The US-China semiconductor conflict has moved past chips. It is now a war over the machines that make chips, the materials those machines depend on, and the question of whether two incompatible technology ecosystems can coexist in a world built on shared supply chains.
By Santhosh Vishwamitra
The Situation in Five Points
- According to industry analysis, China's domestic chip self-sufficiency has risen from around 16 percent to 28 percent in a single year, while Nvidia's China AI-chip revenue is reported to be falling sharply under a 25 percent tariff.
- In March 2025, the Trump administration imposed additional restrictions, blacklisting dozens of Chinese entities. Separately, researchers at Peking University published a claim, as yet independently unverified, of a 2D transistor that may operate significantly faster than TSMC's 3-nanometre devices while consuming less energy.
- Starting January 1, 2026, TSMC, Samsung, and SK Hynix lost their automatic export exemptions for their Chinese fabs and must now apply for annual licences from the US Department of Commerce, giving Washington a renewal lever it can adjust each year.
- The MATCH Act, introduced in the US House and Senate in April 2026 and still under congressional consideration, proposes to ban not only the sale of DUV lithography machines to China but also the maintenance of equipment already installed there. It is proposed legislation, not yet law.
- China's most consequential countermeasure has increasingly moved beyond semiconductors into rare earths and critical materials. China dominates global rare earth processing and magnet manufacturing, with widely cited estimates placing its share of processing at around 90 percent, and its new export licensing regime threatens supply chains for Western defence systems and electric vehicles.
Key Terms
| Term | What It Means |
|---|
| EUV | Extreme Ultraviolet lithography: the most advanced chip-printing technology, made only by ASML in the Netherlands. China has been blocked from buying EUV machines since 2019. Required for chips below 7nm using conventional methods. |
| DUV | Deep Ultraviolet lithography: older and less precise than EUV, but still capable of producing 7nm-class chips using multi-patterning workarounds. Currently the technology SMIC depends on. The MATCH Act, if passed, would restrict DUV exports to China. |
| SMIC | Semiconductor Manufacturing International Corporation: China's largest chip foundry and the state's designated national champion for advanced domestic chip production. Blacklisted by the US in 2020. |
| MATCH Act | Multilateral Alignment of Technology Controls on Hardware Act: proposed bipartisan US legislation introduced April 2, 2026, that would ban the sale and servicing of DUV lithography tools to China and push allied nations to align their export controls with US standards. Not yet law as of mid-2026. |
| Entity List | A US Department of Commerce list of foreign companies restricted from receiving US technology exports without a licence. Huawei, SMIC, and dozens of Chinese semiconductor firms are on it. |
| Rare Earths | A group of 17 elements critical to defence electronics, electric motors, and precision manufacturing. China dominates their processing and is using export licensing as a geopolitical pressure tool. |
| HBM | High Bandwidth Memory: specialised memory chips that sit alongside AI processors. A current bottleneck for China's AI hardware ambitions. Dominated by South Korea's SK Hynix and Samsung. |
| Multi-patterning | A manufacturing technique that achieves fine circuit patterns by running wafers through lithography multiple times. Allows DUV machines to produce chips that would otherwise require EUV, but at higher cost and lower yield. |
Where the War Stands in Mid-2026
The semiconductor conflict between the US and China did not begin in 2022, but that is when it became a formal policy instrument. The Biden administration's October 2022 Bureau of Industry and Security rules blocked the export of advanced logic chips, AI training hardware, and chip manufacturing equipment to China. The restrictions were tightened in October 2023, expanded in December 2024, and supplemented in January 2026 with a 25 percent Section 232 tariff on advanced semiconductor imports.
Each round of controls was met with Chinese adaptation, and each round of adaptation triggered another escalation. The pattern is now structurally embedded in both governments' planning. What makes 2026 different from earlier phases is that the battlefield has widened in both directions simultaneously: the US is moving against the machines that make chips rather than only the chips themselves, while China is moving against the raw materials that Western defence and electronics industries cannot currently source elsewhere.
