AI · US-China Chip War · May 24, 2026 · 6:00 PM ET

Share

Huawei Just Claimed a Path to
1.4nm Chips by 2031. The Real Story Is Whether the U.S. Sanctions Architecture Is Holding.

On Monday, May 25, 2026, in a keynote address at the IEEE International Symposium on Circuits and Systems in Shanghai, He Tingbo— president of Huawei’s semiconductor business and director of its Scientist Committee — unveiled what the company is calling the Tau (τ) Scaling Law and a companion chip-architecture concept, LogicFolding. Bloomberg, the South China Morning Post, and Reuters all carried the announcement. Huawei’s pitch: by reframing semiconductor progress around signal-propagation delay rather than transistor dimensions, the company can deliver 1.4-nanometre-equivalent transistor density by 2031 without access to the ASML extreme-ultraviolet lithography machines that Western export controls have walled off since 2019.

That date matters because it is three years after TSMC plans to mass-produce true 1.4nm chips in 2028, and roughly five years behind where the actual global frontier will be by then. Huawei’s own framing is a tacit admission of the gap: even on its best 2031 case, the company is targeting where TSMC will be in 2028. The cycle has treated this as a breakthrough headline. It is more accurately a corporate-keynote target date attached to a marketing concept, from a company that has spent the last six years saying it can compensate for U.S. sanctions through software, architecture, and the math of signal propagation.

The accountability question this page surfaces is not whether Huawei’s engineers are good — they are. It is whether the U.S. sanctions architecture that was supposed to keep advanced semiconductor manufacturing capability inside democratic-allied jurisdictions is still being enforced with the rigor the December 2024 Biden-era Entity List sweep promised — or whether the May 2026 Trump-Xi Beijing summit, at which USTR Jamieson Greer (R) told Bloomberg TV that chip controls were “not discussed”, has quietly shifted the posture in Beijing’s favor.

• 5yearscurrent process-node gap between TSMC’s leading edge and Huawei + SMIC’s best verified production node · CSIS, TechInsights, SemiAnalysis converge on this figure
• 1.4 nm2031 targetHuawei’s claimed transistor-density target via the Tau Scaling Law and LogicFolding · density-equivalent, not a confirmed 1.4nm critical dimension
• 2028TSMC ships true 1.4nmTSMC’s A14 angstrom-class process is on the published corporate roadmap for mass production in 2028 — three years before Huawei’s 2031 target
• 381chips claimedHuawei’s own count of chips it says were designed and mass-produced using Tau-Scaling-Law principles over the last six years — unaudited
• 200,000chips / yearCommerce Sec Howard Lutnick’s (R) publicly cited estimate of China’s total annual advanced-chip output — the figure he uses to defend the Entity List as effective
• 600,000Ascend 910C / yrSemiAnalysis 2026 production projection for Huawei Ascend 910C units — sits in tension with Lutnick’s 200,000 cap on China advanced output
• 2.9MTSMC dies smuggledWafers Huawei is alleged to have obtained from TSMC via the Sophgo cutout before BIS shut the diversion down — the empirical case that the sanctions have leaks
• ~140Entity List addsEntities added in the December 2024 Biden-administration Huawei/SMIC-supply-chain sweep — the sanctions baseline the Trump administration inherited in January 2025
• $81,620,000,000Nvidia FQ1 2027 revNvidia’s May 21, 2026 quarterly print — the financial context for Jensen Huang’s “largely conceded China” remark · 85% YoY growth

The Shanghai ISCAS keynote — titled “New Semiconductor Path in Practice”— was Huawei’s attempt to reframe what counts as a generational leap in chip design. Western roadmaps (TSMC, Intel, Samsung) track progress through ever-smaller critical dimensions, expressed in nanometres: 7nm → 5nm → 3nm → 2nm → 1.4nm. He Tingbo’s keynote argued that, with EUV lithography off the table for Chinese fabs under U.S. sanctions, the relevant unit of progress is not transistor size but signal-propagation delay— the Greek letter tau. Shrink tau and you get the same performance gains, the argument goes, even if your transistors are physically larger.

The accompanying architectural concept, LogicFolding, was described as a way to shorten the physical wiring inside a chip by folding logic blocks in three dimensions, reducing the distance signals travel between gates. Huawei said its Kirin chips launching in fall 2026 would be the first commercial products to use a LogicFolding-derived architecture. The company further claimed it had already designed and mass-produced 381 chips over the past six years on Tau-Scaling-Law principles, across smartphones, AI accelerators, and embedded use cases.

