China's 50-year semiconductor ambition: from state fabs to Huawei's 'nuclear-level' AI cluster
Apr 16, 2025
Key Points
- China's semiconductor push is a 50-year state mandate dating to 1988, not a recent response to US sanctions—the government deployed $20 billion through the National Integrated Circuit Industry Investment Fund in 2014 to sustain fabs through market cycles.
- US export controls after 2022, including bans on Nvidia's flagship GPUs and restrictions on ASML's advanced lithography equipment, accelerated Chinese chip development rather than halted it; Huawei unveiled the Ascend 910C and CloudMatrix 384 as direct Nvidia alternatives.
- SMIC has advanced from 180nm to near-7nm production over 25 years and, while unable to match Taiwan's cutting edge, operates at scales and timelines immune to market pressure—a structural advantage Western competitors lack.
Summary
China's semiconductor push is not a recent awakening but a 50-year state mandate that has weathered trade wars, sanctions, and technology gaps with patient, orchestrated capital deployment.
The ambition started in 1988 when China's government designated semiconductors a pillar industry. By 1990, the state launched Project 9008 to build a world-class foundry, partnering with Bell Labs to acquire process knowledge. Execution stumbled when bureaucratic delays left the fab obsolete by production. Project 909, ordered by President Jiang Zemin in the mid-1990s, proved more successful and achieved a 0.5-micron process in Shanghai by 1997, but the company hit profitability walls during the dot-com crash, a persistent problem for non-leading-edge fabs facing commodity margins and heavy capex burdens.
China shifted strategy in the late 1990s toward chip design. Huawei's HiSilicon began supplying ASICs for telecom infrastructure, and design houses like Godson emerged. When China joined the WTO in the late 1990s, multinational firms established R&D centers and manufacturing hubs in high-tech parks. Intel's 2007 announcement of a 300mm fab in Dalian marked a turning point, as significant technology transfer to Chinese engineers accelerated. Nvidia, founded in 1993, began establishing R&D centers in Shanghai by the mid-2000s to support growing demand.
By 2004, the gap was stark. China consumed $40 billion in semiconductors annually but produced only $4 billion domestically, a 10% self-sufficiency rate the government deemed unacceptable. The 2008 financial crisis tested resolve. Rather than let fabs fail during demand dips, China deployed stimulus packages to keep semiconductor manufacturing alive. In 2014, the government created the National Integrated Circuit Industry Investment Fund with $20 billion directed toward fabs, design houses, and next-generation research. SMIC and HiSilicon received capital boosts.
Made in China 2025, unveiled in 2015, set the target at 70% domestic chip production by 2025. These state mandates carry enforcement pressure unlike US corporate timelines. Executives cycle through if targets slip. The party sets a goal, and the system delivers.
Nvidia deepened partnerships with Chinese cloud providers—Alibaba, Baidu, Tencent—as Huawei faced mounting US restrictions. In 2019, Huawei landed on the US entity list, barring access to American technology. Huawei's 2024 revenue reached $118 billion, entirely private. That forced HiSilicon into radical self-reliance. In 2020, the foreign direct product rule expanded, restricting advanced chips to Huawei. SMIC also faced sanctions in December 2020.
The 2021 US CHIPS and Science Act coordinated multilateral export controls with the Netherlands and Japan, blocking advanced lithography equipment like ASML's EUV machines, the most critical tools for sub-7nm production. In October 2022, the US banned Nvidia's A100 and H100 flagship GPUs to China. Nvidia responded with the A800, which has lower interconnect bandwidth to comply with export limits but can still train large models. Chinese firms stockpiled them.
In December 2022, SMIC was added to the entity list entirely. That accelerated Chinese companies' own chip development. By 2023, Huawei unveiled the Ascend 910C and the CloudMatrix 384, described by South China Morning Post as a "nuclear-level" product, aimed directly at breaking Nvidia's dominance.
The technology gap remains real but narrower than Western industry players suggest. SMIC has moved from 180nm to advanced nodes over 25 years and won't match Taiwan's 3nm anytime soon. A chip built to 7nm can serve automotive, IoT, and smart home applications, not AI training or flagship phones, but a useful market. More crucially, China's state-backed foundries are inefficient by Silicon Valley standards but operationally immune to market cycles. They accumulate capital relentlessly and cycle executives on fixed political timelines. The US woke late. Semiconductor decoupling accelerated after 2022, but China's 50-year plan was already underway.