The diplomatic layer is also shifting. According to reporting from War on the Rocks, in late 2025 Washington approved restricted H200 exports to selected Chinese customers with conditions attached, and the two sides agreed to a tentative one-year suspension of some measures after emergency consultations in Busan, South Korea. That truce does not reflect de-escalation. It reflects the recognition by both sides that unconstrained escalation carries costs neither has fully modelled. The underlying competition continues regardless of the diplomatic temperature.
Timeline: From Export Controls to Equipment War
Note: Dates up to mid-2026 reflect official policy actions and publicly reported milestones. Future dates from 2027 onward are indicative projections based on current trajectories, not confirmed plans. Verify all milestones against official sources before citing.
| Date | Development |
|---|
| Oct 2022 | Biden administration BIS imposes first major chip export controls targeting advanced AI processors and chip manufacturing equipment bound for China. |
| Oct 2023 | Controls tightened. New performance thresholds added; loopholes closed. |
| Dec 2024 | Controls expanded to cover high-bandwidth memory (HBM) and additional equipment categories. |
| Jan 2025 | BIS issues global AI Diffusion Rule creating a three-tier country licensing framework for AI chip exports. Treasury outbound investment screening rule takes effect. |
| Mar 2025 | Trump administration imposes additional restrictions, blacklisting dozens of Chinese entities. Peking University researchers publish 2D transistor claim (unverified independently). |
| Apr 2025 | Trump administration bans H20-class chips, closing Nvidia's engineered-for-China workaround. |
| Sep–Oct 2025 | Trump extends controls to foreign affiliates. China retaliates with rare earth export licensing requirements. |
| Dec 2025 | Trump announces H200 sales to China will be permitted. China implements its strictest-ever rare earth export regime, effective December 1. |
| ► Jan 2026 | EXISTING POLICY: TSMC, Samsung, and SK Hynix lose automatic exemptions; must apply annually for US licences to operate China fabs. 25% Section 232 tariff on advanced semiconductors takes effect. |
| ► Apr 2026 | PROPOSED: MATCH Act introduced bipartisan in House and Senate. Targets DUV machines and servicing. House Foreign Affairs Committee marks up 20+ export control bills. |
| ► NOW (Jul 2026) | MATCH Act under congressional consideration. China's rare earth licensing approval rate for European firms reported below 25%. SMIC 7nm yields reported between 20% and 40%. Hua Hong readying second Chinese 7nm capacity source. |
| 2027–2028 (projected) | If MATCH Act passes: DUV servicing restrictions could gradually degrade installed SMIC capacity. If it fails: China continues building on its existing DUV base. |
| 2030+ (projected) | China targets photoresist localisation at 50%. CXMT targeting HBM competitive parity. Whether the capability gap widens or narrows depends heavily on policy and yield progress. |
What China Has Built Under Pressure
The most important story in the semiconductor conflict is not the US export control architecture. It is what China has achieved under the constraints of that architecture, and what it has not yet achieved.
SMIC's N+2 and N+3 nodes represent 7nm-class and 5nm-class capabilities reached without access to ASML's EUV lithography, according to industry analysis from Oplexa and CSIS. The tradeoffs are real: DUV-based manufacturing carries higher costs, lower yields, and slower iteration cycles than EUV-based production at TSMC or Samsung. SMIC's advanced node capacity is estimated at about 45,000 wafer starts per month in 2025, expanding toward 60,000 through 2026, according to the same sources.
Yield is the number that matters most and is hardest to verify from outside. SMIC's 7nm process yield rates are reported to be between 20 and 40 percent, according to industry analysts cited in multiple 2026 assessments. The chips function, but reported yields and production costs make them difficult to compete with leading-edge products produced at TSMC or Samsung without substantial domestic support. Outside a state-subsidised market, these chips are not yet commercially viable at global prices.
