Home / Research / AI Chip Export Statistics

Market research · October 13, 2026

AI Chip Export Statistics: 2026 Report

By Darin ten Bruggencate, Founder, AI Industry Reviews

Since August 2022 the US government has repeatedly changed the rules on which advanced AI chips can be exported, and to whom. Those rules decide where AI hardware goes, how much US chipmakers can sell abroad and which countries can build AI data centers.

Our analysts studied US exports in four trade codes, or HS codes, that cover servers, computer units, computer parts and processor chips. The trade data came from UN Comtrade, collected on October 8, 2026. We measured exports by destination from 2022 to July 2026. We also reviewed Nvidia's and AMD's revenue by region and the charges both took after new license rules. The key finding is that exports in the four codes reached $108.53 billion in 2025, up 42.2% from 2024.

In the following sections, we break down US exports by trade code and by destination, with a closer look at China and Hong Kong. We then compare Nvidia's and AMD's sales by region, total the costs the rules imposed on both companies and lay out a timeline of the export controls.

US AI Chip Export Statistics by HS Code

Total US exports in the four trade codes show how much computing hardware leaves the country each year and how that trade has changed since the first chip export rules in 2022.

US Exports by HS Code, All Destinations ($ millions)

• US exports in the four codes reached $108.53 billion in 2025, up 42.2% from $76.32 billion in 2024 and up from $56.63 billion in 2022. In the first seven months of 2026 they were $87.28 billion, 49.8% more than in the same months of 2025.
• Exports of processing units (8471.50), the code that covers servers, nearly quadrupled from $7.75 billion in 2022 to $30.89 billion in 2025. They rose another 97.8% year over year in January to July 2026, to $28.11 billion.
• Computer parts and accessories (8473.30) were the largest of the four codes in 2025 at $35.06 billion, up 49.7% from 2024.
• Processors and controllers (8542.31) grew the least, by 6.8% in 2025 to $34.58 billion and by 7.1% in the first seven months of 2026.
• The growth is in finished servers and parts, not loose chips, which suggests US firms are shipping AI capacity as assembled systems. Processing units rose 97.8% and parts 74.0% in January to July 2026, against 7.1% for processors. If this pace holds, 2026 exports in the four codes will pass 2025's $108.53 billion before year end, with server makers and their suppliers gaining most.

Where US AI Hardware Exports Go

Destinations show which countries buy the most US servers and processors, which helps explain who is affected when the export rules change.

Top Five Destinations by HS Code, 2025 ($ millions and share of US exports in the code)

US Exports of Processing Units (HS 8471.50) to Selected Destinations ($ millions)

• Mexico was the largest destination in all four codes in 2025. It took 58.3% of US exports of computer parts (8473.30), the same share as in 2022 (58.4%), and 62.6% in January to July 2026.
• US exports of processing units to Mexico grew from $715 million in 2022 to $6.17 billion in 2025, and reached $9.40 billion in the first seven months of 2026, 33.4% of the code's total. In the same months of 2025 Mexico's share was 12.7%.
• Processing-unit exports to Taiwan rose from $258 million in 2022 to $2.32 billion in 2025. Exports to Malaysia were $2.10 billion in January to July 2026, up from $176 million a year earlier.
• Exports of processing units to the UAE were $745 million in 2025, up from $75 million in 2022, and $120 million in January to July 2026. BIS moved the UAE into a more favorable country group in a rule published July 14, 2026, so the January to July 2026 figures almost entirely predate that change.
• Much of this trade appears to go to assembly and distribution hubs rather than to the places where AI is used. Mexico takes about 60% of US computer parts exports, and the statistics record only the first foreign destination. Buyers and investors should read jumps such as Malaysia's, from $176 million to $2.10 billion, as signs of where servers are built, not necessarily where they run.

US Exports to China and Hong Kong

China has been the main focus of US export rules on AI chips, and exports to China and Hong Kong show how that trade has moved as the rules changed.

