Technology Export Controls: US-China Chip War and Allies Compliance Pressure

Executive Summary

US export controls have materially slowed but not stopped China's AI compute accumulation, and a new legislative pressure front, centered on the MATCH Act, is now forcing allies to choose between alignment with Washington and protection of their own commercial interests. The core dynamic has shifted since June 2026: the White House and Congress are pulling in opposite directions on export policy, creating a policy environment that Chatham House analysts describe as inconsistent and mercurial, which allies find impossible to follow and Chinese developers have learned to exploit. For allied governments, the pressure is no longer theoretical; the MATCH Act, cleared by the House Foreign Affairs Committee in April 2026, would mandate that the Netherlands, Japan, and South Korea align their controls with US standards within 150 days or face secondary consequences.

• Technology and supply-chain executives: Treat the MATCH Act's 150-day alignment window as a hard planning horizon. If you source semiconductor manufacturing equipment from ASML, Tokyo Electron, or Canon, map your procurement exposure now; additional restrictions on DUV servicing and subcomponent exports are the most moderate-to-high confidence near-term compliance shock.
• Risk officers and investors: The White House-Congress standoff on export policy is not a temporary negotiating posture. Nvidia's inability to recover its Chinese customer base after repeated policy reversals illustrates the permanent risk premium now embedded in China-exposed semiconductor revenue.
• Policy and government affairs professionals: Allied governments are resisting US pressure not on principle but on commercial grounds; the Netherlands lobbied directly against the MATCH Act in late June 2026, signaling that extraction of allied alignment will require either negotiated carve-outs or coercive secondary measures.

Export controls have imposed a real capability ceiling on Chinese frontier AI at the hardware layer, but the ceiling is rising as Huawei's Ascend roadmap and SMIC's DUV workarounds close the gap from below.

What Changed

Since our June 27, 2026 analysis, the MATCH Act was cleared by the House Foreign Affairs Committee on April 22, 2026, and the Netherlands opened a direct lobbying effort against it, with Dutch Trade Minister Sjoerd Sjoerdsma meeting Commerce Secretary Howard Lutnick in Washington in late June 2026. Separately, Bloomberg reported that Huawei plans to manufacture roughly 600,000 Ascend 910C chips in 2026, approximately double 2025 output, and unveiled an Ascend 950 series set to enter production this year. These developments advance two of our prior findings: the allied coordination gap identified in June is now an active geopolitical flashpoint, and China's domestic chip ramp confirms that export controls are delaying but not preventing indigenous capability accumulation.

Our June assessment placed Scenario A (incremental tightening, persistent evasion) at approximately 55%. The MATCH Act's legislative progress and the Netherlands' public resistance raise the probability that allied coordination becomes a sustained contested issue rather than a quietly managed one, which pushes the Scenario B (legislative escalation) pathway from ~30% toward 35-40%, while Scenario C (trade-driven decompression) has narrowed as congressional pressure has intensified.

The Duv Loophole And Why It Is Now The Central Front

The EUV restriction, negotiated by the first Trump administration with the Netherlands in 2019, has held. CSIS and CNAS analysts both conclude that China cannot replicate ASML's EUV technology in the foreseeable future, and CSIS describes the EUV denial as having had "a significant impact on China's semiconductor ecosystem." The problem is that the EUV wall was built while the DUV door remained open.

CNAS reported in May 2026 that Chinese chipmakers have quietly scaled up DUV multipatterning: applying multiple exposure passes to older DUV immersion machines to produce chips approaching advanced node performance at dramatically lower yields and higher cost. SMIC's progress toward 7nm-class and reported attempts at 5nm-class production via this technique illustrates the practical ceiling the regime has imposed, but also its porosity. War on the Rocks analysis from January 2026 found that US and allied firms control approximately 90% of global semiconductor manufacturing equipment and about 92% of overall supply chain value, making Chinese dependency on this tooling deep but not permanent. The question is how long DUV multipatterning can substitute before China's planned indigenous EUV development, however long-dated, changes the calculus.

The MATCH Act's central mechanism is to close this DUV gap by requiring allied governments, primarily the Netherlands, to extend their controls to DUV equipment that China currently accesses freely. The practical implication, as CSIS's transatlantic export control analysis from April 2026 notes, is that CXMT and YMTC, China's two dominant memory chip producers, rely heavily on ASML DUV equipment, and an abrupt cutoff would "amplify price pressure across devices and enterprise servers" at precisely the moment Europe is expanding its own AI infrastructure. The broader economic implications are mutually reinforcing: restricting China's chip capacity tightens global memory supply, which raises input costs for the very allied data center investments Washington wants to incentivize. Short-term gain, long-term cost: the MATCH Act tightens the export control architecture but simultaneously squeezes the allied technology supply chain it is meant to protect.