Process node alone does not determine AI capability. Advanced AI systems also depend on chip architecture, packaging, high-bandwidth memory, interconnects, manufacturing yield, software ecosystems, and the ability to produce large quantities economically. China's progress in individual components of this stack does not automatically mean parity with the leading US-aligned ecosystem. That distinction matters and tends to get lost in the headline framing of "China makes 7nm chips."
Huawei's Ascend AI accelerator series represents a more credible near-term challenge. The Ascend 910C is reported to compete with mid-tier Nvidia GPUs for enterprise AI workloads, according to CSIS and industry sources. Huawei planned to double Ascend production to around 600,000 units in 2025 and aimed for 1.6 million dies across its Ascend line by 2026, according to market intelligence reports. In the Chinese domestic market, where Nvidia hardware is now subject to a 25 percent tariff and annual licence uncertainty, Huawei Ascend is not merely a strategic option. It is increasingly the default for large-scale AI infrastructure purchasing.
In March 2026, Reuters reported that Hua Hong Group, China's second-largest chipmaker, is readying 7nm production capacity at its Huali Microelectronics facility in Shanghai, using DUV equipment and development supported by Huawei-backed domestic equipment supplier SiCarrier. This would break SMIC's monopoly as China's sole advanced node producer, creating redundancy in the domestic advanced chip supply chain. The one area where China has not yet closed the gap is High Bandwidth Memory. China's CXMT is still reported to be below competitive parity with SK Hynix and Samsung on HBM, according to Oplexa's 2026 analysis. AI accelerators require HBM to function at full capability, and no amount of advanced logic chip progress fully resolves that constraint.
The MATCH Act: Proposed Escalation
On April 2, 2026, a bipartisan group of US lawmakers introduced the MATCH Act, proposed legislation that would significantly tighten restrictions on the sale and servicing of semiconductor manufacturing equipment to China. The bill targets ASML's last remaining open door to China: the older DUV machines and the servicing of tools already installed there. It is not yet law. Its passage, scope, and enforcement mechanisms remain subject to congressional negotiation and allied diplomacy.
The servicing provision is the part most commentary has not focused on. Manufacturing tools are not appliances that run unattended. Without support, calibration, parts supply, and technical assistance, equipment availability declines and process stability deteriorates over time. If the MATCH Act passes with its servicing clause intact, the pressure on SMIC and Hua Hong would not arrive as a sudden cutoff. It would be a slow degradation of the installed base that China has spent years accumulating. That is a qualitatively different instrument from banning new equipment sales.
According to ASML's own disclosures and Reuters reporting, China was ASML's largest market in 2025, accounting for around 33 percent of sales. That share was already expected to fall to around 20 percent in 2026. In the first half of 2026, China was ASML's third-largest regional market, behind South Korea and Taiwan. ASML's CEO warned of possible zero growth in 2026 partly due to geopolitical uncertainty. The MATCH Act, if passed in its current form, would cost ASML a substantial further portion of its revenue and create significant political pressure inside the Netherlands and the EU to resist full alignment with US restrictions.
The assumption that export restrictions can permanently freeze a capability gap has repeatedly proved optimistic. According to CSIS and TechPolicy Press analysis, Beijing is funding parallel workarounds across nanoimprint lithography, advanced multi-patterning, and longer-term EUV prototypes. China's stated target is to raise domestic photoresist localisation from 20 percent in 2024 to 50 percent by 2027 to 2030. Semiconductor equipment self-sufficiency stands at about 35 percent. These are not gaps that close overnight, but they are closing.
The Rare Earth Counterattack
China's most consequential countermeasure has increasingly moved beyond semiconductors into rare earths and critical materials. This is not a symmetric response. It is a deliberate exposure of a different vulnerability in the Western industrial base.
China's rare earth export licensing regime, effective December 1, 2025, requires licences for products containing even trace amounts of Chinese-origin rare earth materials or made using Chinese technology, according to reporting from InformedClearly and the European Parliament's 2025 briefing. Firms affiliated with foreign militaries face automatic denial, and military-use requests are rejected outright. Export controls introduced in 2025 and 2026 have triggered reported price spikes of up to sixfold on affected components, with licensing approval rates for European firms reported below 25 percent. These figures come from industry and government sources and may vary by material and application category.
