L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 7 · Lesson 1

The Architecture of Denial: EAR and the Entity List

How the United States built a regulatory framework to slow adversaries' access to the chips that power modern AI
What legal machinery allows Washington to decide which foreign companies can buy American-designed semiconductors — and what happened when it was weaponized against Huawei?

On May 15, 2019, the U.S. Commerce Department added Huawei Technologies and 68 of its affiliates to the Entity List. Within 90 days, Taiwan Semiconductor Manufacturing Company — which fabricated many of Huawei's custom Kirin chips — stopped accepting new orders from the company. By 2021, Huawei's smartphone shipments had collapsed from a peak of 240 million units annually to fewer than 30 million. The instrument that accomplished this was not a military embargo or a UN sanction. It was an administrative rulemaking under the Export Administration Regulations.

The Export Administration Regulations

The Export Administration Regulations (EAR), administered by the Commerce Department's Bureau of Industry and Security (BIS), govern the export of dual-use goods — items that have both civilian and military applications. Semiconductors and the software and technology to design them fall squarely within EAR's scope.

The key structural concept is the Export Control Classification Number (ECCN). Every controlled item receives an ECCN that specifies what export license is needed and to which countries. For advanced semiconductors relevant to AI — including items classified under ECCN 3A090 (high-performance integrated circuits) and 3E001 (technology for their design) — licenses are generally required for exports to countries of concern.

BIS's Entity List is a published register of foreign persons, companies, and organizations that have been determined to act contrary to U.S. national security or foreign policy interests. Once listed, a company cannot receive items subject to EAR from any U.S.-origin supplier without a specific license — which is presumed denied.

The Foreign Direct Product Rule — The Extraterritorial Reach

EAR's normal jurisdiction covers U.S.-origin goods. But the Foreign Direct Product (FDP) Rule extends U.S. jurisdiction to foreign-made products that are the direct product of U.S. technology or software. In May 2020, Commerce amended the FDP Rule specifically to capture Huawei: any chip fabricated anywhere in the world using U.S. semiconductor equipment or EDA (electronic design automation) software required a U.S. export license before being sold to Huawei. Because virtually all advanced chipmakers — TSMC, Samsung, SMIC — rely on U.S. equipment and EDA tools (from companies like Lam Research, Applied Materials, Cadence, and Synopsys), the amended rule effectively subjected global semiconductor supply chains to U.S. jurisdiction.

Key Terms
EARExport Administration Regulations — the primary U.S. framework governing export of dual-use goods including semiconductors and related technology.
Entity ListBIS's register of foreign parties subject to license requirements and a presumption of denial; added by administrative order, not legislation.
ECCNExport Control Classification Number — a five-character alphanumeric code that identifies items and determines applicable license requirements.
FDP RuleForeign Direct Product Rule — extends EAR jurisdiction to foreign-manufactured goods produced using U.S. technology or software.
De minimis thresholdThe percentage of U.S.-controlled content in a foreign product below which EAR jurisdiction does not apply; typically 25%, but reduced to 0% for Huawei-related items.
The Huawei Case in Detail

Before its Entity List designation, Huawei was the world's largest telecommunications equipment vendor and a major smartphone manufacturer. Its HiSilicon subsidiary designed Kirin system-on-chip processors competitive with Qualcomm's Snapdragon series — but all fabrication occurred at TSMC's facilities in Taiwan using equipment from Applied Materials, ASML, Lam Research, and KLA, all subject to U.S. export controls.

The May 2019 designation immediately required TSMC and other suppliers to apply for licenses to continue serving Huawei. In September 2020, after the amended FDP Rule closed the remaining design-house loopholes, TSMC publicly confirmed it would stop all new Huawei orders. MediaTek of Taiwan, which had planned to continue supplying Huawei's non-5G handsets, also announced it would seek licenses — effectively halting supply.

The economic impact was rapid and severe. Huawei's consumer business group revenue fell from ¥467 billion in 2020 to ¥243 billion in 2021. In 2021 Huawei sold its Honor smartphone brand to a consortium of Chinese state and private entities specifically to allow Honor to continue purchasing chips that Huawei itself could not obtain.

Limits of the Tool

The Entity List designation did not eliminate Huawei from advanced technology. By August 2023, Huawei unveiled the Mate 60 Pro smartphone containing a Kirin 9000s chip fabricated by SMIC at a 7-nanometer process node. SMIC itself was added to the Entity List in December 2020, but the chip apparently used older U.S. equipment already present in SMIC fabs before the designation — illustrating the temporal limits of export control enforcement. The incident triggered a formal review by BIS and Congressional demands for tighter controls on chipmaking equipment.

