Version 4

The US chip export restriction debate has entered a new phase marked by concrete customer data, a quantified performance benchmark, and active Congressional legislation. Nvidia has ceded China's AI chip market to Huawei[1][2], and that transfer is now confirmed at the customer level: ByteDance has placed a $5.7 billion order for Huawei AI chips[14], and both ByteDance and Alibaba are planning orders for Huawei's newest chip[15], with China's government reportedly directing major tech firms to stop purchasing Nvidia hardware altogether[16]. DeepSeek research now places Huawei's Ascend 910C at approximately 60% of Nvidia H100 inference performance[17], partially answering the long-standing question about the performance gap. On the legislative front, the House Foreign Affairs Committee passed export control legislation in late April 2026[21], multiple bipartisan bills have been introduced[19][20], and a revenue-sharing model — under which Nvidia and AMD would pay 15% of China AI chip sales to the US government — has been floated as a possible compromise[25].

The story has moved from projection to documentation: the market transfer to Huawei is now confirmed by customer procurement decisions worth billions of dollars, and Huawei's chip performance is now close enough to Nvidia's for production workloads. The policy debate is no longer about whether restrictions will work — they haven't contained Huawei — but about what mechanism the US government will use to manage a situation where its hardware firms are excluded from a massive market while a rival fills the gap. The revenue-sharing model and active Congressional markups suggest the US is groping toward a new policy framework rather than maintaining or reversing the existing one.

The US government's campaign to restrict China's access to advanced AI chips has produced a market outcome that is now documented at both the supply and demand levels. Nvidia, which held roughly 95% of China's AI accelerator market before restrictions tightened, has exited that market entirely. CEO Jensen Huang confirmed the concession publicly in a CNBC interview, calling the export control regime a "failure"[1][2], and Barron's characterized Huang as having effectively won the policy argument[3]. AMD CEO Lisa Su has moved to a similar position, warning against strict controls and reportedly making progress with the US government on easing restrictions[4][5]. On the other side, Anthropic CEO Dario Amodei conducted a personal Congressional lobbying campaign in February 2026, publicly named Nvidia as enabling Chinese AI compute access, and framed semiconductor access as potentially the "only advantage" the US retains over China in AI[6][7][8].

Huawei has absorbed the market Nvidia vacated. Multiple independent estimates place its share of China's AI chip market between 41% and 60%[9][10][11], with $12 billion in projected 2026 AI chip revenue[12] and plans to double Ascend chip output[13]. The commercial transition is now confirmed by customer procurement decisions: ByteDance has reportedly placed a $5.7 billion order for Huawei AI chips[14], and both ByteDance and Alibaba are planning orders for Huawei's newest chip[15]. China's government has separately been reported to have directed major tech companies — including Alibaba, ByteDance, and Baidu — to stop purchasing Nvidia hardware[16], formalizing a domestic-preference push that aligns commercial incentives with national policy. The performance gap between Huawei's chips and Nvidia's banned hardware is now partially quantified: DeepSeek research suggests the Ascend 910C delivers approximately 60% of Nvidia H100 inference performance[17], and a comparison of the newer Ascend 910D against the H100 suggests the gap may be narrowing further[18]. At 60% parity, Huawei's hardware is functional for a wide range of production inference workloads.

On the legislative side, the debate has moved from lobbying to active markup. Representative Baumgartner introduced a bipartisan bill to tighten controls on sensitive chipmaking equipment in early April 2026[19], followed by the MATCH Act introduced by Senators Risch, Ricketts, and Kim on April 8[20]. The House Foreign Affairs Committee passed export control legislation at an April 22 markup[21], supported by advocacy organizations including FDD Action[22][23], and a separate congressional hearing addressed chipmaking tool export control loopholes[24]. This represents the most concrete legislative movement yet on the pro-restriction side. Simultaneously, a revenue-sharing model has emerged as a potential middle path: reports indicate Nvidia and AMD could pay 15% of China AI chip sales revenues to the US government[25], a framework that would allow resumed sales while capturing a financial return for the government.

The regulatory baseline has also shifted. The Bureau of Industry and Security revised its export review policy for advanced AI chips destined for China and Macau[26][27], though the direction of that revision is contested. The Trump administration's record remains contradictory: it reversed Biden-era restrictions in early 2026[28], then imposed the H20-specific ban that triggered $5.5 billion in Nvidia inventory charges and up to $15 billion in projected lost sales[29][30], and may now be considering the revenue-sharing compromise. The Council on Foreign Relations has characterized the framework as "strategically incoherent and unenforceable"[31], a judgment that the oscillating policy record continues to support. Aaron Friedberg's contrarian thesis — that restrictions have counterproductively accelerated Chinese semiconductor self-sufficiency rather than containing it — is now backed by concrete data: China's semiconductor equipment self-sufficiency has reached 35%[32], and Huawei's documented market dominance and customer adoption confirm structural realignment rather than a temporary gap-fill.