SWITCHED ON
The daily technology series nobody asked for but everyone needed
The Chip That Runs the World: Semiconductor Geopolitics
One company on one island produces the majority of the world's most advanced chips. If that changes — for any reason — the consequences for the global economy and military balance of power are difficult to overstate.
Semiconductors are to the twenty-first century what oil was to the twentieth: the resource whose control determines the trajectory of economies, militaries, and geopolitical power. The difference is that oil is distributed across many countries and continents. The most advanced chip manufacturing in the world is concentrated in an area roughly the size of Maryland, in a country whose sovereignty is disputed by the world's second-largest economy.
Yesterday we opened Season Two with nuclear energy's unlikely comeback — the three forces driving its rehabilitation, the SMR promise and the NuScale reality check, the waste problem treated honestly in both directions, and where the global fleet is actually heading. Today we are staying with the theme of critical infrastructure but moving from atoms to transistors. Semiconductors: the chips that run every device, every data centre, every weapon system, and every AI model that exists. The supply chain that produces them is the most concentrated and geopolitically fragile in the history of modern industry, and the race to change that is one of the defining economic and strategic competitions of the decade. If this sounds like it matters, it is because it does. More than almost anything else we will cover this season.
01 — What a Semiconductor Actually Is and Why Making One Is Impossible
A semiconductor is a material — typically silicon — whose electrical conductivity can be precisely controlled, allowing it to function as a switch. Billions of these switches, called transistors, etched onto a chip the size of a fingernail, switching on and off billions of times per second, are how every computer, phone, data centre, and AI system processes information. The more transistors you can fit on a chip, and the smaller and more efficient you can make each one, the more powerful and energy-efficient your computing becomes. This is Moore's Law — the observation that transistor density roughly doubles every two years — and it has driven the past six decades of computing progress.
Producing the most advanced chips currently requires a manufacturing process of staggering precision. The smallest features on leading-edge chips are measured in angstroms — fractions of a nanometre, smaller than some molecules. Achieving this requires machines that use extreme ultraviolet light — with a wavelength of 13.5 nanometres, shorter than soft X-rays — to etch patterns onto silicon wafers with atomic-scale accuracy. These machines, called EUV lithography systems, are produced by a single company: ASML, headquartered in Eindhoven in the Netherlands. Each machine costs approximately $200 million, takes years to build, requires the cooperation of roughly 5,000 suppliers across multiple countries, and contains components that collectively represent the most complex manufactured object ever made. There is no substitute. There is no second source. If ASML stops producing EUV machines, leading-edge chip manufacturing stops globally.
ASML is perhaps the most important company most people have never heard of. Without its machines, there are no advanced chips. Without advanced chips, there is no modern AI, no advanced military electronics, no competitive smartphone industry, and no meaningful progress in computing for the foreseeable future.
The machines ASML produces go primarily to three companies capable of using them to manufacture at the leading edge: TSMC in Taiwan, Samsung in South Korea, and Intel in the United States — though Intel's manufacturing capabilities have lagged its ambitions significantly in recent years. Of these, TSMC is by a substantial margin the most important. It manufactures chips for Apple, NVIDIA, AMD, Qualcomm, and effectively every other fabless semiconductor company that designs chips without owning manufacturing facilities. Its market share in the most advanced process nodes exceeds ninety percent.
02 — The Taiwan Problem
Taiwan's status is one of the most consequential unresolved questions in contemporary geopolitics. The People's Republic of China regards Taiwan as a province that will eventually be reunified with the mainland, by force if necessary. Taiwan's government regards itself as the legitimate government of China, or increasingly simply as a distinct democratic entity with no desire to be governed from Beijing. The United States maintains a policy of strategic ambiguity — selling Taiwan defensive weapons and maintaining unofficial but substantial ties, while not formally recognising Taiwanese independence or explicitly committing to its military defence.
TSMC sits in the middle of this. Its fabs — the manufacturing facilities where chips are made — are located in Taiwan. A Chinese military action against Taiwan, whether invasion, blockade, or targeted strikes, would immediately threaten or eliminate the world's supply of leading-edge chips. The consequences would cascade through every sector of the global economy simultaneously. Every major car manufacturer would halt production within weeks. Every consumer electronics company would face immediate supply crises. Every AI data centre expansion would stop. The US military, which depends on commercial semiconductor supply chains for advanced weapons systems, would face equipment shortages. The economic damage from even a short disruption to TSMC's production would run into the trillions of dollars. This is not a hypothetical risk that analysts have constructed. It is the stated concern of governments, militaries, and corporations across the developed world, and it is the primary driver of the semiconductor policy revolution currently underway.