Lesson 1 Quiz

The Architecture of Denial: EAR and the Entity List
1. Which U.S. government agency administers the Export Administration Regulations and maintains the Entity List?
Correct. BIS within the Commerce Department administers EAR and maintains the Entity List.
Not quite. The Entity List is a BIS/Commerce instrument, distinct from OFAC (Treasury) or DDTC (State) sanctions lists.
2. What was the primary legal mechanism that allowed the U.S. to prevent TSMC from continuing to fabricate chips for Huawei, even though TSMC is a Taiwanese company?
Correct. The amended FDP Rule's extraterritorial reach captured TSMC because it uses U.S. semiconductor equipment and EDA software.
Not quite. The FDP Rule is the key mechanism — it extends U.S. EAR jurisdiction to foreign manufacturers that use U.S.-origin equipment or software.
3. The Huawei Mate 60 Pro's Kirin 9000s chip, revealed in August 2023, demonstrated which limitation of export controls?
Correct. SMIC used legacy U.S. equipment already in its fabs before its December 2020 Entity List designation to achieve 7nm fabrication.
Not quite. The key limitation illustrated is temporal: equipment already in-country before controls tighten can continue producing chips, even from listed entities.

Lab 1: Entity List Mechanics

Explore how the EAR Entity List works through guided conversation

Your Task

You are a trade compliance analyst at a semiconductor equipment company. A senior engineer wants to understand why the company's legal team rejected a large order from a Chinese customer. Walk through the EAR and Entity List framework with the AI tutor.

Suggested opening: "Our legal team flagged an order from a Chinese chip foundry. Can you explain what the Entity List actually is and how we determine whether we can legally fulfill the order?"
AI Tutor — Export Controls Advisor EAR / Entity List
Welcome. I'm your export controls advisor for this session. Ask me anything about the EAR, the Entity List, or how to evaluate whether a specific transaction involving semiconductor equipment requires a license or is prohibited outright.
Module 7 · Lesson 2

October 2022: The Comprehensive Chip Controls

How the Biden administration's landmark rules attempted to cut China off from advanced AI semiconductors and the equipment to make them
What made the October 7, 2022 export control package historically unprecedented — and which specific thresholds defined the new frontier of permissible technology?

On October 7, 2022, the Bureau of Industry and Security published an interim final rule that represented the most sweeping expansion of U.S. semiconductor export controls since the Cold War. The rules targeted not just specific companies but entire categories of capability — establishing performance thresholds below which chips could be freely exported and above which licenses would be presumptively denied. They also imposed a requirement that would prove immediately explosive: U.S. persons working anywhere in the world in advanced Chinese semiconductor production must seek a license — or leave their jobs.

The Three Pillars of October 7

Pillar 1: Advanced Chip Export Restrictions. The rules created new ECCN categories targeting chips with computational performance above defined thresholds used for AI training and inference. For chips with interconnect bandwidth density above 600 Gbps/mm², or with total processing performance above approximately 4,800 TOPS (Tera Operations Per Second) at INT8 precision, Commerce established a license requirement for export to China. This captured Nvidia's A100 and H100 data center GPUs, which were the primary tools for large-scale AI model training.

Pillar 2: Chipmaking Equipment Controls. The rules added new controls on semiconductor manufacturing equipment capable of producing chips at advanced nodes — specifically 16nm logic and below, 128-layer NAND flash, and certain DRAM process nodes. Key tools affected included deposition equipment, etch systems, and metrology tools from Applied Materials, Lam Research, and KLA. These companies were required to immediately halt shipments to Chinese fabs — even before receiving formal license denials.

Pillar 3: The U.S. Person Rule. Perhaps the most novel element: U.S. citizens and permanent residents working in China's advanced chip sector were prohibited from supporting the development, production, or use of advanced ICs without obtaining a BIS license. Within days of the rule's publication, dozens of American engineers and executives working at Chinese chip companies, including SMIC and YMTC, submitted resignations to comply with the requirement.

The Performance Thresholds

The rules set specific numerical thresholds defining "advanced" chips subject to new controls. The key initial thresholds (later tightened in October 2023) were: logic chips with non-planar transistor architecture at 16nm or below; chips with total processing performance (TPP) above approximately 4,800 TOPS at INT8, or 300 TOPS at FP8/FP16; and chips with performance density above 4.6 TOPS/mm² at INT8. These metrics were chosen to capture the specific chips used for large AI model training while leaving consumer and automotive chips unrestricted.

