Semiconductor policy spent much of the past decade arguing over a visible object: the leading-edge chip. That emphasis was understandable. Extreme ultraviolet lithography, transistor density and fabrication geography were measurable symbols of technological rank. By mid-2026, however, the strategic question has shifted. States still want domestic design talent and fabrication footholds, but the practical struggle over advanced compute now turns on a more prosaic issue: who can obtain dependable priority through a chain of tightly linked queues.
The governing scarcity is no longer simply wafers; it is sequenced access to many bottlenecks at once. Fabrication slots, advanced packaging capacity, high-bandwidth memory integration, specialist substrates, power equipment, grid connections, water, data-centre permits and network backhaul all have to align. Miss one queue and the whole stack slips. This matters because artificial intelligence, defence simulation and frontier scientific computing consume not just chips but timed bundles of industrial services. Sovereignty in that environment is less about ownership of a single node than about guaranteed passage across all of them.
From object scarcity to systems scarcity
Classic semiconductor geopolitics was built around a straightforward proposition: leading manufacturing capability confers strategic leverage. That remains true, but it is incomplete. A state can subsidise domestic fabrication and still find that deployable compute arrives late, in smaller volumes than planned, or at materially higher cost because constraints emerge downstream. Advanced packaging has become one of the clearest examples. As chip performance increasingly depends on integrating multiple dies, memory and interconnect technologies in compact packages, the strategic centre of gravity shifts from lithography alone to assembly sophistication and throughput. Industry and research institutions have stressed that packaging is now a key enabling technology rather than a back-end afterthought.
The implication is uncomfortable for industrial policy. Public debate prefers heroic assets: a fab announcement, a new process node, a national champion. Yet the ability to convert semiconductor capacity into sovereign compute often depends on quieter infrastructure with longer planning horizons and thinner labour markets. Packaging engineers, substation transformers, cooling equipment and local planning approvals lack the glamour of lithography, but they increasingly determine whether compute is available when governments and domestic industry need it.
The governing scarcity is no longer simply wafers; it is sequenced access to many bottlenecks at once.
The queue as a strategic instrument
Queues are usually treated as temporary market irritants. In advanced compute they are becoming durable features of power. Capacity in leading fabrication, memory, packaging and data-centre construction cannot be expanded at software speed. It is capital-heavy, permit-bound and skilled-labour intensive. As a result, access is mediated by long booking cycles, prepayments, allocation agreements and public guarantees. The actor with the strongest balance sheet, the deepest diplomatic relationships or the most credible security rationale is often best placed to move to the front.
This creates a subtle but important hierarchy. Formal export controls remain the most visible tool of restriction, especially in advanced computing and semiconductor manufacturing equipment. But even where transactions are legal, practical access may still be rationed by lead times and preferential allocation. A country may be permitted to buy, yet unable to secure delivery windows that match its strategic timetable. In that sense, the queue itself becomes an instrument of statecraft, operating alongside law but not wholly reducible to it.
The governing scarcity is no longer simply wafers; it is sequenced access to many bottlenecks at once.
Why packaging now matters more than rhetoric admits
For years, public discourse over chips implied that value and power resided predominantly in front-end fabrication. That was always a simplification. It is now misleading. Performance gains increasingly derive from combining specialised components rather than relying on a monolithic die alone. High-bandwidth memory, chiplets and advanced interconnects all increase dependence on packaging quality and scale. Research groups and policy analysts have consequently elevated packaging from an operational detail to a strategic capability.
This matters for sovereignty because packaging concentration can nullify upstream investment. A state may support domestic design and even gain some fabrication presence, only to discover that final assembly and integration remain offshore and capacity-constrained. The result is a form of incomplete autonomy. Silicon exists, but usable accelerators do not arrive in the right form factor, with the right memory configuration, on the required schedule. A sovereign fab without sovereign power, packaging and permits is a partial asset, not a full capability.
Data centres are now industrial policy, not just real estate
The second neglected queue sits beyond semiconductors proper. Compute only becomes economically and strategically relevant once installed in facilities with sufficient power, cooling and connectivity. The International Energy Agency has repeatedly underlined the rising electricity significance of data centres and networks. By 2026, this is no niche concern for infrastructure specialists. It is core industrial policy.
Governments that speak fluently about chips but poorly about grid expansion are discussing only half the problem. Large compute clusters require long-lived electricity commitments, not merely short-term market purchases. They also require transformers, switchgear, cooling systems and transmission upgrades, each with their own supply bottlenecks. In several jurisdictions, the limiting factor is no longer investor appetite for data-centre construction but the speed of grid connection and local approval. Compute strategy therefore collides with land-use law, environmental review and public acceptance.
The political economy is awkward. Semiconductor manufacturing offers the tangible narrative of industrial renaissance. Data-centre capacity often invites local opposition over energy use, water and land. Yet a nation serious about sovereign compute cannot avoid these conflicts. If governments will not adjudicate who receives scarce power and permits, the market will do so according to price and incumbent relationships rather than strategic priority.
Europe's lesson: capacity without synchronisation is vulnerability
Europe's semiconductor push has been substantial, with the Chips Act intended to strengthen the region's technological base and resilience. But the European case illustrates the difference between funding assets and synchronising systems. The Union can support research, pilot lines and manufacturing expansion, yet still face fragmentation across energy markets, state-aid regimes, planning processes and telecoms infrastructure. That fragmentation complicates the assembly of a truly sovereign compute stack.
