In August 2022, the United States Congress passed the CHIPS and Science Act with a confidence bordering on optimism. A $52.7 billion investment — the largest direct industrial subsidy in American history — would rebuild domestic semiconductor manufacturing, reduce dangerous dependence on Taiwan, and deny China access to the chips powering the AI revolution. Four years on, the picture is complicated. Fabs are under construction, some producing chips, and supply chain geography is shifting. But the decoupling is partial, slower than planned, and generating geopolitical and economic dynamics that Washington did not fully anticipate.
The Arizona Milestone
The most visible symbol of the CHIPS Act is TSMC’s Fab 21 complex rising from the desert outside Phoenix. After two years of construction delays attributed to a shortage of skilled tradespeople familiar with advanced fab requirements — a gap so severe that TSMC flew in thousands of workers from Taiwan to train local crews — the facility began producing 4-nanometer chips for Apple in late 2024. By early 2026, it has ramped toward its target production volume, and TSMC has broken ground on a third Arizona fab that will eventually produce 2-nanometer process technology.
The numbers are significant. TSMC has committed more than $65 billion in total US investment, making it one of the largest foreign direct investment projects in American history. Commerce Secretary Gina Raimondo, now in her fourth year, called it “the most consequential industrial project since the interstate highway system” at a February ceremony in Phoenix. The chips coming off Fab 21’s lines are going into the iPhone 17 series and Apple’s M5 chip family — meaning that, for the first time in the company’s history, a material share of Apple’s most advanced silicon is being produced on US soil.
Intel’s Harder Road
If TSMC’s Arizona story is one of delayed but real progress, Intel’s parallel effort is more turbulent. The company received $8.5 billion in CHIPS Act direct grants and up to $11 billion in loan guarantees to build advanced fabs in Ohio (Fab 52 and Fab 62, under the New Albany campus) and expand existing facilities in Oregon and New Mexico.
The Ohio fabs were originally scheduled to begin 18A process production in 2025. That timeline slipped. Intel’s 18A node — widely seen as critical to the company’s ambition to compete with TSMC at the cutting edge — has faced yield challenges, and the company’s foundry business posted a $7 billion operating loss in 2025. CEO Pat Gelsinger’s departure in late 2024 added leadership uncertainty at a moment requiring maximum operational focus.
The stakes extend beyond Intel’s own balance sheet. The CHIPS Act was partly predicated on the assumption that Intel would emerge as a domestic alternative to TSMC for the most advanced nodes — providing a second source of leading-edge logic chips and reducing the strategic concentration risk that keeps US national security planners awake at night. If Intel’s foundry cannot achieve competitive yields at 18A, that strategic rationale partly collapses.
Export Controls and China’s Response
The other side of the decoupling equation is what the US and its allies have denied China. Beginning in October 2022 and tightening through 2023 and 2025, the Commerce Department’s export controls have progressively restricted China’s access to advanced logic chips above certain performance thresholds, the equipment needed to manufacture them, and the software tools used to design them.
The controls have had real effects. NVIDIA’s H100 and subsequent Blackwell-architecture GPUs — the primary workhorses of large-scale AI model training — are effectively banned from sale in China. ASML, the Dutch company that makes the extreme ultraviolet lithography machines essential for sub-5nm production, has been barred from exporting EUV systems to Chinese customers since 2019, and US pressure extended that restriction to its older deep ultraviolet systems in 2023.
China’s response has been to accelerate domestic substitution with a force and funding level that analysts are now treating seriously. Huawei’s semiconductor subsidiary HiSilicon has produced AI accelerator chips — the Ascend 910B and 910C series — that, while not matching NVIDIA’s performance per watt, have achieved sufficient capability to support commercial AI inference workloads for Chinese cloud providers including Alibaba, Tencent, and Baidu. SMIC, China’s leading fab, has reportedly achieved 7nm-equivalent production using multi-patterning techniques that bypass the need for EUV equipment, at volumes and yields that remain subjects of debate among semiconductor analysts.
The trajectory is significant. Export controls may have set Chinese AI capability back by two to three years, according to a January 2026 CSIS assessment. They have not reversed it, and they have provided an extremely powerful political argument inside China for sustaining what is now a multi-hundred-billion-dollar domestic semiconductor investment program.
The Costs Nobody Fully Priced In
Rebuilding semiconductor manufacturing capacity in the United States and Europe is proving more expensive than the original CHIPS Act projections suggested. Several factors are compounding costs: chronic shortages of construction workers with clean-room experience, the high cost of ultra-pure water and electricity in some chosen locations, and supply chains for fab equipment that remain heavily concentrated in Japan, the Netherlands, and — in some cases — Taiwan itself.
TSMC’s Arizona fabs are estimated to produce chips at a per-unit cost approximately 30–50% higher than equivalent production in Taiwan, according to analysis from SemiAnalysis published in late 2025. The gap is expected to narrow as the workforce matures and supply chains localize, but it will not close entirely. The implication is that geographic diversification in chip production comes with a persistent price premium — one that will ultimately be paid by the consumers of electronics and the enterprises running AI infrastructure.
The European Chips Act, the EU’s parallel €43 billion program, is running into similar dynamics. Intel’s planned Magdeburg fab in Germany has faced permit delays and cost overruns, and the project’s timeline for advanced production remains uncertain as of this writing.
A New Map of Technology Power
What is emerging from the chaos of export controls, subsidy programs, and supply chain rerouting is not a clean decoupling, but something more complex: a partial geographic diversification of production, a splintering of the technology standards landscape, and a permanent increase in the cost and complexity of building the chips that power the modern economy.
Taiwan remains, for now, the irreplaceable center of advanced semiconductor production. TSMC’s market share in leading-edge logic fabrication stands above 90%, and no amount of CHIPS Act funding changes that reality in the near term. What the past four years have done is begin to build the infrastructure — physical, workforce, and regulatory — that might eventually make a different equilibrium possible.
Whether that equilibrium arrives before the geopolitical risks it is designed to hedge against materialize is the trillion-dollar question that neither policy briefings nor earnings calls can fully answer.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.