Export Controls and AI Chips: How Semiconductors Became the World’s Most Contested Technology

The Most Consequential Trade Policy You Have Never Heard Of

On October 7, 2022, the US Bureau of Industry and Security (BIS) issued export control regulations that sent shockwaves through the global technology industry. The regulations restricted the export of advanced semiconductors, semiconductor manufacturing equipment, and related technologies to China. The goal was explicit: prevent China from acquiring or manufacturing the most advanced chips needed for AI training, supercomputing, and advanced weapons systems.

This was not a tariff or a trade negotiation. It was a technology denial strategy designed to create a permanent capability gap between the United States and its primary strategic competitor by controlling access to the foundational technology of the AI era.

National Security Advisor Jake Sullivan had articulated the doctrine weeks earlier, in a September 2022 speech. He described a shift away from the longstanding “sliding scale” approach of staying a couple of generations ahead in key technologies. Instead, Sullivan argued that given the foundational nature of technologies such as advanced logic and memory chips, the US must maintain as large of a lead as possible.

The implications extend far beyond US-China relations. Export controls are reshaping global semiconductor supply chains, forcing companies to choose sides, accelerating China’s domestic chip development, and creating geopolitical fractures across the technology industry. Three and a half years later, the policy landscape has grown far more complex than the original October 2022 rules — with reversals, expansions, bilateral deals, and a new US administration rewriting the playbook.

---

What the Controls Actually Restrict

The export control regime operates through multiple mechanisms that have evolved significantly since 2022.

Chip Performance Thresholds

BIS established technical thresholds defining which chips cannot be exported to China and other restricted countries. The thresholds target chips with high compute density — the combination of processing performance and interconnect bandwidth that makes a chip useful for AI training.

Key restricted products (as of February 2026):

• NVIDIA H100 — NVIDIA’s flagship AI training GPU, fully banned for China export
• NVIDIA Blackwell series (B100, B200, GB200) — NVIDIA’s next-generation AI chips, fully restricted for China
• NVIDIA H200 — Previously banned, but export to China was authorized in December 2025 under strict conditions (vetted recipients, 25% revenue surcharge paid to the US government). China approved the first major import batch in early 2026, clearing ByteDance, Alibaba, and Tencent for over 400,000 units combined.
• AMD MI300 and MI325X series — AMD’s competing AI accelerators. The MI325X was also cleared for conditional China export alongside the H200.
• Intel Gaudi series — Intel’s AI training chips, restricted

NVIDIA initially designed “China-compliant” chips (the A800 and H800) that fell just below the original performance thresholds. BIS closed this loophole in October 2023 by tightening the thresholds and adding restrictions on performance density. NVIDIA then created the H20 chip specifically for the Chinese market — but even this was caught up in policy turbulence. The Trump administration banned the H20 in April 2025, then reversed course in July 2025, approving licenses for it.

The AI Diffusion Rule — Enacted and Rescinded

In January 2025, the outgoing Biden administration published the “Framework for Artificial Intelligence Diffusion,” which established a global three-tier system for AI chip exports. Tier 1 included the US and 18 close allies (Australia, Canada, Japan, the UK, and others) with few restrictions. Tier 2 covered most other countries, requiring licenses. Tier 3 (China, Russia, and embargoed nations) faced near-total prohibition.

The Trump administration rescinded the AI Diffusion Rule on May 13, 2025, replacing it with a bilateral deal-based approach. Rather than a rules-based framework, the new policy relies on country-by-country negotiations and specific export authorizations — giving the administration more direct leverage but creating less predictability for the industry.

Equipment Controls (Chokepoint Strategy)

The most strategically significant controls target semiconductor manufacturing equipment. The global semiconductor equipment supply chain has extreme concentration:

• ASML (Netherlands) is the sole manufacturer of EUV (Extreme Ultraviolet) lithography machines — required for manufacturing chips at the most advanced nodes. No EUV machine has ever been shipped to China. The informal US-Dutch agreement to block EUV exports dates to 2019 under the first Trump administration, was formalized in 2023, and expanded in January 2025 to also require Dutch government licenses for advanced DUV (Deep Ultraviolet) immersion systems.
• Tokyo Electron, Screen Holdings (Japan) provide critical etching, deposition, and cleaning equipment. Japan agreed to export bans on 23 types of semiconductor equipment to China.
• Applied Materials, Lam Research, KLA (US) provide essential manufacturing and inspection equipment.

This “chokepoint strategy” exploits extreme supply chain concentration: by controlling access to a handful of irreplaceable machines, the US and its allies can effectively cap China’s semiconductor manufacturing capability at generations behind the leading edge. ASML expects China business to drop to around 20% of its total revenue in 2025 (down from higher levels in prior years), reflecting the tightening restrictions.

