What began in 2022 as a targeted set of US export controls on advanced chips bound for China has become, by 2026, something closer to a structural realignment of the entire $600 billion global semiconductor industry — and the most important shift this year isn't a new restriction, it's that Washington's own policy has become genuinely contradictory, oscillating between tightening and loosening within the same twelve-month period.
I've followed the export control timeline, the congressional pushback, and China's domestic manufacturing progress closely through 2026. Here's an even-handed breakdown of where this stands — including the specific policy reversal that has confused industry analysts and lawmakers alike, and what China has and hasn't actually achieved under years of restriction.
How We Got Here: The Restrictions Didn't Start With a Single Decision
The chip war's foundation was laid in October 2022, when the Biden administration's Bureau of Industry and Security (BIS) rolled out export controls targeting advanced logic chips, integrated circuits, and semiconductor manufacturing equipment bound for China. The stated US rationale, articulated by officials including then-Assistant Secretary of Commerce Thea Rozman Kendler, centered on national security: advanced semiconductors enable AI capabilities that could improve Chinese military decision-making, autonomous weapons systems, and — per the policy's framing — facilitate human rights abuses through expanded state surveillance capacity. China's government has consistently characterized these same measures as protectionist overreach designed to preserve American technological dominance rather than address genuine security concerns, and has pursued WTO challenges and retaliatory measures on that basis.
The restrictions targeted AI training hardware above specific computing-power thresholds, with Nvidia bearing the most direct commercial impact since its GPUs were the primary product affected. Nvidia's initial response was pragmatic rather than confrontational: it engineered the H20, a chip purpose-built for the Chinese market and specifically calibrated to fall just below the export control thresholds. For a period, this workaround functioned as intended by all parties — Chinese hyperscalers including Alibaba, ByteDance, and Tencent continued purchasing at scale, generating real revenue for Nvidia while nominally complying with the letter of the restrictions.
The 2026 Policy Whiplash: Banned, Unbanned, Tariffed, Contested
This is the single most important development to understand about 2026 specifically, and it's worth laying out the sequence plainly because the speed of reversal has been genuinely unusual even by this policy area's already volatile standards. In December 2025, the Trump administration reversed the Biden-era "presumption of denial" specifically for Nvidia's H200 chip exports to China. In January 2026, BIS formally shifted its licensing posture to case-by-case review, but attached substantial conditions: third-party testing in the US before any export, a volume cap limiting China-bound shipments to 50% of Nvidia's domestic US sales, and a 25% tariff on each shipment with that tariff revenue flowing directly to the US Treasury. President Trump framed the arrangement in explicitly transactional terms, stating "we're allowing them to do it, but the United States is getting 25% of the chips, in terms of the dollar value."
The commercial stakes behind this reversal were substantial enough to explain the policy attention: Chinese technology companies reportedly placed orders for more than 2 million H200 chips for 2026, and Nvidia CEO Jensen Huang has previously estimated the broader China market could be worth $50 billion annually to the company. But the reversal didn't resolve cleanly even on its own terms — China's government has reportedly indicated it will approve H200 purchases only under exceptional circumstances, with reported friction over concerns about Nvidia's hardware tracking capabilities embedded in the chips. This illustrates a genuinely two-sided dynamic worth taking seriously: even when Washington opens a commercial door, Beijing may choose not to walk through it, for reasons rooted in its own domestic industrial policy goals of building self-sufficient semiconductor capability rather than continued reliance on US-origin chips with built-in tracking or compliance conditions attached.
Congress vs. the White House: A Policy Fight That's No Longer Bipartisan-Simple
What was once a relatively unified bipartisan consensus around restricting China's advanced chip access has genuinely fractured by 2026 into what one analysis described as a triangular standoff between Congress, the White House, and the chip industry itself. The central legislative vehicle is the AI Overwatch Act, introduced by Rep. Brian Mast (R-Fla.) and advanced by the House Foreign Affairs Committee on January 22, 2026. The bill would grant both chambers of Congress a 30-day review window with veto power over export licenses for advanced AI chips to "foreign adversaries" — modeled explicitly on existing congressional oversight mechanisms for arms sales — and would impose a mandatory denial requirement for any chip more powerful than the H200, which would include Nvidia's flagship Blackwell architecture specifically.