US Exports to China and Hong Kong Combined, by HS Code ($ millions and share of US exports in the code)

• China and Hong Kong together took 5.1% of US processing-unit exports in 2022 and 1.7% in 2025. In January to July 2026 they took 0.8%, or $237 million.
• The controls appear to have cut China off mainly from finished US systems, not from chips as a whole. China and Hong Kong's share of processing units fell from 5.1% to 0.8%, and of other computer units from 19.6% to 1.2%. Their processor share is back to 24.9%, so any further loosening would likely show up first in processor exports.
• In processors and controllers (8542.31), the China and Hong Kong share fell from 24.6% in 2022 to 12.1% in 2023, then rose to 22.3% in 2025 and 24.9% in January to July 2026. This code also includes CPUs and microcontrollers that the advanced computing rules do not cover.

Nvidia Revenue by Region

Trade codes mix AI hardware with ordinary computers, so company figures add a second view. The AI Industry Reviews research team followed Nvidia's revenue by where its customers are based.

Nvidia Revenue by Customer Headquarters Location, Fiscal Years ($ millions and share of total)

Nvidia Revenue by Customer Headquarters Location, by Quarter ($ millions)

• Revenue from China-headquartered customers, including Hong Kong, fell 21.4% in fiscal 2026 to $19.68 billion, while Nvidia's total revenue rose 65.5% to $215.94 billion. China's share of revenue fell from 19.2% to 9.1%.
• China revenue dropped from $9.66 billion in the quarter ended April 27, 2025, to $3.99 billion in the next quarter, after the H20 license requirement took effect, and to $2.97 billion in the quarter after that. In the quarter ended July 26, 2026, it was $7.88 billion, 8.2% of revenue.
• Nvidia's growth now depends mostly on US buyers, which ties its results closely to US data center spending. US-headquartered customers made up 69.3% of fiscal 2026 revenue, up from 51.8% two years earlier. China's rebound to $7.88 billion in the latest quarter shows licensed sales can add revenue back, but at 8.2% it remains well below China's earlier 20% share.
• Revenue from Taiwan-headquartered customers was $26.99 billion in the quarter ended July 26, 2026, up from $12.01 billion in the previous quarter. Nvidia estimated that in fiscal 2026, 76% of its Data Center revenue from Taiwan-headquartered customers was attributed to end customers based in the US and Europe.

Until the third quarter of fiscal 2026, Nvidia reported revenue by customer billing location, and Singapore appeared as a large market.

Nvidia Revenue by Customer Billing Location, Earlier Basis ($ millions)

• Singapore accounted for $23.68 billion, or 18.1% of Nvidia's fiscal 2025 revenue, by billing location. Nvidia said customers use Singapore to centralize invoicing, that its products are almost always shipped elsewhere, and that shipments to Singapore were less than 2% of fiscal 2025 revenue.
• In the quarter ended July 27, 2025, Singapore billing was $10.16 billion, 21.7% of revenue. Nvidia said over 99% of controlled Data Center compute revenue billed to Singapore in that quarter was for orders from US-based customers.
• Chipmakers' regional figures say more about where buyers are billed than where chips end up. Singapore was billed 18.1% of Nvidia's fiscal 2025 revenue but received less than 2% of shipments, and China's total for that year was $17.11 billion on one basis and $25.05 billion on the other. Anyone comparing regions over time should stick to one basis and treat hub countries as pass-through points.

AMD Revenue by Region

AMD reports its sales by billing location once a year, which shows where its customers are billed and how much of its business comes from China.

