For four years, Washington has attempted to build a sophisticated industrial wall. This blockade aimed to stop China from getting the computer chips needed for modern military power and artificial intelligence (AI). This policy is failing in mid-2026. The wall is breaking, and the United States is now moving from total containment towards a model of managed trade. What began as an attempt to freeze China’s semiconductor capability in place has instead produced a rival ecosystem, built with state money and running on its own timeline.
How Beijing Responded
The export controls from October 2022 rested on the idea that leading in AI is a matter of national security. However, this led to confusion. Washington tries to restrict Chinese access, yet it also authorizes sales of high-performance chips that were supposedly security threats only months before.
Beijing responded with massive state money. The Ministry of Finance launched Big Fund III with 344 billion yuan, or roughly 47.5 billion dollars, to build a self-reliant supply chain. This forced an industrial mobilization. Huawei now acts as a systems architect and runs its own manufacturing facility in Guanlan. Through its investment arm, Hubble, the company provided capital to more than 60 domestic firms that specialize in materials, chemicals, and equipment.
Stay ahead of the geopolitical week.
MD Briefing delivers expert analysis across five global fronts — the Indo-Pacific, energy, geoeconomics, European security, and the Middle East — every Monday morning. Free.
Chinese firms also reacted with speed. Semiconductor Manufacturing International Corporation (SMIC) is growing its output and expects to reach 80,000 wafers per month by 2027. While the yields remain lower than those using extreme ultraviolet (EUV) lithography machines, the volume of output is changing the competitive landscape. Huawei is using a new design approach to pack more transistors into the same space. This design stacks layers in three dimensions. It claims to achieve 55 per cent higher density than conventional layouts. Huawei aims for performance equivalent to 1.4nm chips by 2031 by changing how the chip is built. The market responded, with stock prices for SMIC and Hua Hong surging more than 16 percent in Hong Kong after the announcement.
This result represents a strategic failure for American planners. Nvidia’s share of the Chinese AI accelerator market had already slid from more than 90 per cent to roughly 50 per cent in early 2026. By May, Jensen Huang confirmed that the figure had fallen to zero.
Beijing’s Countermove
China is not simply building substitutes. It is also tightening its own leverage. On 22 June 2026, Beijing added ten American firms, including a rare-earth mining company, to its own export control list. The order claims extraterritorial reach, barring any organization or individual anywhere from transferring dual-use materials that originated in China. Two days later, the Ministry of Commerce announced a whistleblower mechanism for reporting violations of controls on strategic minerals, complete with a dedicated hotline, effective from 1 July 2026. The mechanism mirrors the enforcement architecture Washington has spent years building. Both capitals are now policing not just their own exports but also the compliance of firms and individuals well beyond their borders.
The Memory Bottleneck
The industry now faces a secondary hurdle in high-bandwidth memory (HBM). This is a type of memory that stacks chips to increase speed and bandwidth for training large models. Because export restrictions made firms like Samsung and SK Hynix unreliable suppliers, Chinese companies are looking inward. ChangXin Memory Technologies (CXMT) is the only domestic player filling this gap, though production remains limited. The company will produce 2 million HBM stacks in 2026, enough for 300,000 Ascend 910C packages, which are Huawei’s premier training processors. SMIC can produce enough parts for over 1 million chips every year, but this lack of memory limits production to a fraction of that volume.
A Policy Contradiction
Washington changed its policy in January 2026. It moved the Nvidia H200 and AMD MI325X from a denied status to a review process, allowing exports of higher-capability chips to maintain corporate revenue. The Council on Foreign Relations noted a flaw: a policy cannot be open, enforceable, and protective of national security at the same time.
Things have become paradoxical. The Department of Commerce opens legal channels for chip sales while it penalizes firms for past behavior. Applied Materials paid 252.5 million dollars in February 2026 for illegal exports. Furthermore, on 1 June 2026, the government closed loopholes that allowed chip servers to reach China through subsidiaries in hubs like Malaysia, Singapore, and the United Arab Emirates. The government fines companies for old actions while it formalizes channels for new trade.
The contradictions have not eased since. The same 1 June guidance affirmed that licensing requirements apply to any company headquartered in China, regardless of where its subsidiaries operate, an attempt to close yet another loophole. In the Senate, the bipartisan Stop Stealing Our Chips Act, which would create a formal whistleblower bounty program within the Bureau of Industry and Security, passed unanimously on 22 May 2026 but has yet to clear the House. Congress is trying to build the enforcement machinery that the executive branch’s own licensing decisions keep undermining.
What Washington Should Do Instead
Analysts at the Brookings Institution now argue the shift is decisive. American firms no longer hold a meaningful share of China’s AI chip market, an outcome that a policy of total restriction was designed to prevent. The lesson is not that export controls failed outright, but that a policy built purely on denial cannot outpace a rival with both the money and the will to build around it.
Technological blockades have a limited life. They give target nations an incentive to build their own alternatives. Washington treated semiconductors as a simple resource, and in doing so, it helped China create an independent tech ecosystem.
United States policymakers need a new way to gain a strategic advantage. They should stop trying to starve a competitor of silicon. Focus should shift towards sustaining American innovation. This requires prioritizing domestic research and development. Securing absolute dominance in foundational design software is also necessary, as is expanding the CHIPS Act to support long-term infrastructure. To protect the national interest, the United States needs to invest in its own foundational capacity rather than relying on blunt instruments to cut off its rivals.