Nvidia's Adaptive Response and BIS's Counter-Response

Within months, Nvidia announced China-specific chip variants — the A800 and H800 — engineered to fall just below the October 7 thresholds by reducing interconnect bandwidth. BIS had set the NVLink bandwidth threshold at 600 Gbps/mm²; Nvidia's China variants reduced this below the threshold while maintaining much of the computational throughput.

In October 2023, BIS tightened the rules substantially. A new performance density threshold of 4,800 TOPS combined with 40 TOPS/mm² captured the A800 and H800. BIS also added new restrictions on chips above 4,800 TOPS total regardless of interconnect bandwidth — closing the workaround. The result was another round of Nvidia engineering: the H20, L20, and L2 chips released for China in 2024, substantially less capable than global variants.

The October 2023 rules also extended the country scope: chip controls that previously applied to China were extended to approximately 40 additional countries deemed proliferation risks, while creating a tiered system allowing allies in Tier 1 (NATO, Japan, South Korea, Australia) largely unrestricted access.

Allied Coordination: Japan and the Netherlands

Unilateral U.S. controls faced an inherent problem: if Japan's Tokyo Electron or the Netherlands' ASML continued selling advanced chipmaking equipment to China freely, U.S. equipment restrictions would simply be circumvented. In January 2023, after months of quiet negotiations, Japan and the Netherlands agreed to align their export controls with U.S. restrictions on advanced chipmaking equipment.

The Netherlands specifically restricted ASML from exporting its deep ultraviolet (DUV) immersion lithography systems — used for chip nodes from 28nm to 5nm — to China without a Dutch export license. ASML had already been barred from exporting its extreme ultraviolet (EUV) systems to China since 2019. The combined DUV restriction was significant because Chinese fabs had been purchasing DUV equipment rapidly to build out capacity that, while not at the most advanced nodes, could produce competitive chips through multi-patterning techniques.

What the Controls Cannot Do

The October 2022 rules and their successors represent the most aggressive use of export controls as industrial policy in the semiconductor era. But analysts at CSET (Center for Security and Emerging Technology) and elsewhere have noted fundamental limits: controls slow but rarely stop determined state-level technology development; China retains substantial legacy chipmaking capacity sufficient for many AI inference applications; and the growing Chinese domestic EDA and equipment industry — while years behind — continues to advance. SMIC's production of 7nm-class chips in 2023 using only equipment obtained before tighter controls underscored that enforcement gaps persist.

Lesson 2 Quiz

October 2022: The Comprehensive Chip Controls
1. Which of the following was a novel element of the October 7, 2022 rules that had not appeared in previous U.S. semiconductor export controls?
Correct. The U.S. person rule was unprecedented — imposing license requirements on the activities of American citizens globally, not just on exports from U.S. territory.
Not quite. The genuinely novel element was the "U.S. person" provision requiring American nationals working in China's advanced chip sector to get a license or resign.
2. Nvidia's A800 and H800 chips were designed specifically for China by reducing which parameter to fall below the October 7, 2022 threshold?
Correct. Nvidia reduced the NVLink bandwidth below 600 Gbps/mm² in the A800 and H800 to comply with the October 2022 threshold, while preserving much raw compute performance.
Not quite. The A800/H800 were engineered to fall below the interconnect bandwidth density threshold — a parameter BIS introduced specifically to capture high-performance AI training chips.
3. Why was the Netherlands' January 2023 agreement to restrict ASML DUV exports to China significant — even though EUV systems were already restricted?
Correct. Multi-patterning allows DUV tools to reach nodes previously thought to require EUV, so unrestricted DUV access would have provided a significant workaround.
Not quite. The key concern was that DUV equipment with multi-patterning techniques could reach advanced chip nodes, making it a viable path to advanced production even without EUV access.

Lab 2: Analyzing the October 2022 Rules

Stress-test the October 7 framework through policy analysis dialogue

Your Task

You are a policy analyst briefing a Senate staff member on the October 7, 2022 export control package. The staffer has questions about what the rules actually accomplish, what they cannot do, and whether allied coordination was sufficient. Engage the AI tutor to deepen your analysis.

