For a decade, the digital sovereignty debate has revolved around the things policymakers can see and regulate: data flows, content rules, cloud jurisdiction, undersea cables. But as writer Eyck Freymann argues in a widely-circulated Rest of World essay published ahead of an expected Xi-Trump summit, the real centre of gravity sits in a cluster of fabrication plants on the western coast of a single island. The China-Taiwan dispute, he writes, has quietly become the most dangerous geopolitical flashpoint for the global digital economy — and Big Tech's deep dependence on Taiwan Semiconductor Manufacturing Company (TSMC) is the reason.
The numbers explain the panic. According to widely cited industry analyses, Taiwan produces roughly 60% of the world's semiconductors and over 90% of the most advanced logic chips — the ones that go into iPhones, Nvidia GPUs, AMD processors, automotive controllers, and the AI training clusters underwriting the current investment supercycle. TSMC alone accounts for the lion's share. There is no near-term substitute. Intel's foundry pivot is years from competitive parity, Samsung's leading-edge yields lag, and China's SMIC is locked out of EUV lithography by US and Dutch export controls.
The chokepoint nobody designed
This concentration was not the product of an industrial strategy gone wrong. It was the natural outcome of three decades of efficient market specialisation: Taiwan got very good at fabrication, and the rest of the industry concluded — rationally — that there was no reason to duplicate the capability. Apple, Nvidia, Qualcomm, MediaTek and AMD all became fabless. Hyperscalers building custom silicon (Google's TPUs, AWS's Trainium, Microsoft's Maia) tape out at TSMC. The supply chain optimised itself into a single point of failure.
That failure mode now sits in the middle of a great-power dispute. A blockade or kinetic crisis in the Taiwan Strait would not just disrupt shipments. As multiple analyses — including the Rhodium Group and Bloomberg Economics — have estimated, it could shave several trillion dollars from global GDP in the first year alone, with knock-on effects across every industry that ships a circuit board. The Freymann piece is right to frame this less as a defence problem than as a digital infrastructure problem.
What the policy response has gotten right
To their credit, governments have moved. The US CHIPS and Science Act of 2022 committed roughly $52 billion in semiconductor incentives, including direct grants funding TSMC's Arizona fabs, Intel's Ohio site, and Samsung's Texas expansion. The EU Chips Act, which entered into force in September 2023, mobilised around €43 billion of public and private investment with a stated goal of doubling Europe's share of global production to 20% by 2030. Japan has poured subsidies into TSMC's Kumamoto fabs. India approved its first commercial fab under a $10 billion incentive scheme.
These are, on balance, the right kind of intervention: industrial policy targeted at a genuine chokepoint, with measurable outputs (wafer starts per month, advanced-node capacity outside Taiwan) rather than vague "sovereignty" rhetoric. The pro-innovation case for them is straightforward — diversifying physical production capacity makes the entire stack above it more resilient, including the open-source software, model weights, and consumer products that depend on chips being available at a predictable price.
What it has gotten wrong
Less defensible is the way "semiconductor sovereignty" has been used to justify a much broader expansion of export controls, outbound investment screening, and de facto decoupling. The October 2022 US export controls and their subsequent BIS updates have steadily widened the perimeter of restricted equipment, chips and design tools. Each tightening makes some marginal contribution to slowing China's frontier AI capabilities — but each also raises the cost of compliance for allied firms, fragments the equipment market that funds the next generation of lithography R&D, and gives Beijing a stronger political case for state-led indigenisation.
The lesson from the past three years is that demand-side controls have a ceiling. SMIC's 7nm Kirin chip in the Huawei Mate 60, Huawei's Ascend AI accelerators, and the rapid build-out of domestic Chinese lithography suggest the controls are slowing — not stopping — the trajectory, while imposing real costs on Nvidia, ASML, Applied Materials and Lam Research. A more proportionate posture would narrow controls to genuinely military-relevant nodes and tools, while accepting that commercial chips will keep flowing both directions.
The sovereignty conversation Big Tech needs to have
For the platforms and AI labs that sit on top of the stack, the TSMC dependency is the sovereignty question that matters most — far more than where a user's profile picture is stored. A serious response involves multi-fab sourcing commitments (already underway at Apple and AMD), second-sourcing critical SKUs at Samsung or Intel Foundry even at a yield penalty, designing for graceful degradation across node generations, and — politically — supporting the diplomatic off-ramps that keep the Strait quiet.
It also means being honest about what cannot be onshored quickly. A leading-edge fab takes four to five years and $20 billion-plus to build; the engineering workforce takes longer. The Arizona and Kumamoto facilities are real progress but, by TSMC's own disclosures, will not match Taiwan's most advanced nodes for years. In the meantime, the most stabilising thing the US, China, and Taiwan can do is keep the commercial relationship boring.
Freymann's framing is uncomfortable for a reason. The digital sovereignty agenda spent a decade arguing about jurisdiction over bytes. The chips those bytes run on were, all along, the actual single point of failure. The right response is more capacity, narrower controls, and a clear-eyed acknowledgment that no amount of data-localisation law will matter if the fabs stop running.