of chip companies are doing quite well, because we’ve evacuated that market. We’ve really largely conceded that market to them.”
Jensen Huang, CEO, Nvidia Corporation. CNBC interview, May 21, 2026.
The China context is significant. Nvidia recorded $17.1 billion in annual sales to customers with a China or Hong Kong address, making China its fourth-largest market. The H20 alone generated an estimated $12 billion to $15 billion in revenue for Nvidia in 2024. Huang has previously stated that revenue from China dropped to half of pre-export control levels. Huang’s “largely conceded” admission is the clearest public acknowledgment from the semiconductor industry that controls have permanently restructured the competitive landscape, not temporarily disrupted it.
This signals a structural, not cyclical, shift in Nvidia’s China business. Investors who are modeling a policy reversal and revenue recovery are pricing in political outcomes that the market evidence does not support. BNP Paribas analyst David O’Connor placed Nvidia’s rolling 12-month H20 revenue hit at $15 billion. That is not a timing problem. That is a permanent market transfer to Huawei.
The Enforcement Crisis No One Wants to Talk About
Here is the tension at the center of chip export controls: the rules are sophisticated. The enforcement is not.
In March 2026, US authorities arrested Yih-Shyan Liaw, co-founder of Super Micro Computer, along with two colleagues. The indictment charged them with conspiring to illicitly ship AI servers containing Nvidia chips worth $2.5 billion to Chinese customers between 2024 and 2025. The alleged operation routed hardware through Taiwan, Malaysia, Vietnam, and the UAE before final delivery to Chinese buyers. Shipments allegedly escalated to $150 million in a single two-month window in early 2025.
That one case is larger than the entire federal enforcement budget. According to reporting by CyberScoop, federal spending on policing export controls amounted to $122 million in all of 2025. The disparity is stark: $2.5 billion in a single documented case versus $122 million to police the entire global system.
“China is betting that its network of smugglers and shell companies can find the leaks in the BIS export control enforcement barrier.”
Gregory C. Allen, Senior Adviser, Wadhwani AI Center, CSIS. Former Director of Strategy and Policy, DoD Joint AI Center. December 2024.
Allen, described by The Economist as “very much at the center of the formulation of current US policy” on chip controls, argues the architecture is correct but structurally underfunded. His view is that the solution is more enforcement resources, not policy relaxation. But the resource gap is not a rounding error. At a 20-to-1 ratio between a single smuggling case and the annual enforcement budget, no amount of marginal funding increases will close it quickly.
The January 2026 BIS settlement against a European company for an in-country transfer made by its Chinese subsidiary demonstrates that export control liability is extraterritorial. Multinational organizations must audit every third-party data center partner, reseller, and logistics provider for Entity List exposure, not just direct sales channels.
The Case Against Controls: What the Critics Get Right
The strongest argument against the current policy architecture is not ideological. It is empirical. And it starts with a chip company you may not have expected to be at the center of the story.
In May 2024, Chinese AI startup DeepSeek released a best-in-class open-weight model reportedly trained on Nvidia A100 chips stockpiled before the October 2022 controls took effect. Its even more impressive DeepSeek-R1 reasoning model was trained on H800 chips, which were not restricted until October 2023. China built frontier AI capabilities on hardware it had legally acquired before the controls existed.
The “hardware-only” critique extends beyond DeepSeek. Berkeley researchers Ritwik Gupta, Leah Walker, and Andrew Reddie published peer-reviewed evidence in November 2024 showing that US chip export controls are widely permeable, with Chinese AI labs accessing restricted hardware through circumvention. Their paper examined Tencent and other major Chinese AI labs and found systematic evidence of diversion at scale.
“US export controls on chips and hardware alone will not prevent China from further developing advanced AI.”
Chatham House, Digital Society Programme. Published April 29, 2026.
The Chatham House position, published in April 2026, synthesizes the research: algorithms, open-source model weights, and software innovation can partially decouple AI capability from chip access. If that argument is correct, then the economic costs of controls (lost revenue, accelerated Chinese self-sufficiency, damage to allied relationships) may exceed the strategic benefits.
The counterargument from policy hawks rests on a time-window thesis. If transformative AI arrives by 2027, then denying China frontier compute today buys decisive advantage. Anthropic CEO Dario Amodei has suggested that “super powerful AI” could emerge by 2026 to 2027, with significant military implications for whichever nation leads. That forecast is the load-bearing assumption for the entire policy framework. If it is wrong, or if the timeline extends by even two or three years, the United States will have permanently ceded a $15 to $17 billion annual market to Huawei and funded the domestic chip self-sufficiency it sought to prevent.
What the data shows is genuinely mixed. China domestically produces AI chips equivalent to only 1 to 2% of US production in 2026, suggesting controls have maintained a massive compute gap. But Chinese models’ share of global AI token usage grew from approximately 1% in 2025 to approximately 30% in 2026, according to the American Enterprise Institute. Compute access translates to AI market share. China’s inference footprint is growing explosively even under controls. That is not the outcome the 2022 policy architects predicted.
What to Watch in the Next 6 to 18 Months
The policy landscape is currently bifurcated and unstable. The Trump administration is simultaneously relaxing export rules (the H200 case-by-case pathway) and intensifying enforcement actions (the Super Micro indictment, additional Entity List additions). Those two directions are not contradictory from a political standpoint but they create serious compliance uncertainty for any organization in the supply chain.