Suggested opening: "The October 2022 rules set specific chip performance thresholds. How did Nvidia respond, and why did BIS have to tighten the rules again in 2023?"
AI Tutor — Policy Analysis Advisor Oct 2022 Controls
Ready to dig into the October 7, 2022 rules. I can walk you through the performance thresholds, the U.S. person rule, how Nvidia responded with the A800 and H800, what happened in October 2023, and the role Japan and the Netherlands played. What aspect would you like to analyze first?
Module 7 · Lesson 3

China's Countermoves and Domestic Substitution

How Beijing responded to semiconductor export controls with industrial policy, investment mobilization, and accelerated domestic development
What concrete investments and policy instruments did China deploy in response to U.S. chip controls — and which areas showed genuine progress versus persistent gaps?

The October 2022 rules prompted Beijing to accelerate programs already underway. The China Integrated Circuit Industry Investment Fund — known as the "Big Fund" — had been established in 2014 with an initial capitalization of ¥138.7 billion. A second phase raised an additional ¥200 billion in 2019. In May 2024, the government announced a third phase: ¥344 billion ($47.5 billion) — the largest single investment in the fund's history — focused specifically on semiconductor equipment, materials, and EDA software: precisely the areas where U.S. export controls had proven most damaging.

The Big Fund and Industrial Policy

China's semiconductor industrial policy rests on several pillars. The National Integrated Circuit Industry Investment Fund (NICICIF, or "Big Fund") provides equity investment and low-cost capital to domestic chipmakers, equipment companies, and materials suppliers. The "Made in China 2025" initiative (launched 2015) set targets for domestic semiconductor content in Chinese electronics — 40% by 2020 and 70% by 2025 — targets that have not been met but that have concentrated industrial attention and public investment.

The Sci-Tech Innovation Board (STAR Market), launched in 2019 on the Shanghai Stock Exchange, was designed specifically to provide IPO access for technology-intensive companies that might not meet traditional profitability requirements. By 2023, dozens of semiconductor companies — including Cambricon (AI chips), Horizon Robotics (edge AI), and SMIC's suppliers — had listed on STAR Market, accessing public capital markets for chip development.

Documented Progress: Where China Has Gained Ground

Memory Chips: YMTC (Yangtze Memory Technologies Corporation) developed its 232-layer 3D NAND flash memory — competitive with Samsung and Micron's leading products — by 2022. YMTC was added to the Entity List in December 2022, but had already built out substantial production capacity. The technology demonstrated that Chinese firms could reach world-class memory specifications, though production volume and yield remained behind leading competitors.

AI Inference Chips: Huawei's Ascend 910B AI accelerator, produced by SMIC at a 7nm-class process, demonstrated competitive AI training performance when benchmarked against older Nvidia A100 chips. Chinese cloud providers including Baidu, Alibaba, and ByteDance have purchased Ascend 910B units as a domestic alternative to restricted Nvidia products. Baidu's ERNIE Bot and several other large language models were trained or fine-tuned on Ascend hardware.

EDA Software: Domestic EDA companies including Empyrean Technology (华大九天) and Primarius Technologies have developed point tools competitive for specific design tasks, particularly analog and mixed-signal design. Full-flow digital EDA — the complete design toolchain needed for advanced logic chips — remains dominated by the U.S. trio of Cadence, Synopsys, and Mentor (Siemens EDA), with domestic Chinese alternatives years behind.

The SMIC 7nm Revelation — August 2023

When TechInsights performed a teardown of the Huawei Mate 60 Pro in August 2023 and identified its Kirin 9000s chip as fabricated at SMIC on a 7nm-class process, the finding carried multiple implications. It showed SMIC had achieved advanced-node production — something the U.S. intelligence community believed was 5–7 years away. But it also showed the process likely used ASML DUV immersion tools purchased before controls tightened, potentially with yields substantially lower than TSMC or Samsung. High-volume production at sub-7nm nodes without EUV remains an open question with significant cost and yield implications.

Persistent Structural Gaps

Despite investment and progress in select areas, China faces structural semiconductor gaps that industrial policy has not yet bridged. Semiconductor equipment remains the most critical gap: China has no domestic equivalent of ASML, no viable EUV lithography program, and limited capability in advanced etch, deposition, and inspection equipment. Domestic alternatives to Applied Materials, Lam Research, and KLA exist for trailing-edge applications but cannot replicate the capability needed for advanced nodes.

Advanced packaging is an area where China has made more substantial progress. JCET, Tongfu Microelectronics, and others have developed advanced packaging capabilities including fan-out wafer-level packaging and 2.5D interposer technology. Huawei's Mate 60 Pro chip used advanced packaging to achieve its performance, partially compensating for process node limitations. But CoWoS-class HBM packaging — required for Nvidia H100-class performance — remains largely outside Chinese production capability.