This legislative pushback reflects a genuine, substantive disagreement about strategy rather than partisan theater. In February 2026, a bipartisan group of eight lawmakers wrote directly to Commerce Secretary Howard Lutnick and Secretary of State Marco Rubio, arguing that the current entity-specific control regime has fundamentally failed, since once specialized manufacturing equipment crosses into China, the US government has extremely limited practical ability to enforce end-use restrictions afterward. Their letter specifically argued for country-level controls on "chokepoint" tools — equipment China currently has no domestic equivalent for — rather than continuing the current patchwork of entity-by-entity restrictions, which they characterized as a permanent and irreversible loss of American leverage with each new chokepoint tool that enters the country regardless of its nominal end-use restrictions.
The White House's competing position, by contrast, has prioritized stable trade negotiations ahead of President Trump's planned visit to Beijing, with the administration publicly downplaying chip-restriction confrontation and approving exports of higher-tier chips while suspending further new restrictions through much of 2026. The Department of Commerce, caught between these two pressures, appears to be pursuing a middle path: avoiding new restrictive rules that would complicate trade talks, while demonstrating enforcement resolve on existing rules specifically to pre-empt the kind of congressional override that the AI Overwatch Act represents. In June 2026, BIS issued a clarifying notice affirming that licensing requirements for advanced AI chips apply to all businesses headquartered in China, including their subsidiaries located outside Chinese territory — a clarification widely interpreted as closing a loophole that had reportedly allowed continued chip flows to China-linked entities operating internationally.
What China Has Actually Achieved Under Restriction — and What It Hasn't
This is where an even-handed accounting matters most, because both "the restrictions are clearly working" and "China is easily working around them" are oversimplifications of a more genuinely mixed picture. China's domestic champion, SMIC (Semiconductor Manufacturing International Corporation), has made real, independently verified progress: the company has pushed production to the 7-nanometer node using sophisticated multi-patterning techniques on deep ultraviolet (DUV) lithography tools that were already inside China before export controls tightened around extreme ultraviolet (EUV) equipment specifically. SMIC has reportedly even produced some 5-nanometer chips using DUV-based workarounds, though independent analysts are clear that this comes at meaningfully higher cost, substantially lower yields, and slower iteration cycles than EUV-based production at TSMC or Samsung.
The capacity numbers illustrate both the progress and its limits simultaneously. SMIC's total advanced-node capacity is estimated at roughly 45,000 wafer starts per month in 2025, expanding toward 60,000 by the end of 2026 — genuine growth, but a small fraction of what TSMC alone produces at advanced nodes. In March 2026, Reuters reported that Hua Hong Group, China's second-largest chipmaker, is preparing 7nm production capacity at its Huali Microelectronics facility in Shanghai, which would meaningfully break SMIC's current monopoly on Chinese advanced-node manufacturing if it materializes as reported. Separately, estimates cited in 2026 industry analysis suggest China's overall capacity to produce genuinely advanced chips sits at roughly 1–4% of US production capacity, a figure some analysts project declining further to 1–2% in 2026 specifically because US and allied chipmakers continue scaling faster than China's domestic alternatives can close the gap.
One structural argument made by some China-policy analysts deserves inclusion precisely because it's a genuinely different framing than the "is the gap closing or not" question: that the restrictions function less like a temporary tariff and more like what one analysis termed an "industrial firewall" — preventing China from absorbing the kind of iterative learning loops that historically allowed it to catch up in other sectors (telecommunications equipment, electric vehicles) through legitimate market access and reverse engineering. Under this framing, the relevant metric isn't whether China can produce some advanced chips today, but whether it can replicate the cumulative manufacturing-equipment expertise that took decades for the US, Netherlands, and Japan to build — a different and substantially harder challenge than catching up in chip design alone. Commerce Department data and independent analyses cited in 2026 reporting indicate US and allied firms still control roughly 90% of global semiconductor manufacturing equipment and approximately 92% of overall supply chain value, underscoring the depth of Chinese dependency specifically on the equipment layer rather than just the chip-design layer.