F-35 joint strike fighters require rare earth-derived magnets in their actuators, radar systems, and targeting equipment. Major Western missile programmes depend on rare earth permanent magnets for guidance accuracy. Electric vehicles and wind turbines require them for motors. The European Parliament's 2025 briefing noted that over 80 percent of European companies depend on Chinese supply chains for critical minerals essential to defence and electric vehicles, though coverage and methodology of that figure should be verified against the primary source before use.
The strategic logic: Washington is attacking China's ability to manufacture at the frontier. Beijing is attacking Washington's ability to build anything that requires rare earth-derived components. Neither side can yet fully absorb the other's attack. That mutual vulnerability is what is preventing full escalation while making the conflict drag on without resolution.
Who Gets Caught in the Middle
| Actor |
Position |
US Pressure |
China Pressure |
| ASML (Netherlands) |
Only maker of EUV; major DUV supplier to China |
MATCH Act (proposed) threatens China revenue; pressure to align with US controls |
Loss of largest 2025 market; retaliation risk on other trade |
| TSMC (Taiwan) |
Manufactures most of the world's advanced chips |
Annual US licence renewals for China fabs; pressure for Arizona expansion |
China could threaten Taiwan itself; mainland Chinese market access at risk |
| South Korea (Samsung / SK Hynix) |
Dominates DRAM and NAND; HBM leader |
Annual licence requirements; pressure to reduce China manufacturing exposure |
Significant manufacturing presence in China; rare earth dependency |
| Europe (ASML / Infineon / others) |
Critical equipment and design tools |
Pressure to join US control architecture; MATCH Act multilateral alignment clause |
Rare earth approvals reported below 25% for European firms; supply disruption |
| India |
Emerging fab destination; growing chip design sector |
Courted as alternative supply chain partner; Tier I AI Diffusion Rule status |
Rare earth dependency; manufacturing capability still early-stage |
| Vietnam / Malaysia / Thailand |
Packaging and assembly hubs; alleged chip diversion transit points |
End-use verification pressure; risk of losing export privileges |
Trade relationship dependency on China |
The enforcement gap is structural and documented. Super Micro Computer's co-founder is alleged to have diverted $2.5 billion of Nvidia servers to China through a scheme routed via Southeast Asia, according to reporting from The Next Web and multiple industry sources. The allegation demonstrates both the scale of demand for restricted chips and the limits of an enforcement regime that relies on declared end-use and corporate compliance teams. Third-country routing will persist as long as demand for restricted chips exceeds legal supply, and the pressure it places on smaller nations hosting those routes is a recurring feature of the control architecture's failure points.
Three Scenarios for the Next Five Years
How to read these scenarios: These are editorial projections based on current policy trajectories and publicly available technical data as of mid-2026, not forecasts. "Most Likely" reflects the author's assessment of the most plausible path given current signals, not a prediction that it will occur. Semiconductor policy has moved faster than most analysts predicted; treat all projections as directional rather than fixed.
Best Case: Managed Bifurcation with Limited Escalation
The MATCH Act passes in a modified form that restricts new DUV sales but preserves a narrow servicing window for existing equipment, giving ASML and the Netherlands enough political cover to comply without a full commercial rupture. China's rare earth licensing regime remains in place but is applied selectively, allowing most Western defence supply chains to secure licences at elevated cost. Both sides tacitly acknowledge that full decoupling would be mutually destructive and manage a parallel-ecosystem arrangement: two separate semiconductor supply chains, one serving the US-allied world and one serving China and its partners, that share mature-node technology but diverge at the frontier. Trade in non-strategic semiconductors continues. The outcome is stable but costly for every actor caught between the two systems.