Export Control Evasion: The Third-Country Problem

Multiple documented cases emerged after October 2022 of restricted chips reaching Chinese end-users through intermediaries in third countries. Reuters investigations in 2023 found Nvidia A100 chips appearing in Chinese research institutions after being routed through distributors in Singapore, Malaysia, and Taiwan. BIS responded with enhanced end-use verification requirements and a new "red flag" guidance emphasizing that U.S. exporters must investigate suspicious re-export patterns. The challenge illustrates why unilateral controls have inherent enforcement limits: a chip that legally enters Singapore can be diverted onward in ways difficult to detect without active enforcement cooperation from Singapore authorities.

Lesson 3 Quiz

China's Countermoves and Domestic Substitution
1. China's "Big Fund" Phase 3, announced in May 2024 with ¥344 billion ($47.5 billion), was specifically focused on which areas most affected by U.S. export controls?
Correct. Big Fund Phase 3 specifically targeted the upstream elements of the chip supply chain — equipment, materials, and EDA — where U.S. controls had created the sharpest bottlenecks.
Not quite. Big Fund Phase 3 targeted semiconductor equipment, materials, and EDA software — the exact nodes in the supply chain where export controls had been most effective.
2. Which Chinese company's teardown in August 2023 revealed that SMIC had achieved 7nm-class fabrication, and what chip was it?
Correct. TechInsights' teardown of the Huawei Mate 60 Pro identified the Kirin 9000s as a SMIC 7nm-class product, a development that surprised U.S. intelligence assessments.
Not quite. The revelation came from TechInsights' teardown of the Huawei Mate 60 Pro and its Kirin 9000s chip — fabricated at SMIC.
3. What makes third-country chip re-export a difficult enforcement problem for BIS, based on documented cases after October 2022?
Correct. Reuters documented A100 chips appearing in Chinese institutions after legal import into Singapore — diversion that requires active third-country enforcement cooperation to detect and prevent.
Not quite. The core problem is that once chips legally enter a friendly third country, onward diversion is difficult to detect without active enforcement cooperation from that country's authorities.

Lab 3: China's Semiconductor Strategy

Evaluate Beijing's industrial policy responses through structured analysis

Your Task

You are a strategic analyst at a think tank preparing a report on whether China's domestic semiconductor substitution strategy is working. Use the AI tutor to stress-test your assessments about where China has made genuine progress and where structural gaps remain.

Suggested opening: "My report argues that China has made real progress in AI inference chips and memory, but still faces critical gaps in chipmaking equipment. Am I reading the evidence correctly?"
AI Tutor — Semiconductor Strategy Analyst China Industrial Policy
Let's evaluate China's semiconductor substitution strategy together. I can discuss the Big Fund phases, YMTC's memory progress, Huawei's Ascend chips and SMIC's 7nm achievement, the EDA gap, equipment self-sufficiency challenges, and the third-country re-export problem. What's your opening thesis?
Module 7 · Lesson 4

The Diffusion Rule, Ally Tensions, and Strategic Limits

How Washington's attempt to extend chip controls globally fractured allied relationships and exposed the boundaries of export control as geopolitical strategy
When the U.S. moved to restrict AI chip exports to U.S. allies and imposed its framework on trading partners, what political and commercial costs emerged — and what does this reveal about the limits of technology denial as strategy?

On January 13, 2025, the Biden administration published its final major semiconductor export control action: the Framework for Artificial Intelligence Diffusion. The rule created a three-tier system governing global AI chip exports — Tier 1 allies with near-unrestricted access, Tier 2 countries with per-entity caps on how many advanced AI chips they could import annually, and Tier 3 adversaries subject to existing strict controls. The response from affected countries was immediate and, for Washington, unexpectedly sharp. Singapore, Malaysia, India, Mexico, and other Tier 2 nations objected publicly. Semiconductor companies warned of competitive damage to U.S. exports. The incoming Trump administration indicated it would revise the framework before implementation.

