The debate has outgrown the fab
Public discussion of semiconductors still tends to orbit the same familiar symbols: the most advanced logic chips, the smallest geometries, the latest restrictions on equipment and the prestige attached to domestic fabrication. Those matters remain important. Yet they no longer capture the full shape of the contest. Semiconductor production has become inseparable from a wider set of infrastructural questions that governments are only beginning to confront.
A leading-edge facility is not merely a factory. It is a dense concentration of electricity demand, water treatment, chemical supply, specialist tooling, ultra-clean environments, transport links, highly trained workers and long-lived financing. It also depends on stable regulation and a predictable planning system. In that sense, chipmaking now looks less like a conventional manufacturing sector and more like strategic infrastructure in its own right.
This matters because many governments have announced ambitious semiconductor plans without fully reckoning with the surrounding physical and institutional burden. Industrial policy can help attract investment. It cannot, on its own, conjure grid capacity, speed up transmission build-out, train thousands of technicians or eliminate local bottlenecks. The chip race is therefore becoming an infrastructure contest: slower, more prosaic and more politically demanding than the headlines suggest.
The decisive bottleneck is no longer only the fab itself, but the industrial ecosystem that keeps the fab running every hour of every day.
Demand is broadening faster than policy assumptions
The new urgency is not hard to explain. Semiconductor demand is being pulled in several directions at once. Artificial intelligence has intensified appetite for advanced compute and memory. Carmakers require more chips per vehicle as drivetrains, safety systems and infotainment become increasingly electronic. Defence establishments continue to worry about secure access to trusted supply. Renewable energy systems and grid modernisation rely on power electronics. Consumer devices may be cyclical, but the digitisation of the wider economy is not.
That diversification changes the policy problem. It is no longer enough to think only in terms of the handful of most advanced processors that dominate public attention. Mature-node chips remain essential to industrial equipment, vehicles and medical devices. Analogue and power semiconductors matter to electrification. Advanced packaging has become a strategic chokepoint as performance gains depend more heavily on connecting multiple dies efficiently. In other words, resilience cannot be reduced to a single frontier technology.
Reports from the Semiconductor Industry Association and analyses by the OECD have both pointed to the growing complexity of the supply chain and the concentration of key steps in a small number of geographies. That concentration creates vulnerability not merely to geopolitical shocks, but also to natural disasters, local utility failures and transport disruption. A robust semiconductor strategy must therefore account for the full stack of dependencies, from design software to specialty gases to substations.
Electricity is emerging as the central constraint
If one infrastructural factor is likely to dominate the next decade, it is power. Semiconductor fabrication is electricity-intensive, and advanced compute is becoming similarly demanding once chips leave the factory and enter data centres. This links two previously separate debates: industrial strategy and the economics of electricity systems. A country may succeed in attracting fabrication capacity only to discover that it cannot supply enough reliable power at a competitive price.
This is not a temporary inconvenience. The International Energy Agency has warned that electricity demand from data centres, artificial intelligence and digital infrastructure is set to rise sharply in many markets. At the same time, fabs require extremely high reliability; even brief interruptions can be costly. That raises the value of grid redundancy, local storage, backup systems and speedy interconnection procedures. It also elevates the importance of power quality, not just power volume.
The decisive bottleneck is no longer only the fab itself, but the industrial ecosystem that keeps the fab running every hour of every day.
The implication is straightforward but uncomfortable. Semiconductor ambition now competes with other electricity-intensive priorities, including household electrification, heat pumps, electric vehicles and broader industrial decarbonisation. Governments cannot wish away these trade-offs. They will need to plan generation, networks and demand management with a degree of coordination that liberalised power markets have often struggled to deliver.
In practice, this will favour jurisdictions that can combine abundant generation with efficient grid expansion and credible long-term pricing. It may also favour those willing to treat strategic industrial loads differently from ordinary commercial demand. That is politically sensitive. But pretending that all loads are equal in strategic value is increasingly implausible.
In semiconductors, kilowatt-hours and grid stability are starting to matter as much as tax credits and ribbon-cuttings.