The White House-Congress Split And What It Means For Beijing's Procurement Strategy

The most consequential development since the June 2022 export control regime began is not any single restriction, but the fracturing of US policy coherence. East Asia Forum reported in March 2026 that the Trump administration is "downplaying the issue publicly while approving the export of higher-tier chips to China and suspending further export restrictions," a posture designed to preserve space for trade negotiations. Meanwhile, the House Select Committee on China Chair John Moolenaar has written multiple letters to Commerce Secretary Lutnick opposing H200 exports, and the AI OVERWATCH Act, pushed through the House Foreign Affairs Committee by Chair Brian Mast on January 22, 2026, would give Congress veto power over export licenses.

The result, as Oplexa and the Semiconductor Industry Association both document, is that Chinese hyperscalers face a procurement environment where the rule can change every six months. Tech Policy Press noted in April 2026 that Chinese technology companies reportedly ordered more than 2 million H200 chips for 2026, but Nvidia has not recovered its customer base because procurement certainty, not just chip availability, drives enterprise ordering decisions. This uncertainty spills into an accelerating domestic substitution dynamic: Huawei's decision to open-source its Ascend software stack and CANN AI toolkit is a direct response to the ecosystem disadvantage that has historically constrained Ascend adoption. If the software gap closes, the hardware gap becomes less decisive.

What is not being reported: public discourse focuses on whether China can match Nvidia chip-for-chip. The more strategically significant question is whether China can build sufficient inference capacity at acceptable cost for the military, surveillance, and dual-use applications that matter most to the PLA. At 7nm-class process and with SuperPoD cluster architectures, China's military AI applications, which Mayer Brown's January 2026 congressional testimony summary noted include nuclear weapons design, cyber warfare targeting, and autonomous drone control, do not require frontier-node chips. They require sufficient compute at sufficient scale, and that threshold may already be achievable with existing Huawei Ascend production levels.

Allied Nation Pressure: Who Faces What And When

The table below maps the current position of key allied nations on semiconductor export control alignment, drawing on CSIS, Chatham House, and government trade press reporting.

The CSIS analysis published in May 2026 emphasizes that no allied country has put end-user controls on Chinese AI and semiconductor firms "like the United States has, at least not publicly," which means that Chinese procurement routed through allied-country intermediaries or cloud nodes faces materially weaker scrutiny than US-origin transactions. This aligns with the June 2026 Mapshock finding that international governance is fragmenting rather than converging: the MATCH Act is an attempt to reverse that fragmentation by imposing US standards extraterritorially, but it is provoking the allied resistance that is now visible in Dutch lobbying.

The interplay between US legislative pressure and allied commercial interests creates a political dynamic that CSIS's Emily Benson noted in April 2026 risks "sharply reducing the market share of an allied champion firm, while bolstering Chinese indigenization efforts." Every blunt instrument applied to ASML accelerates the incentive for China to succeed in its own lithography development. The broader geopolitical implications include a possible scenario where overly coercive US pressure fractures the coalition, leaves ASML selling to neither China nor maintaining US alignment, and creates a vacuum Chinese equipment makers attempt to fill. That scenario remains low confidence on current evidence, but the Dutch lobbying campaign in June 2026 suggests the fracture risk is not theoretical.

Counterarguments

1. The "delay not denial" framing may underestimate how much time matters. The analytical position, supported by Chatham House, is that export controls impose delay rather than permanent denial of Chinese AI capability. But the counterargument is that in the AI race, delay is strategically decisive. If the US maintains a 21-to-49-times compute advantage in 2026 (per Semiconductors Insight estimates), and that advantage persists through the 2027-2030 window when AI-enabled military applications mature, the delay may be sufficient to lock in structural US military AI superiority regardless of China's eventual convergence. The evidence cannot currently discriminate between these two views because the application-level military AI capabilities of both sides remain classified.

2. Huawei's SuperPoD cluster architecture may be more capable than per-chip comparisons suggest. The framework for assessing the capability gap compares individual chips, where Ascend 910C's 7nm-class SMIC process puts it roughly a generation behind Nvidia's Blackwell. But Huawei chairman Eric Xu explicitly dismissed this comparison frame, and Tom's Hardware's analysis confirms that Huawei's scaling-first architecture, using proprietary interconnects to bind thousands of Ascend chips, is designed to achieve system-level throughput that bypasses per-chip efficiency disadvantages. If inference workloads, which dominate deployed military AI, scale well on cluster architectures rather than requiring frontier-node chips, the capability gap narrative rests on a flawed unit of analysis.

3. Allied export control authority gaps are more remediable than CSIS's analysis implies. CSIS's May 2026 analysis emphasizes that allied countries lack the legal authority for US-style Entity List controls and end-use restrictions. But the Netherlands, Japan, and South Korea have all implemented controls in the past three years using existing authority frameworks, demonstrating that the legal authority argument often understates political will as the binding constraint. Japan's March 2023 action on 23 SME categories used existing Foreign Exchange and Foreign Trade Act authority without new legislation. The more honest framing is that allied governments have more legal flexibility than they publicly acknowledge, and the gap between US controls and allied controls reflects commercial and diplomatic choices, not purely legal constraints.