The AI Diffusion Rule: Architecture and Controversy

The Framework for AI Diffusion established the most geographically comprehensive AI chip control regime attempted by any government. Under the rule, Tier 1 comprised 18 close allies — including the UK, EU members, Japan, South Korea, Australia, Canada — and faced essentially no new restrictions on AI chip imports. Tier 2 covered most of the rest of the world, including Singapore, India, Israel, UAE, Mexico, Brazil, and dozens of others. Tier 2 countries faced a new national cap: their companies could import advanced AI chips up to a certain limit (roughly 1,700 H100-equivalent chips per company per year without a license, with higher limits subject to enhanced verification requirements). Tier 3 — China, Russia, and other adversaries — remained subject to the existing strict controls.

The rule's stated rationale was preventing AI capability diffusion — the concern that large data centers in relatively permissive Tier 2 countries could be used to provide AI compute services to restricted Tier 3 users, effectively circumventing the chip-level controls through cloud access rather than physical export.

Allied and Partner Objections

Singapore's Ministry of Trade and Industry stated it was "deeply concerned" about the rule and would raise objections through diplomatic channels. The Singapore government noted that its semiconductor and data center sector was a significant portion of its economy, and that being treated similarly to less-aligned nations was both economically damaging and diplomatically offensive. Israel, a close security partner of the U.S., expressed similar concerns about finding itself in Tier 2 alongside countries with which it had fundamental policy differences. India's government and technology industry objected to caps that would limit data center expansion at a moment of rapid growth. The Semiconductor Industry Association noted that Tier 2 restrictions would harm U.S. chip exports to major markets without clear national security justification.

Ally Coordination: The AUKUS and Multilateral Export Control Challenge

Beyond the diffusion rule controversy, U.S. policymakers faced a recurring structural challenge: getting allied nations to maintain comparable export control regimes requires ongoing diplomatic investment and creates commercial friction that tests alliance durability. Japan's January 2023 agreement to restrict chipmaking equipment exports was significant but came after months of negotiations and required Japan to accept costs — lost revenue for Tokyo Electron, a leading equipment maker — in exchange for U.S. commitments on other alliance dimensions.

The Wassenaar Arrangement, the multilateral regime governing dual-use export controls, proved inadequate to the speed of the U.S. AI chip control agenda. Wassenaar decisions require consensus among 42 member states and typically take years to implement. The U.S. moved on months-long timescales with bilateral agreements, essentially bypassing the multilateral framework that had governed dual-use export controls since the Cold War's end. This bilateral approach was faster but created a less durable and consistent framework than multilateral consensus would provide.

The CHIPS Act: Controls and Incentives Together

The CHIPS and Science Act, signed August 9, 2022 — just two months before the October 7 export control rules — provided the complementary instrument to denial: domestic investment. The law appropriated $52.7 billion for U.S. semiconductor manufacturing and research, including $39 billion in manufacturing incentives and $13.2 billion for R&D. Recipients of CHIPS Act manufacturing funding are prohibited by law from expanding advanced semiconductor manufacturing capacity in countries of concern for 10 years — a provision known as the "guardrails."

TSMC's planned Arizona fabs (TSMC Arizona), Intel's Ohio and Arizona expansion, Samsung's Texas fab, and Micron's Idaho and New York DRAM facilities were all announced or accelerated in the context of CHIPS Act incentives. The combination of export controls (restricting technology to adversaries) and CHIPS Act investment (building domestic capacity) reflected a coherent if contested industrial policy strategy.

Strategic Limits: What Export Controls Can and Cannot Achieve

A growing body of analysis from institutions including CSET, RAND, and the Brookings Institution has examined what export controls have demonstrably accomplished and where their limits lie. Controls have slowed Chinese access to the most advanced AI training chips — the H100-class systems. They have created measurable cost and capability gaps for Chinese AI researchers using domestic alternatives. The forced resignation of American engineers from Chinese chip companies represented a genuine talent drain.

But controls have not stopped Chinese large language model development: DeepSeek's R1 model, published in January 2025 and trained at a fraction of the cost of comparable U.S. models, demonstrated that algorithmic efficiency improvements can partially compensate for compute restrictions. Controls have not prevented China from achieving 7nm fabrication at SMIC. They have not eliminated the re-export problem. And they have created political costs with allies and trading partners that complicate broader technology governance cooperation.

The DeepSeek Moment — January 2025

When Chinese AI lab DeepSeek released its R1 reasoning model in January 2025, achieving benchmark scores competitive with leading U.S. models at dramatically lower training costs, it prompted a reexamination of export control strategy in Washington. If algorithmic innovation could partially substitute for raw compute — through techniques like mixture-of-experts architecture, knowledge distillation, and reinforcement learning from human feedback — then chip-level controls might be necessary but not sufficient to maintain AI leadership. The model was reportedly trained using approximately 2,048 H800 chips (the China-market variant just below the October 2022 thresholds) — a cluster size well within what Chinese labs could legally assemble before the October 2023 tightening. The episode illustrated that export controls set ceilings on hardware access but cannot constrain algorithmic ingenuity.