The challenge is not simply that Europe lags in one segment. It is that different member states hold different pieces of the puzzle, while the timelines for electricity infrastructure, skills formation and industrial construction do not naturally align. A cluster may secure semiconductor investment but remain exposed on packaging, memory, power or downstream deployment. The strategic question is therefore less whether Europe can produce more chips than whether it can orchestrate a coherent priority lane from design through installation.
A sovereign fab without sovereign power, packaging and permits is a partial asset, not a full capability.
Export controls have not replaced logistics
Much commentary on compute geopolitics still treats export controls as the master variable. They are plainly important, and recent rules on advanced computing items and semiconductor manufacturing equipment have sharpened the boundary between lawful and restricted access. Yet controls do not abolish the economics of scarcity. They reorder them.
When controls tighten, compliant buyers compete more intensely for whatever capacity remains available within legal channels. That can lengthen lead times and strengthen the bargaining power of firms and states already positioned inside preferred networks. Restrictions also redirect capital toward substitution efforts, which in turn raise demand for mature-node chips, tools and engineering labour that many presumed abundant. In practice, control regimes and queue dynamics reinforce one another. The law says who may participate; allocation decides who receives what, and when.
In compute politics, queues have become instruments of statecraft.
The hidden role of memory, substrates and power equipment
Public semiconductor debates often compress the supply chain into a duel between design and fabrication. The more revealing picture is broader. Advanced accelerators depend heavily on memory technologies whose production is concentrated and capital intensive. They also rely on substrates and packaging materials where capacity additions can take years. Then there is the electrical apparatus that brings data centres online: transformers, switchgear and backup systems whose own supply chains have come under pressure from wider electrification trends.
This is why aggregate claims about new chip investments can mislead. A headline sum tells little about the availability of complementary bottlenecks. If memory integration lags, if substrate supply tightens, or if substations cannot be energised on schedule, a nominal compute build-out underdelivers. Policymakers seeking sovereignty therefore need a discipline that semiconductor strategy has often resisted: accepting that the decisive constraint may sit in an adjacent sector with lower prestige.
National accounting is bad at measuring usable compute
Another problem is conceptual. States count fabs, patents, investment pledges and occasionally chip output. They are much worse at measuring assured compute availability. Yet for strategic purposes, usable compute is what matters: the volume of high-performance processing domestic actors can reliably access within defined security, cost and latency conditions.
This suggests the need for a different analytical frame. Rather than asking only how many advanced chips a country can make, one should ask how much priority compute it can guarantee to research laboratories, defence agencies, health systems and domestic industry during periods of stress. That metric would include semiconductor throughput, packaging access, electricity reliability, cloud and data-centre capacity, and the legal ability to retain domestic preference when markets tighten. It would treat compute as a compound national capability, not a commodity purchased one transaction at a time.
In compute politics, queues have become instruments of statecraft.
The labour constraint is sharper than many strategies admit
Behind every queue is a labour market. Advanced packaging, semiconductor process engineering, power systems, thermal management and data-centre operations all require specialised personnel. Training pipelines are slower than capital announcements. Even wealthy states cannot conjure experienced engineers, technicians and construction managers overnight.
This matters because labour shortages create second-order dependencies. Countries may attract investment yet rely on imported expertise to build, qualify and operate facilities. That may be efficient in ordinary times, but it weakens claims of sovereignty if visas, security restrictions or geopolitical tensions disrupt staffing. The politics of compute is therefore also a politics of vocational systems, university-industry collaboration and migration policy. A strategy focused only on capital expenditure risks purchasing shells without the people to run them.
What strategic autonomy now really requires
If compute sovereignty is becoming an allocation problem, strategic autonomy requires a correspondingly administrative response. The essential task is to build credible priority mechanisms across the full chain. That does not mean autarky. No major economy can internalise every semiconductor and infrastructure segment efficiently. It means identifying which chokepoints must be secured by domestic capacity, which can be covered by trusted external arrangements, and which require standing rights to priority access under stress.
In practice, that pushes governments toward a less romantic form of state capacity. They must coordinate energy regulators with industrial ministries, planning authorities with national security officials, and research funding with grid investment. They must decide whether the marginal public euro or pound is better spent on another symbolic manufacturing subsidy or on the unglamorous enablers that actually shorten queues. For many jurisdictions, the latter will yield more sovereign compute than the former.
The end of the fab-centric myth
The strategic map of semiconductors has not ceased to matter. It has become more intricate. Leading chips remain indispensable, and the geography of fabrication still shapes alliances and vulnerabilities. But the old fab-centric myth no longer captures where leverage lies. The decisive question in 2026 is not merely who can make advanced silicon. It is who can turn design, manufacturing, packaging, electricity and installation into reserved compute on a dependable timetable.
That is a less theatrical story than the race for the smallest node. It is also more faithful to the way industrial systems confer power. Nations that understand this will treat queues, permits and power contracts as strategic assets. Those that do not may discover that they own pieces of the semiconductor future without commanding the compute it was supposed to deliver.
A sovereign compute strategy worthy of the name must therefore begin with an unflattering admission. The chip is only the most visible component in a larger choreography of constraints. Real autonomy lies in synchronising the whole sequence. Anything less is procurement dressed up as strategy.