“A single chip is still behind by a generation compared with the U.S. industry, but we compensate via mathematics.”

Ren Zhengfei · Huawei founder · paraphrased remarks attributed by People's Daily · the strategic frame Huawei has used since 2023

That “compensate via mathematics” line is the rhetorical core of the entire announcement. It is also the part that no independent teardown shop or competing engineering team has yet been able to validate. ISCAS is a respected IEEE technical conference, but a keynote claim is a claim, not a peer-reviewed measurement. The 381 chips Huawei cites are not enumerated in any public paper trail; the LogicFolding architecture is described in press materials but not specified at the level a fab engineer could replicate. As of publication, the announcement is best read as a Huawei marketing target that the cycle has covered as a scientific breakthrough.

The case for skepticism does not rest on geopolitics. It rests on the existing public record of what Huawei has actually been shipping versus what it has been claiming. In October 2025, Bloomberg reported on a TechInsights teardown of Huawei’s flagship AI accelerator, the Ascend 910C. The teardown found:

• TSMC-fabricated compute dies inside the package, sourced via a chain that ran through Sophgo, a Chinese front-design house that BIS later cut off the export-control map. The Department of Commerce has documented roughly 2.9 million dies reportedly diverted from TSMC’s production to Huawei via this and adjacent cutouts before the enforcement action.
• Non-Chinese high-bandwidth memory— Korean-made HBM from SK Hynix and Samsung — stacked alongside the compute dies. China does not yet have a domestic HBM supplier capable of replacing this content at the bandwidths AI inference requires.
• SMIC N+2 7nm-class logic elsewhere in the supply chain, with poor yieldsper SemiAnalysis — meaning the indigenous Chinese node Huawei does control is producing far fewer working dies per wafer than TSMC’s equivalent-generation processes did six years earlier.

In other words: as recently as late 2025, the most-advanced AI chip Huawei was shipping was substantially notChinese-made in its highest-value components. The company’s May 2026 announcement of a 2031 path to 1.4nm density should be read against this baseline. The story of the last five years is not that Huawei caught up to TSMC; it is that Huawei found ways to keep accessing TSMC-fabricated content despite the Entity List, and that BIS eventually closed the specific cutout it identified.

The five-year gap is consistent across every generation. TSMC shipped 7nm in 2018; Huawei’s Kirin 9000S in the 2023 Mate 60 Pro — the chip that caught the first Trump-administration sanctions architecture off guard — was on SMIC’s N+2 7nm-class process. TSMC shipped 5nm in 2020; TechInsights documented Huawei reaching 5nm-class in 2025. The Kirin 9030, Huawei’s 2026 mobile SoC, claims SMIC N+3 — what some describe as “3nm-class” on density — but Tom’s Hardware’s reporting flags that the actual performance cannot match TSMC’s 5nm-generation products from 2020.

“Huawei is very, very strong. They had a record year, they'll likely, very likely, have an extraordinary year coming up, and their local ecosystem of chip companies are doing quite well, because we've evacuated that market.”

Jensen Huang · Nvidia CEO · CNBC interview on Nvidia Q1 FY2027 earnings call · May 21, 2026

The technological ceiling Huawei has been operating under since 2019 is not a single executive order — it is a layered regulatory architecture administered by the U.S. Commerce Department’s Bureau of Industry and Security (BIS). The three load-bearing instruments are:

• The Entity List. A roster of foreign companies and individuals subject to license requirements for any U.S. export, re-export, or in-country transfer. Huawei was added in May 2019; SMIC was added in December 2020; the December 2024 Biden-era sweep added roughly 140 additional entities across the Huawei/SMIC supply chain.
• The Export Administration Regulations (EAR). The underlying statutory framework that governs dual-use exports — including the controls on advanced semiconductor manufacturing equipment, EDA tools, and design software that the Chip Race depends on.
• The Foreign Direct Product Rule (FDP). The extraterritorial reach mechanism that lets BIS block third-country fabs (including TSMC in Taiwan) from selling to Entity-Listed Chinese customers if any U.S. technology was used in the production process — which, for every advanced-node fab on Earth, it is.