Foundry Access Controls

TSMC (Taiwan Semiconductor Manufacturing Company) fabricates approximately 90% of the world’s most advanced chips at the 3nm node and below, with a broader foundry market share of around 67%. US controls restrict TSMC from manufacturing advanced chips (below 7nm) for Chinese customers, including Huawei. This cuts off China’s primary route to advanced chips even without domestic manufacturing capability.

---

China’s Response: Self-Sufficiency at Any Cost

China has responded to export controls with the most ambitious semiconductor self-sufficiency push in industrial history.

Huawei and SMIC’s Continued Progress: In August 2023, Huawei launched the Mate 60 Pro smartphone containing a 7nm chip manufactured by SMIC (Semiconductor Manufacturing International Corporation) using older DUV lithography through multi-patterning techniques. Since then, SMIC has continued advancing: the Kirin 9030 processor in Huawei’s Mate 80 Pro Max represents China’s most advanced domestic semiconductor manufacturing to date, using sophisticated Self-Aligned Quadruple Patterning (SAQP) to push DUV-based technology further. Yields remain lower and costs higher than EUV-based production at TSMC or Samsung, but the chips are proving viable for AI and 5G workloads.

Huawei Ascend AI Chips: Huawei plans to double output of its Ascend AI accelerator line in 2026, targeting 1.6 million dies — including approximately 600,000 units of the Ascend 910C. SMIC’s advanced node capacity (7nm and below) is expected to reach 60,000 wafers per month in 2026 and 80,000 in 2027. While these chips do not match NVIDIA’s performance or efficiency, they represent a meaningful step toward AI self-sufficiency. The two major bottlenecks remain advanced fab capacity and HBM (High Bandwidth Memory) supply.

State Investment: China’s “Big Fund III” (National Integrated Circuit Industry Investment Fund Phase III), established in May 2024, raised $47.5 billion — the largest semiconductor investment fund in history. With a 15-year horizon (2024-2039) and 19 state-owned investors led by the Ministry of Finance, the fund targets manufacturing, equipment, materials, and HBM production. Combined with provincial and municipal funds, China is investing well over $100 billion in domestic semiconductor capability.

Talent Acquisition: China continues aggressively recruiting semiconductor engineers globally, particularly from TSMC, Samsung, and Intel, offering substantial salary premiums. Taiwan has enacted laws restricting the recruitment of its semiconductor workforce by Chinese companies.

Alternative Supply Chains: China is developing domestic alternatives across the semiconductor supply chain — wafer fabrication, lithography (SMEE is developing a Chinese EUV-equivalent machine, though it remains years behind ASML), packaging, testing, and EDA design tools. Progress is real, but the technology gap at the leading edge remains significant.

---

The US Reshoring Push: CHIPS Act Results

The US CHIPS and Science Act (2022) allocated $52.7 billion in subsidies for domestic semiconductor manufacturing and R&D. By late 2025, the broader semiconductor investment wave had reached over $630 billion across 140 projects in 28 states, creating 500,000 jobs according to the Semiconductor Industry Association.

TSMC Arizona — Already Producing: TSMC’s Fab 21 Phase 1 in Arizona is fully operational, producing 4nm and 5nm chips at a 92% yield rate — reportedly surpassing TSMC’s flagship fabs in Taiwan by approximately four percentage points. Production started at 10,000 wafers per month and is scaling to 30,000. Phase 2 (3nm) is on track for equipment installation in mid-2026 with production targeted for 2027. Phase 3 (2nm) broke ground in April 2025. TSMC CEO C.C. Wei announced the Arizona site will become a “gigafab” cluster of six fabs, two advanced packaging facilities, and an R&D center, with total investment reaching $165 billion.

Intel 18A Breakthrough: Intel’s Fab 52 in Arizona has entered high-volume manufacturing on the Intel 18A process (1.8nm-class) — the first time a US-based facility has manufactured at this scale below the 2nm threshold. Intel 18A uses RibbonFET gate-all-around transistors and PowerVia backside power delivery. Yields are expected to reach target cost levels by end of 2026. Intel’s “Silicon Heartland” project in Ohio, however, has been delayed, with the first fab now pushed to 2030.

Samsung Taylor, Texas: Samsung’s fab is focused exclusively on 2nm Gate-All-Around (GAA) technology, with mass production pushed to late 2026. Samsung’s CHIPS Act award was reduced from a proposed $6.4 billion to $4.7 billion after due diligence.

Micron: Micron’s memory manufacturing facility in New York broke ground, with its Idaho site expected to begin production next year. The facilities are expected to onshore approximately 40% of Micron’s DRAM production over the next two decades.

