The Silicon Curtain: How Export Controls Reshaped the Chip Industry

From blanket bans to case-by-case deals, the US semiconductor strategy toward China has undergone a dramatic transformation. The consequences are reverberating through every layer of the global technology supply chain.

On January 13, 2026, the US Department of Commerce's Bureau of Industry and Security published a rule that would have been unthinkable two years earlier. After years of tightening restrictions on advanced chip exports to China, Washington formally opened a pathway for Nvidia to sell its powerful H200 processor to Chinese customers — under strict conditions.

The next day, Chinese customs authorities told agents the chips were not permitted to enter the country.

That whiplash sequence — America offering to sell, China refusing to buy — captures the strange new reality of the semiconductor trade. What began in October 2022 as a straightforward American effort to deny China access to cutting-edge chips has evolved into something far more complex: a shifting, contradictory, and deeply consequential struggle over who controls the world's most important technology.

This is the story of how we got here, what the rules look like now, and why it matters.

The Origins: October 2022

The modern chip war began with a single sweeping decision. In October 2022, the Biden administration imposed the most aggressive semiconductor export controls in decades. The rules restricted China's access to advanced AI chips, the equipment used to manufacture them, and the expertise of American engineers working in Chinese fabs.

The logic was clear: advanced semiconductors power everything from artificial intelligence to precision weapons. By choking off China's access to the most advanced chips and the tools to make them, Washington aimed to freeze Beijing's progress at the technological frontier.

The immediate targets were Nvidia's A100 and H100 processors — the workhorses of AI training — and the lithography machines made by the Dutch firm ASML, without which no country can manufacture chips at the leading edge. The Netherlands and Japan, under heavy US pressure, agreed to impose parallel restrictions on their own equipment exports.

The controls were updated and tightened in October 2023 and again in late 2024, closing loopholes that Chinese firms and American chipmakers had exploited to continue trading.

The Chip Designers' Response

American chip companies did not accept the restrictions passively. Nvidia, the dominant force in AI processors, responded by designing modified chips calibrated to fall just below the control thresholds. The A800 and H800 were created specifically for the Chinese market — less powerful than their unrestricted counterparts, but still capable of supporting significant AI workloads.

When Washington tightened the thresholds again, Nvidia designed the H20 — a further downgraded chip that could legally enter China. Nvidia's leadership has consistently maintained that overly restrictive controls risk pushing Chinese customers toward domestic alternatives without meaningfully slowing Beijing's AI progress.

On the Chinese side, Huawei emerged as the flagship effort at chip self-sufficiency. Its Ascend 910 series processors, manufactured by China's SMIC on relatively mature 7nm-class processes, became the standard-bearer for domestic AI compute. But the performance gap remained vast. According to publicly available benchmarks, Huawei's AI chip output amounts to roughly 1 to 4 percent of Nvidia's aggregate computing power — and that gap is projected to widen through 2027 (Council on Foreign Relations, December 2025).

A New Policy Direction

American chipmakers were losing billions in potential Chinese revenue. Allies were grumbling about the compliance burden. And China's domestic chip industry, while still far behind, was making incremental progress — partly spurred by the very restrictions designed to contain it.

On December 8, 2025, President Trump announced a new policy: the United States would allow Nvidia to sell H200 chips to approved Chinese customers, subject to security conditions and a 25 percent revenue fee to the US government.

The formal regulation published in January 2026 established the new framework (Bureau of Industry and Security, January 15, 2026). License applications for the H200 and AMD MI325X — chips with a total processing performance below 21,000 and DRAM bandwidth under 6,500 GB/s — would be reviewed case by case rather than automatically denied. Exporters had to certify that domestic US demand was met first, that Chinese buyers had adequate security procedures, and that chips were tested by independent US labs before shipment. China-bound volumes could not exceed 50 percent of domestic US sales.

The policy shift triggered immediate debate. The Council on Foreign Relations published an analysis describing the new framework as "strategically incoherent," warning that even under the volume caps, China could receive enough H200 chips to increase its total installed AI compute by 250 percent in a single year (CFR, January 14, 2026). The tension reflects a fundamental disagreement within Washington: Commerce officials and industry argue that some controlled trade is better than driving China to full self-sufficiency, while hawks in Congress and the national security establishment counter that any advanced chips sold to China will inevitably support military and surveillance applications.

Beijing's Countermove

China's response to the January relaxation was unexpected: rejection. Chinese customs authorities instructed agents not to clear H200 shipments. State media outlets had already labeled Nvidia's less powerful H20 chip as a potential security risk, citing concerns about tracking capabilities and remote access.

The message was strategic. By refusing American chips on its own terms, Beijing signaled that it would not accept a supply relationship that could be turned on and off at Washington's discretion. Instead, China is doubling down on domestic alternatives.

Huawei's Ascend 910C is being actively promoted as the national standard for AI compute. Current benchmarks suggest it reaches roughly 60 percent of the H100's inference performance — a significant gap, but one that Chinese engineers are working to close. The Ascend 920, expected in late 2026, aims for closer parity.

The result is an accelerating decoupling. Western AI labs optimize for Nvidia's CUDA ecosystem. Chinese labs are building around Huawei's architecture. Two parallel technology stacks are emerging — what some analysts have begun calling the "Silicon Curtain."

The TSMC Factor

At the center of all this sits Taiwan Semiconductor Manufacturing Company. TSMC fabricates the vast majority of the world's most advanced chips, including Nvidia's entire product line. This concentration makes Taiwan the most strategically sensitive piece of territory in the global economy.

The United States has invested heavily to reduce this dependency. The CHIPS and Science Act of 2022 allocated over $52 billion in subsidies to build domestic fabrication capacity. TSMC is constructing plants in Arizona. Intel is expanding in Ohio. Samsung is building in Texas.

But leading-edge chip fabrication is extraordinarily difficult to relocate. TSMC's new Arizona facility has faced delays and cost overruns, and industry experts estimate it will take years before American fabs can match the output and yield rates of Taiwanese facilities.

Until then, the uncomfortable reality persists: the most critical input for both American AI dominance and Chinese AI ambition is manufactured on an island that both superpowers consider strategically vital.

What Happens Next

The semiconductor export control regime is now in permanent flux. The January 2026 rules are widely expected to be revised again — possibly tightened by congressional action, possibly loosened by further executive decisions. Companies across the supply chain face what industry analysts describe as a new category of risk that is not technical but regulatory, interpretive, and highly political.

For enterprise technology leaders worldwide, the lesson is clear: infrastructure planning can no longer be built on static assumptions. The specs of a chip matter less than the geopolitical narratives surrounding it.

For China, the path toward semiconductor self-sufficiency is long but irreversible. Every restriction accelerates the political will and state funding directed at domestic alternatives.

And for the United States, the fundamental strategic question remains unresolved: is it better to maintain controlled trade that preserves leverage and revenue, or to pursue maximum restriction that risks pushing China to build the very capabilities Washington seeks to deny?

This is part of Tech Cold War's ongoing coverage of the semiconductor battleground. Subscribe to receive weekly analysis on the technologies reshaping global power.