Captive Market: How US Export Controls and Beijing's Self-Sufficiency Push Drove Cambricon's 108% Revenue Surge to ¥6 Billion in H1 2026

Cambricon Technologies reported H1 2026 revenue of RMB 5.996 billion (up 108.13% YoY) and net profit of RMB 2.311 billion (up 122.61% YoY), with adjusted non-GAAP profit rising even faster at 137.3% to RMB 2.166 billion . The South China Morning Post reported that Cambricon "capitalized on a nationwide rush to replace restricted foreign technology" as Beijing pushed state-owned enterprises to substitute foreign chips with domestic alternatives . The surge was not driven by a single product win—it was the direct result of a structural demand shock created by the convergence of US export controls and Beijing's domestic substitution mandates .

Direct Drivers of the Surge

Massive volume ramp of MLU-series chips. Cambricon's MLU590 accelerator, built on a domestic 7nm SMIC process and reported to approach ~70% of Nvidia A100 inference performance, became the chip of choice for Chinese data centers unable to buy Nvidia's best . The company planned to more than triple output to 500,000 AI accelerators in 2026 , and its market share tracked that ambition: IDC data showed Cambricon shipped roughly 116,000 cards in 2025, ranking alongside Baidu's Kunlunxin .

Explosive nationwide AI infrastructure buildout. China's data center and AI computing market exceeded ¥500 billion in 2025, with domestic AI chips alone accounting for approximately ¥100 billion . Chinese chip suppliers—led by Huawei and Cambricon—captured nearly 80% of the domestic AI server market by mid-2026 , compared with negligible domestic share just two years prior.

Sustained 55% gross margins across seven consecutive quarters, driven by pricing power in a supply-constrained market where domestic alternatives face foundry capacity limits at TSMC and SMIC rather than design bottlenecks .

Government subsidies were a minor factor. Only ¥91.72 million in subsidies (about 4% of net profit) flowed into the result, confirming that organic commercial demand, not direct state handouts, drove the profit surge .

The US Export Control Catalyst

US bans locked out Nvidia's best. The H100 and A100 have been formally barred from China since October 2023 under a presumption-of-denial policy . Even when the Trump administration approved H200 sales to 10 Chinese firms in late 2025, Beijing instructed domestic buyers to reject them . Nvidia itself reported it had yet to generate any revenue from H200 sales to China as of February 2026 . By mid-2026, Nvidia's China AI chip revenue was on track to fall from ~$17 billion toward near zero .

The May 2026 loophole closure. On May 31, 2026, the US Commerce Department closed the last major loophole—Chinese firms purchasing advanced Nvidia chips through overseas subsidiaries—ensuring a permanent wall between Nvidia and the Chinese market .

The result was a captive market. Chinese cloud providers and AI companies had no choice but to buy domestic. Cambricon, Huawei (Ascend), Alibaba (T-Head), and Baidu (Kunlunxin) collectively claimed nearly half the local market by April 2026, up from roughly 16% domestic self-sufficiency in Q4 2024 . The Brookings Institution summarized the situation starkly: "Ball game's over—the US is out of the AI chip market in China" .

Beijing's Tech Self-Sufficiency Push

Policy-driven replacement of foreign hardware. The SCMP reported that Cambricon "capitalized on a nationwide rush to replace restricted foreign technology" as Beijing pushed state-owned enterprises and critical infrastructure operators to substitute foreign chips with domestic alternatives . China's Ministry of Industry and Information Technology issued a new procurement list certifying domestic AI chips as "secure and reliable" for government purchase, with only two companies making the initial cut: Cambricon and Huawei .

Domestic self-sufficiency accelerated dramatically. China's chip self-sufficiency rate hit 28% in Q4 2025, up from 16% a year earlier . Chinese chip suppliers—led by Huawei and Cambricon—were forecast to capture nearly 80% of the domestic AI server market in 2026, further squeezing Nvidia, AMD, and other foreign suppliers . TrendForce later revised its forecast: domestic solutions are now expected to command close to 90% of China's high-end accelerator market in 2026, with foreign vendors retaining roughly one-tenth of a market growing at more than 83% year-on-year in unit shipments .

R&D spending rose nearly 30% in H1 2026, indicating Cambricon is reinvesting heavily into its next-generation Siyuan 690 chip, expected in H2 2026 and reportedly designed to rival Nvidia's H100 in performance . The company also deepened cooperation with leading financial and internet companies, driving large-scale deployment of its products across both sectors .

Caveats and Risks

Sequential growth slowed. The stock fell ~7% after the earnings release as the pace of quarter-over-quarter growth decelerated, suggesting the initial demand rush may be maturing . While revenue and profit more than doubled year-on-year, the sequential quarterly growth came in "far below what analysts had anticipated" .

Working capital strain. Cambricon's inventory ballooned to ¥8.25 billion (an 83% quarter-over-quarter jump) and operating cash flow fell 65.83%, signaling that it is building stockpiles faster than it can convert them to cash—a potential risk if demand softens or if Huawei's dominant ~49% market share squeezes Cambricon's growth runway .

Foundry bottleneck. SMIC's 7nm capacity, not chip design, is the binding constraint . With Huawei and Alibaba also competing for the same limited capacity, Cambricon's ability to sustain 100%+ revenue growth depends on SMIC scaling output—which remains uncertain under US equipment restrictions .

In short, Cambricon's H1 2026 results are a direct reflection of a captive market created by US export controls and turbocharged by Beijing's domestic substitution mandates. The company did not suddenly build a better chip than Nvidia—it simply became the best available alternative in a market from which Nvidia was forcibly excluded.