---

The Global Fallout

Export controls have created cascading effects across the global technology industry.

NVIDIA’s Revenue Whiplash

NVIDIA derived 20-25% of its data center revenue from China before the restrictions took full effect. CEO Jensen Huang stated that NVIDIA’s China market share fell from 95% to effectively zero. By mid-2025, NVIDIA was forecasting zero China revenue in its outlook.

The December 2025 policy reversal — allowing H200 exports under strict conditions and a 25% government surcharge — partially reopened the market. NVIDIA is preparing shipments and plans to bring additional China production capacity online, taking orders from Q2 2026. But the Blackwell generation remains restricted, and the policy instability itself has pushed Chinese customers toward domestic alternatives, particularly Huawei’s Ascend AI chips.

Gulf States: From Restricted to Preferred Customers

Saudi Arabia and the UAE were initially caught in export control crossfire. Under Biden-era rules, they were limited to approximately 1,700 advanced chips annually. The Trump administration dramatically reversed course: in November 2025, the Commerce Department authorized exports to Saudi Arabia’s HUMAIN and the UAE’s G42, both state-backed AI ventures. Saudi Arabia’s HUMAIN purchased 18,000 NVIDIA GB300 chips to build up to 500MW of data centers, with several hundred thousand more expected over five years. The strategic logic is explicit: ensure the Gulf builds its AI infrastructure on a US-centric tech stack rather than turning to Chinese alternatives.

Middle Powers Navigate the Split

Countries that are neither close US allies nor Chinese partners face difficult choices:

• Singapore, Malaysia, India: These countries host semiconductor manufacturing and assembly operations serving both US and Chinese customers. Export controls create compliance burdens and force companies to segregate supply chains.
• Southeast Asian nations: Vietnam, Thailand, and Malaysia have attracted semiconductor investment from companies diversifying away from China but must navigate export control compliance to maintain US technology access.
• Enforcement challenges: Banned NVIDIA GPUs (A100, H100) have been found in Chinese university research and startup documentation, highlighting the difficulty of enforcing chip-level trade restrictions across complex global supply chains.

The Innovation Dilemma

Export controls create a strategic paradox: by cutting off China’s access to the most advanced chips, they simultaneously reduce the revenue that US chip companies use to fund R&D and incentivize China to develop alternatives that could eventually compete. If China succeeds in building a fully independent semiconductor supply chain — even if it takes a decade — the US will have created a competitor that no longer depends on American technology. The partial reopening of H200 exports reflects this tension: a recognition that total restriction may be counterproductive if it accelerates Chinese self-sufficiency while starving US companies of revenue.

---

The Semiconductor Market: Approaching One Trillion Dollars

The global semiconductor market continues its extraordinary growth trajectory, driven primarily by AI infrastructure demand. The World Semiconductor Trade Statistics (WSTS) projects the market will reach approximately $975 billion in 2026 — approaching the one-trillion-dollar milestone — representing 26% growth over 2025’s estimated $772 billion. Logic chips (37% growth) and memory chips (28% growth) are the primary drivers, both fueled by AI and data center buildout. Global semiconductor equipment sales are projected to reach a record $156 billion by 2027.

This growth makes the stakes of export controls even higher: control over who can build, buy, and deploy the most advanced chips is control over the commanding heights of the global economy.

---

TSMC and the Taiwan Question

The centrality of TSMC to the entire export control regime underscores the geopolitical significance of Taiwan. TSMC manufactures approximately 90% of the world’s most advanced chips at the 3nm node and below, with over 90% market share at 3nm specifically. A Chinese military action against Taiwan would not only be a humanitarian and geopolitical crisis — it would trigger a global semiconductor disruption of unprecedented proportions.

This reality has driven the geographic diversification now underway: TSMC’s Arizona gigafab, its fabs in Kumamoto (Japan) and Dresden (Germany), and the broader CHIPS Act investment wave all aim to ensure that advanced chip manufacturing exists outside Taiwan. But even with these investments, Taiwan will remain the center of gravity for leading-edge semiconductor production for years to come.

---

---

Sources

• Bureau of Industry and Security — Export Administration Regulations
• BIS — October 2022 China Semiconductor Controls
• ASML — Annual Report and Export Control Compliance
• TSMC — Geographic Diversification Strategy
• Chris Miller — Chip War (2022)
• SIA — Semiconductor Supply Chain Investments
• CSIS — China’s Semiconductor Industry
• US CHIPS and Science Act
• WSTS — Global Semiconductor Market Approaches $1T in 2026
• Council on Foreign Relations — China’s AI Chip Deficit
• Tom’s Hardware — NVIDIA H200 Export Developments
• Deloitte — 2026 Global Semiconductor Industry Outlook