Water, chemicals and land are not secondary issues
Electricity tends to dominate public discussion, but fabs also depend on vast quantities of ultrapure water, tightly controlled chemical inputs and land that can support highly specialised facilities. Taiwan’s repeated struggles with drought have offered a reminder that environmental stress can become industrial risk. Climate change makes such stresses more likely in many regions, whether through water scarcity, flooding or extreme heat.
This has two consequences. First, site selection is becoming more complicated. A location that appears commercially attractive may prove environmentally fragile over a twenty-year horizon. Secondly, resilience increasingly depends on circular systems: water recycling, diversified chemical suppliers, robust storage and contingency planning for transport disruption. These are not glamorous policy areas, but they may determine whether a fabrication ecosystem is genuinely durable.
Land-use politics adds another layer. Semiconductor facilities require not just plots for factories, but surrounding space for utilities, treatment plants, warehousing and supplier co-location. Planning delays, local opposition and fragmented permitting can therefore become hidden strategic liabilities. Countries keen to host advanced manufacturing often discover that national ambition collides with local process. Where institutions cannot reconcile the two, promises outrun delivery.
Skills pipelines matter more than subsidy headlines
Politicians understandably prefer announcing large funding packages to discussing vocational training and immigration rules. Yet semiconductor capacity cannot be scaled by money alone. It requires process engineers, materials scientists, tool technicians, construction specialists, clean-room operators and maintenance staff. These are not interchangeable roles, and many labour markets are already tight.
The issue is especially acute because semiconductor projects arrive in waves. A cluster can suddenly require thousands of construction workers and a smaller but highly specialised permanent workforce. If several projects move at once, labour shortages can drive delays and cost escalation. Universities can help, but higher education alone is too slow and too narrow a response. Apprenticeships, technical colleges and mid-career retraining are at least as important.
Migration policy also deserves a more honest treatment. In practice, many advanced manufacturing ecosystems rely on international talent, especially during ramp-up phases. Restrictive immigration settings can therefore undermine domestic industrial goals. That does not mean abandoning local workforce development; it means recognising that skills strategy must operate on multiple timescales. The countries that treat people as infrastructure, rather than as an afterthought to capital expenditure, will be better placed to sustain semiconductor expansion.
The weakest links are often off the main stage
In semiconductors, kilowatt-hours and grid stability are starting to matter as much as tax credits and ribbon-cuttings.
A notable feature of the semiconductor supply chain is that strategic vulnerability often lies in less celebrated segments. Specialty chemicals, silicon wafers, photoresists, advanced packaging substrates, industrial gases and precision components can all become bottlenecks. So can the equipment maintenance ecosystem that keeps tools operational. These dependencies tend to be less visible than frontier logic chips, but they are no less consequential.
The lesson from recent disruptions is that resilience requires breadth. It is not enough to subsidise one or two flagship facilities if upstream and downstream capabilities remain thin. Nor is it enough to count nominal capacity without considering utilisation, equipment availability or packaging throughput. Policymakers need a more granular map of the ecosystem than many currently possess.
That suggests a shift from symbolic industrial policy towards systems policy. Rather than asking whether a country has a fab, the more useful question is whether it can support a complete and adaptive production network. In many cases, the answer will involve alliances, specialisation and strategic interdependence rather than full autarky. The aim should be resilient integration, not a fantasy of total self-sufficiency.
Resilience in chips will not come from trying to make everything everywhere, but from knowing which dependencies are tolerable and which are dangerous.
Export controls buy time, not capacity
Much recent commentary has focused on export controls and technology restrictions. These tools can indeed slow an adversary’s access to particular capabilities, especially in advanced manufacturing equipment. But they are best understood as measures that buy time rather than substitutes for domestic capacity-building. Time is valuable only if used well.
There is a risk that governments overestimate what denial strategies can achieve on their own. Semiconductor ecosystems evolve. Firms adapt. Research diffuses unevenly but persistently. Workarounds emerge, particularly in mature nodes, packaging and system-level optimisation. Meanwhile, the restricting countries still face their own internal constraints: grid congestion, permitting delays, skills shortages and fiscal trade-offs. A policy centred too heavily on constraint may neglect the harder work of construction.
This is why industrial policy in semiconductors must be judged not by legislative volume but by implementation quality. How quickly are sites connected to power? How many technicians are being trained? Are transport links adequate? Is there enough housing for workers near major facilities? These sound like ordinary administrative questions. In reality, they are now central to technological power.