Most Likely: Escalating Controls with Partial Enforcement
The MATCH Act passes or is implemented through executive action in a form that restricts new DUV exports but struggles to enforce the servicing ban due to allied resistance, particularly from the Netherlands. China continues building on its existing DUV base. SMIC yields improve gradually over three to five years. Hua Hong becomes a second advanced node producer. The HBM gap narrows but does not close by 2030. Rare earth export restrictions tighten periodically, producing supply shocks in specific components but not a full cutoff. The US export control architecture keeps the capability frontier moving faster in the West than China can independently replicate, but the gap is measured in years, not decades. Third-country enforcement remains a structural leak. The conflict settles into a sustained slow-motion technology decoupling that costs everyone something and resolves nothing.
Worst Case: Full Equipment Cutoff and Rare Earth Crisis Collide
The MATCH Act passes in its most aggressive form, including the servicing ban, within twelve months. China's rare earth licensing denial rate for Western defence contractors rises above 50 percent. A military incident in the Taiwan Strait, not necessarily a full invasion, triggers emergency export controls against any entity doing business with China in semiconductors. The global chip supply chain fragments in months rather than years. US allies and non-aligned nations are forced to choose which supply chain to join. The effect on global AI development and economic growth is severe, because the AI infrastructure buildout depends on chip supply that neither fragmented ecosystem can independently sustain at the required scale.
What Most People Miss About This Conflict
The framing most people use for the chip war is a race: who gets to the frontier first, who controls the most advanced process node, who wins on the most prestigious technical metric. That framing misses the more consequential dynamics operating underneath it.
The first is the policy incoherence problem on the US side. According to TechPolicy Press analysis from April 2026, the administration was loosening controls on finished chips while Congress attempted to tighten them, and the executive branch withdrew a draft rule in February 2026 that would have further restricted global AI chip sales. Nvidia's H200 was banned, then unbanned, then subjected to new conditions that independent experts described as contradictory. The shift from automatic exemptions to annual licence renewals for TSMC, Samsung, and SK Hynix is structurally significant, but the terms can change with every renewal cycle. This inconsistency is not a bug. It reflects a genuine tension between the national security goal of limiting Chinese AI capability and the commercial interest of US chip companies in maintaining revenue from the world's largest chip-consuming market. That tension has not been resolved; it has been managed through ambiguity.
The second overlooked factor is what happens to the non-aligned world. India, Vietnam, Malaysia, the UAE, and dozens of other nations are not choosing between the US and China. They are trying to benefit from both simultaneously. The AI Diffusion Rule's three-tier country classification grouped nations by their alignment with US controls, but most of the world falls into a middle tier that is neither fully trusted nor excluded. Managing those middle-tier relationships, while preventing third-country diversion, is probably the most practically difficult enforcement problem the US faces.
Third: the Peking University 2D transistor research deserves attention, but with care. Researchers at Peking University reported in March 2025 that they had developed a 2D transistor using new materials that they claimed could operate 40 percent faster than TSMC's 3nm devices while consuming 10 percent less energy. The claim remains independently unverified. The research has not been replicated by external institutions as of mid-2026. It should be treated as a scientific claim under review, not a confirmed breakthrough. But the strategic ambition it represents is real: rather than catching up to TSMC on silicon-based EUV lithography, pivot to a different material substrate where no established player has a head start. If that bet eventually validates, export controls on conventional chip manufacturing equipment become partially irrelevant. If it does not, China remains dependent on DUV workarounds for the foreseeable future.
Editorial Analysis: What Is Really Happening
Editorial Analysis · Santhosh Vishwamitra
The semiconductor conflict is often described as a technology competition. I think that framing underestimates it. What is actually happening is a fight over who gets to define the infrastructure of the next fifty years of economic activity. AI depends on chips. Advanced manufacturing depends on chips. Defence systems depend on chips. Whoever controls the production of frontier semiconductors controls a chokepoint that runs through virtually every other strategic industry.