03 — The Chips Acts and the Race to Reshore
The US CHIPS and Science Act, signed into law in August 2022, allocated approximately $52 billion in subsidies and tax credits for domestic semiconductor manufacturing, research, and workforce development. It was the largest industrial policy intervention in US history in the technology sector, and it represented a decisive break from the decades-long consensus that governments should not pick winners in technology markets. The argument for picking this particular winner was the strategic concentration problem — the calculation that allowing the world's most critical technology manufacturing to remain concentrated in a geopolitical flashpoint was a risk that market forces alone would not adequately address.
TSMC is building fabs in Arizona — two initially, with a third announced — with significant US government subsidy. Intel is building new fabs in Ohio and Arizona. Samsung is building in Texas. The European Chips Act is funding expansion in Germany and elsewhere. Japan has attracted a TSMC fab to Kumamoto with substantial government support. The global reshoring effort is real, substantial, and proceeding.
It is also slower and more expensive than the announcements suggested. The Arizona TSMC fabs have faced repeated delays, cost overruns, and a highly publicised shortage of the specialised workforce required to operate leading-edge fabs — a workforce that took Taiwan decades to develop and cannot be conjured quickly in Phoenix. The initial target of producing chips at the leading process node by 2024 slipped to 2025, then to 2026. The chips being produced in the US will cost more than equivalent chips from Taiwan for the foreseeable future, because the learning curves, ecosystem density, and workforce depth that make TSMC's Taiwan operations so efficient take decades to build, not years.
04 — China's Semiconductor Ambitions and the Export Control War
China produces a large volume of mature-node chips — less advanced but still essential for a wide range of applications including automotive, industrial, and consumer electronics. What China cannot currently produce is leading-edge chips. Its most advanced domestic manufacturer, SMIC, is operating at process nodes several generations behind TSMC, constrained by the inability to acquire EUV lithography machines following US-led export controls that have prevented ASML from shipping its most advanced equipment to China.
The export control escalation has been one of the most significant technology policy moves in recent years. The Biden administration's October 2022 export controls — substantially tightened in subsequent rounds — restricted not just EUV machines but a wide range of advanced semiconductor manufacturing equipment, chip design software, and the chips themselves from reaching China. The controls apply to any company using US technology anywhere in its supply chain, which given the ubiquity of US technology in semiconductor manufacturing effectively means most of the global supply chain. Countries that were previously selling to China have faced US pressure to align with the restrictions.
China's response has been to accelerate domestic investment in semiconductor capability to a degree that is historically unprecedented. Hundreds of billions of renminbi in state funding, thousands of semiconductor engineers recruited from overseas, and a policy priority that places semiconductor self-sufficiency alongside AI development as the central technology objectives of the state. The results are real but partial — China has made genuine progress in mature nodes and in specific areas of the supply chain, but the gap to the leading edge remains substantial and the export controls are designed to prevent it from closing. Whether they succeed depends on whether China can develop indigenous EUV capability — a challenge that requires mastering optics, light sources, precision mechanics, and materials science at the absolute frontier of human capability, without access to the global supply chain that ASML took decades to develop.
05 — NVIDIA and the AI Chip Moment
No discussion of semiconductor geopolitics in 2026 is complete without NVIDIA, which has become arguably the most strategically important company in the world by virtue of producing the graphics processing units that turned out to be the essential infrastructure of the AI revolution. GPUs, originally designed for rendering video game graphics, proved extraordinarily well-suited to the parallel matrix calculations that train and run neural networks. NVIDIA's CUDA software platform, developed over fifteen years as a programming environment for its GPUs, created a switching cost and developer ecosystem that competitors have been unable to replicate despite years of trying.
The result is that NVIDIA holds a dominant position in AI training hardware that gives it extraordinary leverage over the trajectory of AI development globally. Every frontier AI lab — OpenAI, Anthropic, Google DeepMind, Meta AI — runs its most advanced training on NVIDIA hardware. The export controls on advanced AI chips have become one of the primary instruments of US technology policy toward China, with successive rounds restricting increasingly capable NVIDIA chips from Chinese buyers. NVIDIA has responded by developing restricted versions of its chips for the Chinese market, which are then restricted further, in an ongoing regulatory and commercial dance that neither side has fully resolved.
The concentration of AI compute capability in NVIDIA's product line, manufactured by TSMC in Taiwan, is the most acute single-point-of-failure in the global technology ecosystem. It is not lost on anyone paying attention that the hardware running the AI systems that may define the next century of human development is made by one company's designs, on another company's machines, using a third company's equipment, in a geopolitical flashpoint. The fragility is extraordinary. The dependence is total. And the race to change it — through domestic manufacturing, alternative chip architectures, and supply chain diversification — is one of the defining industrial competitions of the next decade.
Tomorrow we are zooming from the hardware layer to the human layer — the future of work in an AI economy. What jobs are actually at risk, what the historical evidence on technology and employment tells us, why this time might genuinely be different, and what the policy responses look like from the countries that are thinking hardest about it. See you then.
Switched On is a daily technology series covering the ideas, systems, and arguments shaping the digital world. Opinionated. Witty. Occasionally wrong. Always worth the argument.