The Trump-administration officials currently administering this architecture:

• Howard Lutnick (R), U.S. Secretary of Commerce — political principal over BIS · publicly defends the Entity List with the “200,000 advanced chips a year” estimate for China’s total output, framed as evidence the controls are working.
• Jeffrey Kessler (R), Under Secretary of Commerce for Industry and Security — runs BIS as a day-to-day matter, owns Entity List adds, FDP enforcement, and license-review policy.
• Marco Rubio (R), U.S. Secretary of State — led the State Department piece of the May 2026 Trump-Xi Beijing summit; became the first U.S. Secretary of State on the Chinese-government sanctions list to visit China while in office.
• Jamieson Greer (R), U.S. Trade Representative — on Bloomberg TV after the Beijing summit, said chip controls were not discussed in the U.S.-China bilateral talks.
• President Donald Trump (R)— set the top-line softer posture toward Beijing during the May 2026 summit, including the public “opening China to American business was my first request to President Xi” framing.

The Trump-Xi Beijing summit in mid-May 2026 was the largest U.S.-China bilateral engagement of the second Trump administration. Rubio, Lutnick, Greer, and the President himself were in the delegation. The pre-summit framing on the U.S. side emphasized trade deals, agricultural purchases, fentanyl cooperation, and a broad reopening narrative. The post-summit framing — per Greer’s on-air Bloomberg TV remarks — was that chip controls were not discussed.

“Chip controls were not discussed in Beijing.”

USTR Jamieson Greer (R) · Bloomberg TV interview · post-summit, May 2026

There are two ways to read “not discussed.” The charitable read: the U.S. delegation made a deliberate choice not to put the most contentious file on the summit table, preserving unilateral flexibility on BIS enforcement for after the diplomatic window closed. The skeptical read — the one the House Select Committee letter five months earlier was already anticipating — is that “not discussed” in a bilateral setting with Xi Jinping is functionally an unspoken concession: Beijing now knows that the senior trade-policy principal of the United States did not raise chip controls as a live priority in the most consequential bilateral of the year. That itself is a signal, regardless of intent.

Nvidia CEO Jensen Huang, who travelled with the Trump delegation and met with Chinese officials in connection with the summit, gave the most consequential post-summit on-the-record assessment when, on the May 21 Nvidia Q1 FY2027 earnings call — the day after Nvidia booked $81,620,000,000 in quarterly revenue, up 85% year-over-year — he conceded that Nvidia had “largely conceded” China’s AI chip market to Huawei. The statement was framed as a business reality, not an indictment of the controls; but its substance is that Huawei is now the dominant AI-accelerator vendor inside the second-largest economy on Earth, and that Nvidia does not expect to recover that market in the near term.

This is the editorial center of the page. Huawei’s ISCAS announcement is a marketing statement attached to a 2031 target date, made by a company that has spent six years working under the most aggressive technology-export regime the United States has ever imposed on a single foreign firm. The right object of accountability is not the foreign engineers doing what their government has asked them to do. The right object of accountability is the U.S. policy architecture that, by design, was supposed to ensure no announcement like this would ever need to be made — because the Chinese semiconductor industry would be five or more generations behind the U.S. frontier with no credible path to close the gap.

The five-year gap remains. The independent record — TechInsights, SemiAnalysis, CSIS — says it remains. But the December 2025 House Select Committee letter says the H200 license posture is now putting that gap at risk. The May 2026 Beijing summit said, with the silence of an unaddressed agenda item, that chip controls are no longer the trade-policy priority they were in the Biden administration’s last December. The May 21 Nvidia earnings call said the China market is gone. The May 25 Huawei keynote said Beijing’s national-champion has a company-stated path to 2031.

The two questions worth tracking from this point forward:

• Does BIS add or remove entities in the back half of 2026? The Entity List is the public ledger of whether the sanctions architecture is being expanded or quietly walked back. Adds are the leading indicator that the architecture is still being enforced; removes (or license-broad grants that functionally undo a listing) are the leading indicator that it isn’t.
• What independent teardown emerges from the fall-2026 Kirin launch? Huawei has said the fall-2026 Kirin will be the first LogicFolding-derived commercial product. TechInsights will tear it down within weeks of release. That teardown — not the ISCAS keynote — will be the first independent evidence of whether the architectural claims have substance.

This page will be updated when either signal lands.