Market Share: Nvidia's China Position Has Genuinely Eroded
One concrete, measurable indicator worth highlighting: Nvidia, which once commanded over 90% of the Chinese AI chip market, has seen that share decline to approximately 50% as of early 2026, according to industry tracking. This decline reflects a combination of factors rather than any single cause — the 25% tariff attached to H200 shipments, Beijing's own "buy local" industrial policy mandates encouraging Chinese firms toward domestic alternatives, and the general uncertainty created by a full year of policy back-and-forth that has pushed Chinese hyperscalers to accelerate investment in domestic alternatives, particularly Huawei's Ascend chip line, regardless of whether US licenses are ultimately granted or denied.
This points to a dynamic that several 2026 analyses flag as genuinely important and easy to overlook in the day-to-day policy coverage: market share lost to domestic Chinese alternatives during a period of US policy uncertainty doesn't automatically return once licenses are eventually granted or restrictions eventually loosened. Once a Chinese hyperscaler has invested in retraining its engineering teams, rebuilding software toolchains, and validating production workloads on Huawei's Ascend architecture instead of Nvidia's CUDA ecosystem, reversing that switch carries real switching costs that persist independent of whatever the current export license status happens to be. This is one of the more durable, structural consequences of 2025–2026's policy volatility, regardless of how the specific restrictions evolve from here.
The Equipment Chokepoint: Where Enforcement Is Genuinely Difficult
It's worth being direct about the area where US policy critics, including the bipartisan congressional letter referenced above, have the most legitimate technical complaint: semiconductor manufacturing equipment, not finished chips, may represent the more genuine and harder-to-close chokepoint in this entire competition. ASML, the Dutch company that holds an effective monopoly on the most advanced EUV lithography systems required for cutting-edge chip production, has continued selling less-advanced but still meaningful lithography equipment to China — reportedly doubling sales of advanced (non-EUV) lithography equipment to China from 2022 to 2023, and again from 2023 to 2024, according to the congressional letter's own figures. Because this equipment, once inside China, is extremely difficult for the US to monitor or enforce end-use restrictions on afterward, critics argue that the current entity-by-entity control approach systematically underestimates how much cumulative manufacturing capability China continues to accumulate regardless of chip-level export restrictions.
Reading the Sequence: Managed Bifurcation, Not Clean Decoupling
The most analytically careful 2026 assessments converge on describing the current state as "managed bifurcation" rather than either a clean technological decoupling or a return to the pre-2018 status quo of relatively open semiconductor trade. Neither side appears to be pursuing — or able to achieve — a complete severing of the relationship: full decoupling would meaningfully hurt US toolmakers and chip designers (who derive substantial revenue from Chinese sales) nearly as much as it would hurt their Chinese customers, which is precisely why Washington has repeatedly opted for conditional, licensed, tariffed access rather than an absolute freeze, even as congressional hawks push for exactly that kind of harder line.
At the same time, the broader trajectory described across multiple independent 2026 analyses is one of accelerating, structural separation rather than stabilization — both the US and China are simultaneously building increasingly independent supply chains, independent chip architectures, and, in the most consequential long-term sense, increasingly incompatible AI infrastructure ecosystems built around different chip platforms (Nvidia/CUDA on one side, Huawei Ascend and other domestic Chinese alternatives on the other). The practical reality for global enterprises and investors is that policy unpredictability itself — not any single restriction or its reversal — has become a permanent operating condition for companies like Nvidia and AMD doing business with China, rather than a temporary headwind that will resolve into a stable long-term equilibrium anytime soon.
What This Actually Means for Different Audiences
For understanding the chip industry broadly: this is best understood not as a story with a winner and loser determined yet, but as a structural realignment of a $600 billion global market whose competitive contours through roughly 2035 are actively being shaped by decisions made in 2025–2026 specifically. Both the US national-security rationale and China's characterization of the restrictions as protectionist contain genuine substance from each government's own stated perspective, and reasonable analysts disagree sharply about which framing better predicts the next several years.