Industrial policy is entering its awkward middle age
The first phase of renewed semiconductor policy was theatrical: large numbers, strategic language, declarations of sovereignty and grand factory announcements. The second phase is administrative. It involves public procurement, utility regulation, local permitting, environmental review, workforce agreements and patient monitoring of projects that may take years to reach volume production. This phase is less photogenic and far more decisive.
That awkward transition will test political systems. Democracies are often good at announcing initiatives and less adept at maintaining operational discipline across election cycles. Yet semiconductors require exactly that discipline. Fabs are expensive, but the larger challenge is continuity: ensuring that roads, substations, pipelines, wastewater systems and training partnerships are delivered in sequence and on time.
There is also a fiscal question. Subsidy competition can escalate quickly, especially when multiple jurisdictions seek to host similar investments. The danger is not merely waste. It is that governments spend heavily on attraction while underinvesting in the public goods that actually determine long-run competitiveness. A grid upgrade may be less politically visible than a grant, but it often has wider and more durable economic value.
Resilience in chips will not come from trying to make everything everywhere, but from knowing which dependencies are tolerable and which are dangerous.
Compute strategy now depends on what happens after manufacturing
Even if manufacturing capacity expands, the broader compute landscape presents a second infrastructural challenge: deployment. Advanced chips increasingly derive strategic importance from where and how they are used. Data centres, networking infrastructure, cooling systems and power contracts all shape access to compute. In some respects, the politics of semiconductors is merging with the politics of cloud-scale infrastructure, though the underlying issue is more general than any one delivery model.
This creates a feedback loop. Demand for compute justifies more chip investment, but the usefulness of that investment depends on downstream capacity to install and run systems efficiently. Countries that produce chips but cannot deploy them at scale may capture less value than expected. Conversely, countries with strong digital infrastructure but weak manufacturing may still wield influence through control of compute clusters, research ecosystems and application development.
The strategic objective should therefore be coherence across the stack. Semiconductor policy cannot sit in isolation from energy planning, digital infrastructure strategy, research funding and competition policy. The compute economy is an ecosystem, not a sector. The states that recognise this early will make fewer category errors.
What a serious strategy would look like
A more credible semiconductor strategy would begin by ranking constraints realistically. First, secure abundant and reliable electricity, including transmission and local resilience. Secondly, streamline permitting without hollowing out environmental scrutiny; speed matters, but so does legitimacy. Thirdly, build technical workforce pipelines across vocational and university levels, supported by pragmatic migration rules. Fourthly, invest in the less glamorous layers of the supply chain, including packaging, materials and maintenance capabilities. Fifthly, coordinate with allies around specialisation rather than chasing duplication in every segment.
Such a strategy would also use metrics that discourage self-congratulation. Instead of counting announcements, governments should track construction timelines, utility readiness, workforce graduation rates, packaging capacity, water recycling performance and the share of critical inputs with diversified sourcing. These indicators are less dramatic than political speeches, but they are far more telling.
Above all, serious strategy requires admitting that semiconductors are not a standalone trophy industry. They are a foundational input into defence, health, transport, communications and scientific research. Treating them as infrastructure does not mean nationalising the sector or ignoring markets. It means acknowledging that markets alone will not solve coordination problems of this scale and strategic importance.
The winners will be the best organisers
It is tempting to view semiconductor competition as a race for technical brilliance alone. Technical excellence remains indispensable. But the next decade is more likely to reward organisational competence: the ability to align utilities, regulators, educators, financiers, researchers and manufacturers around a common industrial timetable. That is a different kind of national strength, and a rarer one.
The countries most likely to succeed will not necessarily be those with the loudest rhetoric or even the largest subsidies. They will be those that can build substations as readily as laboratories, train technicians as seriously as they court investors, and sustain policy continuity after the launch event has passed. In an era when compute is becoming a general-purpose strategic asset, the humble arts of coordination may prove more decisive than any single technological breakthrough.
That is the deeper lesson of the semiconductor moment. Chips may be tiny, but the systems that support them are vast. The contest is no longer simply about who can etch the smallest features into silicon. It is about who can construct, power and govern the industrial landscape on which modern compute depends. On that measure, the race has barely begun.