Indicators To Watch

Near-term watch list: (1) US House floor vote on MATCH Act (Q3 2026): passage without Dutch carve-out language is the clearest escalation signal for ASML's operating environment and allied diplomatic posture; (2) Huawei Ascend 950DT launch (Q4 2026): production volumes and PLA procurement signals at launch will indicate whether the Ascend roadmap is tracking to plan or slipping; (3) SMIC Q3 2026 earnings (October 2026): yield and revenue data will provide the most direct observable evidence of whether DUV multipatterning is scaling as Beijing hopes.

Decision Relevance

Scenario A (~50%): Legislative progress stalls, White House holds on H200 framework, allied coordination advances incrementally without MATCH Act. The MATCH Act does not reach a full House floor vote in 2026, the Netherlands reaches a negotiated accommodation with BIS on DUV servicing notification requirements, and Japan expands its SME controls modestly. If you are an AI hardware developer or cloud infrastructure provider with international deployment plans, this scenario extends the current policy ambiguity: plan for case-by-case BIS review of any frontier model deployment and document end-user nationality rigorously. If you lack direct China exposure, monitor ASML quarterly China revenue as the most reliable leading indicator of whether DUV controls are tightening in practice.

Scenario B (~35%): MATCH Act passes with 150-day allied alignment mandate, Netherlands faces formal compliance deadline, ASML DUV servicing restrictions tighten significantly. This scenario materially reduces China's ability to expand SMIC-produced chip capacity and spikes global memory prices as CXMT and YMTC face equipment servicing disruptions. If you have semiconductor manufacturing or memory procurement exposure, begin identifying non-Chinese supply alternatives for DRAM and NAND now; the price spike CSIS and CNAS both flag would arrive with limited warning. If you advise on technology policy, this scenario accelerates Chinese domestic EUV development funding by removing any remaining incentive for Beijing to preserve existing equipment partnerships.

Scenario C (~15%): Trade deal with China includes explicit semiconductor provisions, White House grants H200 and potentially Blackwell access in exchange for rare earth and other concessions, MATCH Act shelved. This remains the lowest-probability scenario given congressional pressure documented by East Asia Forum, the Mayer Brown analysis of legislative momentum, and the structural bipartisan consensus around restricting Chinese military AI development. If you are evaluating China market re-entry for AI products, do not plan on this scenario as a base case; the legislative countermeasures would moderate-to-high confidence arrive faster than commercial relationships could be rebuilt.

Why The Euv Wall Held But The Duv Door Did Not Close

The causal mechanism linking the EUV restriction to China's sustained capability gap is well-established: the first Trump administration negotiated with the Netherlands in 2019 to deny ASML EUV exports to China, CSIS confirms this has foreclosed China's path to sub-7nm production using conventional lithography approaches, and CSIS notes that "producing this kind of complex machinery entirely within China is not moderate-to-high confidence to happen in the foreseeable future." The mechanism is strong for EUV and moderately strong for chipmaking equipment broadly.

The DUV gap is different. CNAS's May 2026 analysis documents the mechanism by which China exploits unconstrained DUV access: multipatterning on DUV immersion systems produces 7nm-equivalent chips at lower yield and higher cost. The causal link from continued DUV access to Chinese chip production capacity is a hoop test: it must be present for China's near-term production ramp to hold, and the evidence that it is present is the SMIC production data and Huawei's output targets. What would falsify the mechanism is evidence that SMIC's multipatterning yields are too low to support Huawei's 600,000-unit target, which SMIC Q3 earnings will begin to clarify.

(for DUV-based production path; STRONG for EUV denial).

• The most material unknown is actual PLA AI compute deployment and the specific performance thresholds required for priority military applications. All civilian-facing analysis of Huawei Ascend capability is based on commercial product specifications; classified PLA system requirements and actual cluster deployments are not observable in open sources.
• Huawei's production targets are self-reported through Bloomberg sources described as anonymous people familiar with the plans; yield rates at SMIC's N+2 process node are not independently verified. If actual yields are significantly below plan, the capability gap assessment would be more favorable to the US position than this analysis reflects.
• This assessment does not cover Chinese AI model development independent of compute constraints. DeepSeek's efficiency gains documented in early 2026 demonstrate that algorithmic improvements can substitute for raw compute in some domains, which Chatham House identifies as a fundamental limitation of the hardware-centric export control approach.
• Allied government positions are assessed on the basis of public statements and lobbying actions; internal policy deliberations in the Netherlands, Japan, and South Korea are not observable, and the gap between stated positions and eventual policy decisions may be significant.
• The White House-Congress divergence on export policy introduces a high degree of aleatory uncertainty into the 6-12 month policy trajectory. The legislative calendar, trade negotiation timelines, and domestic political dynamics in an election-adjacent year in multiple allied nations create genuine unpredictability that no analytic framework can resolve.