TSMC's $100bn bet and industrial policy's reckoning

Taiwan Semiconductor Manufacturing Company announced this week that it would commit an additional $100 billion to expanding its manufacturing capacity on American soil, a pledge that brings its total US investment to a figure that would have been regarded as fantastical five years ago. The announcement came as China reported its weakest economic growth in years, with Q2 GDP coming in at 4.3% — significantly below Beijing’s official 5% target and well under the pace needed to sustain the investment levels that have made Chinese chip ambitions credible. Taken together, these two data points tell a story about the industrial competition that will define the next decade more than any treaty or tariff: who can manufacture the most advanced semiconductors, at scale, reliably, and in politically safe locations. The answer, for now, is TSMC — and TSMC has chosen America.

The received wisdom

The progressive-technocratic consensus on the CHIPS Act and the associated pressure on TSMC to build in the US holds that this is an unambiguous win for American workers, American national security, and the broader project of reducing strategic dependence on a single choke point in the Taiwan Strait. On this reading, the TSMC investment is industrial policy working exactly as intended: government incentives, security imperatives, and corporate calculation aligning to reshore critical manufacturing that markets alone would never have delivered. The Biden CHIPS Act of 2022, even from a right-of-centre perspective, had a genuine national security rationale that transcended the usual arguments about picking winners. Taiwan’s vulnerability to Chinese military pressure — Al Jazeera and others have noted how Hormuz-style chokepoints are the new vocabulary of great-power leverage — makes chip geography a matter of first-order strategic importance.

A different read

The case for celebrating the TSMC announcement requires more scepticism than the headlines suggest. Begin with the dependency problem. The announcement does not represent an American chip industry; it represents TSMC’s American branch. The distinction matters enormously. American firms — Intel above all — have spent decades losing ground to TSMC because they made the rational short-term decision to outsource manufacturing and focus on design. The result is that the United States now lacks not just the factories but the human capital pipeline, the supply-chain ecosystems, and the institutional knowledge to produce cutting-edge chips without TSMC’s involvement. A $100 billion TSMC plant in Arizona is better than no advanced chip manufacturing in America, but it is not the same as a domestically rooted semiconductor industry with multiple competitive players and deep engineering bench strength.

The historical parallel that comes to mind is the British car industry of the 1970s and 1980s. Having allowed domestic producers to decline through a combination of union militancy, managerial failure, and short-sighted government policy, Britain found itself dependent on foreign investment — Japanese plants in Sunderland and Derbyshire — to maintain any manufacturing presence. Those plants employed British workers and contributed to GDP, but the technology, the intellectual property, and the strategic decisions remained in Tokyo. When political or economic circumstances change, the branch plant is always the first to be reconsidered. American policymakers celebrating TSMC’s announcement would do well to ask: what happens if Taiwanese domestic politics, or a change in the geopolitical temperature, or simply a different CEO’s judgment, leads TSMC to conclude that its Arizona investment is no longer commercially rational?

There is also a fiscal question that the boosterism around industrial policy tends to elide. The CHIPS Act subsidies that underwrote the original TSMC commitment ran to tens of billions of dollars of public money. The Wall Street Journal and others have reported that the total cost of incentivising semiconductor reshoring substantially exceeds any straightforward cost-benefit calculation. At a moment when American debt-service costs are consuming an ever-larger share of federal revenue, the opportunity cost of semiconductor subsidies is real. The honest version of the industrial policy argument is not “this is free money”; it is “this is a costly insurance premium against a catastrophic but low-probability event.” That argument is worth making — but it should be made honestly, not dressed up as an economic growth strategy.

Finally, China’s GDP miss at 4.3% deserves careful interpretation. The slowing of Chinese growth does not mean the Chinese chip challenge has been neutralised. Beijing’s semiconductor ambitions have a political durability that market logic alone does not determine. The Chinese state has absorbed enormous losses in chip investment because it is willing to pay for strategic autonomy in ways that no Western listed company can match. SMIC, China’s flagship foundry, is three to five years behind TSMC in process technology — but three to five years is not forever, and state-directed industries have a poor record of respecting timelines that look comfortable in the present.

What to watch

Watch the pace at which Intel and other US firms use the CHIPS Act framework to rebuild domestic capabilities — this, not TSMC’s Arizona campus, is the long-run test of whether industrial policy has actually worked. Track whether TSMC’s Arizona production costs come in anywhere near competitive with its Taiwan operations, since a sustained cost disadvantage will eventually create political pressure within the company. Monitor China’s chip investment trajectory against its GDP growth: a slowing economy that nonetheless maintains defence-grade semiconductor spending is a state making a strategic bet, not a state in retreat. And note whether the TSMC announcement prompts any reciprocal European investment decisions — the EU’s own CHIPS Act is behind schedule and could become the weak link in the democratic alliance’s technology supply chain.

— J