Lesson 4 Quiz

The Diffusion Rule, Ally Tensions, and Strategic Limits
1. The AI Diffusion Framework published in January 2025 placed which types of countries in "Tier 2" — subject to per-entity caps on advanced AI chip imports?
Correct. Tier 2 included major U.S. trading partners and security partners — Singapore, India, Israel, UAE — which generated significant diplomatic blowback.
Not quite. Tier 2 was a broad category including many U.S. partners and allies that objected to being placed below Tier 1 nations in the framework.
2. What does the Wassenaar Arrangement represent, and why was it inadequate for the speed of U.S. AI chip control efforts?
Correct. Wassenaar's consensus requirement among 42 members makes it too slow for the fast-moving semiconductor technology control agenda, pushing the U.S. toward bilateral agreements.
Not quite. Wassenaar is the multilateral dual-use export control framework with 42 members — its consensus requirement made it too slow, pushing the U.S. to negotiate bilateral agreements with Japan and the Netherlands instead.
3. What did DeepSeek's R1 model release in January 2025 demonstrate about the limits of semiconductor export controls?
Correct. DeepSeek R1 achieved competitive performance using legally available H800 chips — demonstrating that algorithmic innovation can partially offset hardware restrictions.
Not quite. DeepSeek showed that algorithmic ingenuity — mixture-of-experts, efficient training — can compensate for hardware limitations, meaning chip controls are necessary but not sufficient to maintain AI leadership.

Lab 4: Strategic Limits of Export Controls

Debate the effectiveness and costs of technology denial as geopolitical strategy

Your Task

You are preparing testimony for a Congressional hearing on whether U.S. semiconductor export controls are achieving their stated national security goals. The AI tutor will challenge your arguments and push you to engage with evidence for and against the controls' effectiveness.

Suggested opening: "I want to argue that export controls have been effective but face serious limits — the DeepSeek case and the SMIC 7nm achievement both suggest we're overestimating what chip restrictions can accomplish. Help me stress-test this argument."
AI Tutor — Strategic Policy Advisor Export Control Strategy
Ready to debate the strategic effectiveness of U.S. semiconductor export controls. I can engage with the evidence on both sides — what the controls have demonstrably accomplished (talent drain, cost imposition, slowing advanced training chip access), what they haven't stopped (DeepSeek, SMIC 7nm, re-export), the ally costs of the diffusion rule, and the question of whether algorithmic efficiency undermines hardware-level control strategies. What's your opening position?