For evaluating Nvidia and AMD's China exposure specifically: policy unpredictability is now a structural feature of their China strategy rather than a temporary headwind, and the H200 framework's commercial pathway back into China comes with real margin-reducing conditions (the 25% tariff, the 50% volume cap relative to domestic sales) attached. Market share already lost to Huawei's Ascend chips during the uncertainty period is unlikely to fully reverse even if licensing conditions ease further.
For evaluating China's domestic semiconductor progress: meaningful, independently verified gains have occurred at SMIC and potentially Hua Hong specifically, but those gains remain concentrated at older process nodes using workaround techniques that carry real cost and yield penalties relative to TSMC's EUV-based production. The gap at the most advanced nodes, where genuine technological frontier competition occurs, remains substantial and is not closing quickly by most independent estimates, even as the gap at less-advanced "legacy" nodes — which still represent the bulk of real-world chip demand for automobiles, consumer electronics, and industrial applications — has narrowed considerably less dramatically than headline coverage of the restrictions sometimes implies.
FAQ
What is the H200 chip controversy in 2026?
In December 2025, the Trump administration reversed Biden-era restrictions on Nvidia's H200 chip exports to China, then in January 2026 created a licensing framework requiring third-party US testing, a volume cap limiting China shipments to 50% of domestic US sales, and a 25% tariff with revenue flowing to the US Treasury. The policy has drawn bipartisan congressional criticism, with some lawmakers arguing it represents a dangerous loosening of restrictions on chips powerful enough to advance Chinese AI and military capabilities.
What is the AI Overwatch Act?
A bill introduced by Rep. Brian Mast and advanced by the House Foreign Affairs Committee in January 2026 that would give Congress 30 days to review and potentially block any export license for advanced AI chips to foreign adversaries, modeled on existing oversight of arms sales. It would also impose a mandatory denial requirement for chips more powerful than the H200, including Nvidia's Blackwell architecture, and is currently in conflict with the White House's more permissive licensing approach.
How advanced is China's domestic chip manufacturing in 2026?
SMIC has reached 7-nanometer production using DUV multi-patterning workarounds and has reportedly produced some 5-nanometer chips, though at meaningfully higher cost and lower yield than TSMC or Samsung's EUV-based production. Estimates suggest China's overall advanced-chip production capacity remains at roughly 1–4% of US capacity, with some analysts projecting that share narrowing further to 1–2% in 2026 as US and allied capacity continues scaling.
Has Nvidia lost market share in China?
Yes, substantially. Nvidia once commanded over 90% of the Chinese AI chip market but has seen that share decline to approximately 50% as of early 2026, driven by the 25% tariff on H200 shipments, Beijing's domestic-preference industrial policy, and a year of US policy uncertainty that pushed Chinese hyperscalers toward domestic alternatives like Huawei's Ascend chip line.
Why is semiconductor manufacturing equipment considered a bigger chokepoint than chips themselves?
Once specialized manufacturing equipment crosses into China, the US has limited practical ability to enforce end-use restrictions on how it's subsequently used. A bipartisan group of lawmakers has specifically criticized the current entity-by-entity control approach, citing reports that ASML's sales of advanced lithography equipment to China roughly doubled year-over-year in both 2023 and 2024, arguing that country-level controls on "chokepoint" equipment would be more effective than the current patchwork system.
Is the US-China chip conflict likely to result in full technological decoupling?
Most 2026 analysis describes the trajectory as "managed bifurcation" rather than clean decoupling — both governments continue allowing conditional, licensed, and tariffed chip trade rather than a complete freeze, partly because full decoupling would meaningfully hurt US chip designers and toolmakers who derive substantial revenue from Chinese sales. At the same time, both countries are simultaneously building increasingly independent and incompatible supply chains and AI infrastructure ecosystems, suggesting accelerating long-term separation even without a single dramatic break.