AMD Sales by Customer Billing Location ($ millions and share of total)

• AMD's sales billed to China and Hong Kong rose 24.4% in 2025 to $7.75 billion, 22.4% of its revenue. The share was 22.1% in 2022, dropped to 15.1% in 2023 and rose to 24.2% in 2024.
• Sales billed to Taiwan grew 57.1% in 2025 to $5.19 billion. Singapore accounted for $4.28 billion, or 12.4%.
• AMD is more exposed to China than Nvidia, which makes it more sensitive to any change in the rules. China and Hong Kong were 22.4% of AMD's 2025 sales by billing location, against 9.1% of Nvidia's fiscal 2026 revenue by headquarters, though the two bases differ. If licensed MI308 and MI325 sales grow, AMD stands to gain more from looser rules and lose more from tighter ones.

Export-Rule Charges at Nvidia and AMD

Nvidia and AMD each told investors what the new license requirements cost them, and those amounts put a dollar figure on what the rules meant for each company.

Charges and Revenue Effects Tied to US Export Licensing Disclosed in SEC Filings

• Nvidia said that on April 9, 2025, the US government told it that exports of its H20 chip to China, including Hong Kong and Macau, and to Country Group D:5 countries would need a license. The resulting $4.5 billion charge was less than Nvidia had first expected because it was able to reuse some materials.
• Export rules now carry direct costs that can cut a chipmaker's margins in a single quarter. Nvidia's gross margin fell to 60.5% from 78.4% in the quarter of its $4.5 billion H20 charge, and AMD took about $800 million on MI308. With an expected 15% government share and a 25% tariff that Nvidia says it absorbs, a licensed China sale is likely to earn less than the same sale did before the rules.
• Both companies said US officials expected the government to receive 15% of revenue from licensed sales of these chips to China. Nvidia's later filings say "15% or more." Both said the government had not published a regulation requiring the payment.
• Nvidia's licensed H200 shipments to China were less than 1% of its Data Center revenue in the quarter ended July 26, 2026. Chips shipped under the H200 licenses must be inspected in the US first and face a 25% tariff on import for that inspection, which Nvidia said it has not passed on to customers. AMD's MI325 licenses carry the same inspection and tariff terms.

Timeline of US AI Chip Export Controls

The export rules have changed many times since 2022, and a dated list helps readers match each shift in the trade figures to the rule in force at the time.

US Export-Control Actions on AI Chips, August 2022 to July 2026

• Nvidia says the government announced license requirements for its A100 and H100 chips bound for China in August 2022. The BIS rule dated October 7, 2022, set license requirements for advanced computing chips, computers containing them and some chipmaking equipment. Rules dated October 17, 2023, and December 2, 2024, widened the controls to more chips, more destinations and high bandwidth memory.
• The AI Diffusion rule published January 15, 2025, would have required licenses for advanced computing chips worldwide. BIS announced on May 13, 2025, that it would rescind it before its compliance date of May 15, 2025.
• The January 15, 2026 rule changed license review for chips such as the Nvidia H200 and AMD MI325X bound for China and Macau from a presumption of denial to case-by-case review. Exporters must certify that shipments to China and Macau will not exceed 50% of the product's shipments to US customers and that each shipment is checked by a third-party lab in the US.
• Because the rules have changed direction several times since 2022, exporters and buyers should plan for further change rather than a settled regime. The suspension of the 50% affiliates rule, published September 30, 2025, is set to end November 9, 2026, unless BIS extends it. US officials said in September 2026 that the related US-China trade truce now runs to January 10, 2027, but BIS had not published a formal extension as of mid-October.

What AI Chip Export Trends Mean for 2027 and Beyond

American AI hardware is being shipped abroad faster than ever, but mostly as finished servers routed through assembly hubs, not as chips sold into China. That shift means US chipmakers now earn most of their money from American data center builders, so their fortunes rise and fall with US spending on AI rather than with access to China.

The rules have changed many times since 2022 and will keep changing. Licensed sales to China can bring some revenue back, but the payments the government expects on those sales and new tariffs make each one less profitable. The next date to watch is the end of the suspension of the rule covering affiliates of restricted companies, because it decides which foreign buyers can get US chips.

So if you own chip stocks or buy AI hardware outside the US, keep track of each change to the export rules, because a single decision can move sales and delivery times within weeks.