Three specific developments to monitor:
- The Chip Security Act. The legislation, advancing through the US House in late April 2026, would require companies to verify that semiconductors used in AI remain in authorized locations. It would add a physical tracking layer to the current paper-based compliance system. If it passes, compliance obligations for cloud providers and data center operators will change materially.
- The AI OVERWATCH Act. Pushed through the House Foreign Affairs Committee in January 2026 by Chair Brian Mast, this bill would grant Congress veto power over AI chip export licenses, a power that currently belongs to the Department of Commerce. If it advances, executive flexibility on licensing decisions narrows significantly.
- SMIC’s capacity expansion. SMIC’s advanced node capacity is estimated at approximately 45,000 wafer starts per month in 2025, expanding toward 60,000 wspm through 2026, according to Oplexa’s Global Semiconductor Supply Chain Risk and Forecast Report. SMIC is producing these chips without EUV lithography, validating the argument that controls cannot fully stop Chinese chip development and providing the most direct test of the time-window thesis.
For investors, the smarter position is to treat BIS rule changes and Entity List updates as leading indicators for semiconductor stock moves, not lagging ones. Companies selling export control compliance infrastructure and KYC tools face direct structural tailwinds regardless of which direction the policy moves.
FAQ: Chip Export Controls Explained
What are chip export controls?
Chip export controls are US government regulations administered by BIS that restrict which advanced semiconductors can be sold to specific countries or companies. Since October 2022, the rules have primarily targeted China’s access to AI chips above defined performance thresholds, preventing companies like Nvidia from selling H100 and equivalent GPUs without a license. Controls also cover chip-making equipment and, since January 2025, AI model weights.
Why is the US restricting chip exports to China?
The US restricts chip exports to China to prevent Chinese military and intelligence agencies from using advanced AI computing power for weapons development, nuclear programs, and surveillance. BIS has specifically cited military-intelligence and WMD end-use risks as the policy rationale. The stated goal is to maintain a US lead in frontier AI and deny China the compute needed to match US capabilities.
What is the Entity List in chip export controls?
The Entity List is a BIS register of foreign companies and individuals that pose national security risks. Exporting any item to an Entity Listed company requires a specific BIS license with a presumption of denial. As of mid-2026, hundreds of Chinese semiconductor and AI companies are listed, including Huawei (added 2019), SMIC subsidiaries, and major AI firms. Being listed effectively cuts a company off from the US technology supply chain.
What is the Foreign Direct Product Rule (FDPR)?
The FDPR extends US export control jurisdiction to products made abroad using US equipment, software, or technology. Because virtually all advanced chips are made on American-origin equipment, it means TSMC, Samsung, and other non-US fabs must comply with BIS rules when selling to restricted end users. It was expanded in 2024 to cover AI model weights, taking US jurisdiction from hardware into software for the first time.
Has China been able to circumvent chip export controls?
Yes, extensively. The Super Micro case alone involved $2.5 billion in allegedly illicit chip transfers routed through Southeast Asia and the Middle East. SMIC has produced 7nm-class chips using pre-control DUV equipment, and DeepSeek trained frontier AI models on chips legally acquired before 2022 controls took effect. Federal prosecutions confirm that industrial-scale smuggling networks have operated throughout the control regime.
What is the AI Diffusion Rule?
The AI Diffusion Rule was a Biden-era regulation issued January 13, 2025, that created a global three-tier licensing framework for advanced AI chips. Tier 1 included 18 US allies exempt from licensing. Tier 2 covered most of the world with per-country compute caps. Tier 3 covered China and Russia with presumption of denial. The Trump administration paused the rule in May 2025 and replaced it with a narrower H200 licensing pathway in January 2026. A replacement framework remains under development.
What does the January 2026 policy change mean for H200 chip exports?
BIS shifted the H200 from a presumption of denial to a case-by-case review standard for exports to China and Macau. This does not open the market broadly. It means companies can apply for specific export licenses for specific customers and expect those applications to be evaluated on their merits rather than automatically rejected. The White House simultaneously announced a 25% tariff on chips meeting the same performance thresholds.
What You Now Understand
Chip export controls are not a trade dispute. They are an attempt to use the US position at the chokepoints of global semiconductor supply chains to slow China’s AI development by denying access to the most powerful training hardware on earth. The architecture is sophisticated. The enforcement is structurally underfunded. The policy is producing outcomes its architects did not predict.
China is domestically producing only 1 to 2% of US chip output. But its models now account for roughly 30% of global AI token usage. Nvidia has conceded the Chinese market to Huawei. Huawei is heading toward $12 billion in annual AI chip revenue. Shipments of one million H200s would increase China’s total installed AI compute by 250% relative to domestic production alone, according to the Council on Foreign Relations. That is the territory into which the January 2026 policy relaxation has stepped.
Whether you read that as a dangerous concession or a rational acknowledgment of an unenforceable status quo depends on which load-bearing assumption you accept about the AI development timeline. What is not in dispute is the scale of the economic stakes, the inadequacy of current enforcement resources, and the fact that the policy is still being written in real time.
The three things worth watching closely: the Chip Security Act’s physical tracking provisions, SMIC’s wafer capacity numbers (they will tell you how much the controls have actually delayed China’s timeline), and the next BIS Entity List update, which remains the single most market-moving document in the semiconductor industry.
Stay Ahead of the Next BIS Decision
The Neural Loop delivers policy shifts, enforcement actions, and market implications directly to your inbox before they move markets. No noise. Just signal.
Subscribe to The Neural Loop