Module 7 Test

The Export Control Dimension — 15 questions · 80% to pass
1. The Export Administration Regulations (EAR) govern the export of which category of goods?
Correct. EAR governs dual-use goods — items with both commercial and potential military applications, including semiconductors.
Incorrect. EAR covers dual-use goods. Purely military items fall under ITAR (International Traffic in Arms Regulations), a separate regime.
2. Which U.S. government bureau administers the Entity List?
Correct. BIS within the Department of Commerce administers both EAR and the Entity List.
Incorrect. BIS (Bureau of Industry and Security) maintains the Entity List. OFAC maintains the SDN list, DDTC administers ITAR.
3. The Foreign Direct Product Rule extends U.S. export control jurisdiction to foreign-manufactured products under what condition?
Correct. The FDP Rule captures foreign products made using U.S. technology or software, giving U.S. controls extraterritorial reach.
Incorrect. The FDP Rule applies when the foreign product is the direct product of U.S. technology or software — regardless of where it's physically made or the component content ratio.
4. When was Huawei added to the BIS Entity List, effectively beginning the campaign to cut it off from advanced chip supply?
Correct. Huawei and 68 affiliates were added to the Entity List on May 15, 2019.
Incorrect. Huawei was added to the Entity List in May 2019. SMIC was added in December 2020. The broader chip rules came in October 2022.
5. What was TSMC's response after the amended Foreign Direct Product Rule was applied to Huawei in September 2020?
Correct. TSMC announced it would stop all new Huawei orders after the amended FDP Rule closed remaining workarounds.
Incorrect. TSMC stopped accepting new Huawei orders — because all its fabrication relies on U.S. equipment and software, the FDP Rule gave it no viable path to continue.
6. The October 7, 2022 export control rules' "U.S. person" provision required what of American citizens working in China's advanced chip sector?
Correct. The rule required U.S. persons to get a BIS license to continue work in China's advanced chip sector — effectively forcing many to resign.
Incorrect. The U.S. person rule required obtaining a BIS license (presumptively denied) to continue supporting advanced Chinese chip production — causing many American engineers to resign.
7. Nvidia's A800 and H800 chips, designed for the Chinese market after October 2022, were engineered specifically to fall below which threshold?
Correct. Nvidia reduced NVLink interconnect bandwidth density below 600 Gbps/mm² in the A800 and H800 to comply with October 2022 thresholds.
Incorrect. The A800/H800 were specifically engineered to stay below the 600 Gbps/mm² interconnect bandwidth threshold — the parameter BIS chose to define AI training chip performance.
8. Which two countries, besides the U.S., agreed in January 2023 to align semiconductor equipment export controls with the U.S. framework?
Correct. Japan (home to Tokyo Electron) and the Netherlands (home to ASML) were the critical allies whose cooperation was needed to make equipment controls effective.
Incorrect. Japan and the Netherlands were the key partners — Tokyo Electron and ASML are the dominant non-U.S. chipmaking equipment suppliers whose cooperation was essential.
9. YMTC achieved what milestone in memory chip technology that demonstrated Chinese competitiveness, before its Entity List designation in December 2022?
Correct. YMTC's 232-layer 3D NAND flash matched the layer counts of Samsung and Micron's leading products — a genuine technical achievement.
Incorrect. YMTC developed 232-layer 3D NAND flash memory, matching world-class competitors. It was added to the Entity List in December 2022 after demonstrating this capability.
10. China's "Big Fund" Phase 3, announced in May 2024, raised approximately how much capital for semiconductor investment?
Correct. Phase 3 of the NICICIF raised ¥344 billion ($47.5 billion) — the largest single investment in the fund's history, focused on equipment, materials, and EDA.
Incorrect. Phase 3 was ¥344 billion, approximately $47.5 billion — dwarfing the previous phases and the U.S. CHIPS Act's $52.7 billion in scale relative to the targeted sectors.
11. The CHIPS and Science Act signed in August 2022 included "guardrail" provisions that prohibited recipients of CHIPS Act manufacturing funds from doing what?
Correct. CHIPS Act guardrails prohibit recipients from expanding advanced chip manufacturing in countries of concern (primarily China) for a decade.
Incorrect. The guardrail provision prohibits CHIPS Act recipients from expanding advanced semiconductor capacity in countries of concern — primarily China — for 10 years.
12. The AI Diffusion Framework published in January 2025 created a three-tier system. What was the stated rationale for imposing caps on Tier 2 countries?
Correct. The diffusion rule aimed to prevent Tier 3 adversaries from accessing AI compute through cloud services hosted in Tier 2 countries.
Incorrect. The stated rationale was preventing "capability diffusion" — the concern that Tier 2 data centers could give Tier 3 users effective AI access through cloud compute, bypassing physical chip controls.
13. Which of the following best describes the Wassenaar Arrangement's limitation in the context of AI chip controls?
Correct. Wassenaar's consensus among 42 members operates on multi-year timescales — incompatible with the rapid pace of AI chip technology and policy.
Incorrect. Wassenaar's consensus requirement among 42 member states makes it too slow for AI-era export controls, pushing the U.S. to negotiate bilateral agreements with Japan and the Netherlands instead.
14. DeepSeek's R1 model, released in January 2025, was reportedly trained on approximately how many chips — and which chip type?
Correct. DeepSeek R1 used ~2,048 H800 chips — legally available at the time of training, demonstrating that efficiency gains can partially offset hardware restrictions.
Incorrect. DeepSeek reportedly trained R1 on ~2,048 Nvidia H800 chips — the legal China-market variant — demonstrating that algorithmic efficiency can partially substitute for access to top-tier restricted hardware.
15. Which of the following most accurately describes what U.S. semiconductor export controls have demonstrably accomplished — and what they have failed to prevent?
Correct. This balanced assessment reflects the documented evidence: controls have imposed real costs and slowed some capabilities, but have not stopped the most significant Chinese AI and chip advances.
Incorrect. The evidence shows a mixed picture: controls have imposed real costs (talent drain, compute access limits), but have not prevented SMIC's 7nm chip, DeepSeek's efficient LLMs, or re-export through third-country intermediaries.