The MATCH Act's servicing clause is the detail that reveals the most about US intent. Banning new equipment sales is aggressive but manageable over time. Banning the maintenance of installed equipment is a different category of action entirely. It is designed to make the equipment China already has degrade over time, not just prevent future acquisition. That is not a trade restriction. It is closer to a siege tactic applied to industrial capacity. If it passes in its full form, it signals that the US is prepared to accept significant allied commercial damage, ASML's China revenue is not trivial, in exchange for long-term capability degradation of Chinese semiconductor manufacturing.
China's rare earth licensing regime is the mirror image. It is not designed to stop Western defence production tomorrow. It is designed to make the cost visible and unpredictable. A reported sixfold price spike on rare earth components does not ground the F-35 fleet. It creates procurement crises, contract renegotiations, and budget holes across Western defence programmes. China is making the point that mutual dependency is real, not theoretical. It is not a chip for a chip. It is rare earths. That asymmetry is deliberate.
What the next five years will clarify is whether the US can maintain allied cohesion on export controls as the commercial costs compound. The Netherlands, Japan, and South Korea are formal or informal partners in the US chip control architecture. Each has significant commercial exposure to China. History suggests that sustained allied cohesion on costly economic restrictions is harder to maintain than it appears at the point of signing. The MATCH Act's multilateral alignment clause is an attempt to solve that problem through legislation. Whether it succeeds depends on how much allied governments decide they have to lose from non-compliance versus from compliance.
Three things I am watching closely. First, whether the MATCH Act passes with the servicing clause intact, or whether allied lobbying strips it before final passage. Second, whether China's CXMT closes the HBM gap by 2028, because that is the single technical development that would most change China's AI hardware independence picture. Third, whether the Peking University 2D transistor research produces independently verified results within two to three years. A genuine material-science route around silicon EUV constraints would not end the conflict, but it would change what the conflict is fundamentally about.
The semiconductor war is not close to resolution. The fight over equipment, materials, and allied alignment is harder to manage and harder to reverse than the fight over finished chips. The supply chains that took thirty years to build are being deliberately fractured. Whatever the new configuration looks like, it will not resemble the integrated global semiconductor supply chain that enabled the early 2020s AI buildout. Both sides are building something new. Neither has fully modelled what it will cost.
Key Takeaways
- The MATCH Act is proposed legislation, not yet law. If passed, it would ban both the sale and maintenance of DUV machines in China, a qualitative escalation beyond all previous controls. Its passage and final scope remain uncertain.
- China's domestic chip self-sufficiency is reported to have risen from around 16 to 28 percent in one year. SMIC is producing 7nm-class chips using DUV multi-patterning at yields reported between 20 and 40 percent. Commercially uncompetitive globally, but functional within a state-subsidised domestic market.
- Process node alone does not determine AI capability. Progress in 7nm logic chips does not mean China has closed the overall gap with US-aligned semiconductor ecosystems, which include advanced packaging, HBM, software, and yield advantages that are harder to quantify but equally important.
- China's rare earth export licensing regime, effective December 2025, affects supply chains for defence systems and electric vehicles. Approval rates for European firms are reported below 25 percent. These figures come from industry and government sources; coverage varies by material category.
- TSMC, Samsung, and SK Hynix now operate their China fabs under annual US licence renewals. This is existing policy. Washington holds a renewal lever it can adjust each year.
- The enforcement gap is structural. Third-country diversion through Southeast Asia is documented and persistent. No policy relying on declared end-use has closed it.
- The Peking University 2D transistor claim has not been independently verified as of mid-2026. Monitor it, but do not treat it as a confirmed capability.
- Allied cohesion, from the Netherlands, Japan, and South Korea, is the most important variable in whether US export controls remain effective over the medium term. Commercial pressure on all three is mounting.
Read More from Vishwamitra Chronicles:
Science vs Myth: The Truth About Dark Matter
Fast Read: 11 August
Big Numbers: $ 2 Trillion Space Economy Explained
Field Guide: How Vaccine travels
Student Explainer: NATO and History of NATO
Comments
Post a Comment