Technological Sovereignty

Technological sovereignty is the claim that a state can hold a technology — not merely buy its output or regulate its use, but retain the capability to make it, on its own territory, through a period in which somebody outside refuses to help. Three instruments pursue it and are usually discussed as one: subsidy, restriction (export controls, investment screening, research-security rules), and alignment (friend-shoring, allied control lists, standards positions).

Established Each leg is owned elsewhere on this map and the conjunction is owned nowhere. Economic Resilience owns the shock behaviour of economic systems and the finding that governs the subsidy leg: reshoring reduces resilience. Future Trade Systems owns the rules layer and the measurement of what the post-2018 reallocation moved. Innovation Ecosystems owns the causal evidence on whether public money produces innovation. This brief owns what happens when a state runs all three at once and calls the result sovereignty.

Established The finding this brief lands is about the unit of account. Every major sovereignty programme states its objective in units that cannot fail: euros mobilised, share of global production value, fabs announced. None states it in the units of the thing it claims to buy — a named capability sustained for a named number of months after a named supplier stops shipping. Because the objective is never written that way, the programmes cannot be scored and the insurance cannot be priced.

Frontier The standard economic critique misses the point these programmes are making, which is why it has not slowed them. The critique says autarky is expensive, and the measured costs below are large. But a sovereignty programme buys an option that pays out in a state of the world — blockade, embargo, war over a producing island — where relative prices are not the operative variable, and an option can be worth an absurd-looking price. What cannot be defended is buying it without writing down the strike, the payout or the event.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established The money is real and known to within the usual authorisation-versus-disbursement gap. The United States CHIPS and Science Act of 2022 authorised about $52.7bn — roughly $39bn in manufacturing incentives and $13.2bn in research and workforce programmes — alongside a 25% investment tax credit uncapped by appropriation and larger in present value. The European Chips Act, Regulation (EU) 2023/1781, describes itself as mobilising about €43bn to 2030. Japan has committed trillions of yen, India near $10bn, Korea a multi-hundred-billion-dollar cluster. A programme is not its authorisation: schedules have been renegotiated repeatedly and at least one United States award became a government equity position rather than a grant.

Established Exactly one large programme states a target that can fail, and it is on course to fail it. The European Chips Act sets 20% of global semiconductor production value by 2030 against a starting point usually given as 8–10%. The European Court of Auditors, in its 2025 special report on the microchips strategy, judged the target unlikely to be met and projected a share in the low teens. That is a public-audit body scoring a programme against the programme’s own published number. Nothing equivalent exists for the United States, Japan, Korea or India, because none published a number a measurement could contradict.

Established The best cost estimate for full self-sufficiency comes from the industry lobbying for the subsidies. The Semiconductor Industry Association and Boston Consulting Group estimated that a fully self-sufficient regional model would need upward of $1tn in incremental upfront investment and raise chip prices roughly 35–65%, and that the ten-year cost of ownership of a leading-edge United States fab runs about 25–50% above the same fab in Asia. (industry association; the gap is exactly what the subsidy was sized to close) The 25–50% figure is the standing per-unit price of domestic production, and it does not end when the construction subsidy is spent.

Established What the largest sovereignty-adjacent intervention achieved is a change of intermediary, not of dependence. China’s share of United States imports fell from 21.6% in 2017 to 16.5% in 2022 while the dollar value rose from $505.1bn to $531.3bn. Over the same period Vietnam’s imports from China rose from about 9% to about 40% and Mexico’s from about 1% to about 20%. The measured price was about 9.8% higher import unit values from Vietnam and 3.2% from Mexico. A World Bank analysis finds the largest gainers were the economies most integrated with China, and that for over 70% of affected products one supplier took at least three-quarters of the lost share. The tier-one import share moved; the exposure did not.

Established Export controls impose a measured cost on the firms of the country imposing them. A Federal Reserve Bank of New York staff study of firms hit by United States export controls reports cumulative abnormal returns near −2.5% within twenty days, an average market-capitalisation loss near $857m per firm and about $130bn in aggregate, with revenue down about 8.6%. Frontier The corresponding quantity — how far a control set back the target, in months or in nodes — has never been estimated with a credible identification strategy in public, so the debate substitutes announcements for measurement in both directions.

Established Export restriction has stopped being an exception and become an ordinary instrument. The International Energy Agency reports more than half of a broad group of energy-related minerals now under some form of export control. On the other side of the ledger, China introduced licensing regimes covering gallium and germanium in 2023, graphite the same year, antimony in 2024 and a broad rare-earth regime from 2025. Frontier Those measures have been announced, partly suspended and renegotiated more than once; the durable fact is the licensing architecture, not any week’s coverage list.

Frontier The evidence on whether subsidy buys capability is better than it was and still does not answer this question. A well-instrumented study of a United Kingdom regional investment programme finds positive effects on employment, investment and net entry and none on total factor productivity. The long-run study of the United States Small Business Innovation Research programme finds awardees grew faster than matched firms over ten years, with the effect confined to regions that already had venture activity. Public money reliably buys plants, jobs and entry where an ecosystem exists, and has not been shown to buy productivity or to create an ecosystem where none was — which is what most sovereignty programmes attempt.

Established The fragmentation these programmes respond to is, in the flows data, much smaller than the rhetoric. Excluding Russia and the United States–China pair, East-to-West trade outperformed world trade by about 6% and West-to-East by about 10%, and the average distance travelled by a dollar of manufactured trade rose. The best-identified bloc effect is near 12% against about 67% for the Cold War. The consequence is uncomfortable for both camps: the world is not fragmenting at a rate that justifies emergency industrial policy, and it is fragmenting in exactly the two places these programmes target.

Frontier One sovereignty objective in the record states a falsifiable number, and it is not a technology objective. China has operated an explicit grain self-sufficiency target near 95% and held roughly half of global grain reserves in 2022. (think tank with a stated strategic interest; used for quantities, not framing) Food sovereignty gets a ratio and a stockpile because grain is storable. Technological sovereignty gets neither, because the thing at risk is a running production system with a tool fleet, a supplier network and an operating workforce. That asymmetry makes the unit-of-account problem structural rather than a drafting failure.

Frontier Did the semiconductor export controls work? The field has one contested datum and two readings of it. In September 2023 a Chinese handset shipped with a domestically fabricated processor at a 7-nanometre-class node, made without extreme-ultraviolet lithography. One reading: the control failed, the capability arrived anyway. The other: the control worked, because the part came from multi-patterning on deep-ultraviolet tools at a yield and cost never disclosed. Both are consistent with the public evidence, and the variable separating them — yield at volume — is the one number nobody outside the fab has.

Frontier Does subsidy create capability or relocate it? The flagship plants of this round are built by incumbent foreign operators using imported recipes, tool sets and, early on, engineers. That is technology transplantation, a defensible objective: a transplanted fab keeps producing if its parent’s territory becomes unreachable. It is not sovereignty over the technology, because the recipe, the service contracts and the next node stay outside. The distinction has an observable test nobody runs: what fraction of the plant’s critical inputs, by value and by substitutability, is single-sourced outside the jurisdiction after ramp?

Frontier Are sovereignty and resilience the same objective? The evidence says no and the programmes assume yes. Economic Resilience shows the dominant instrument raises volatility. If sovereignty is an insurance option against refusal — an event outside the distribution those studies sample — higher ordinary volatility is a premium, not a contradiction. That reconciliation is available and no government has made it in public, because making it requires naming the event being hedged, which is diplomatically costly. A defensible argument is being advanced in an indefensible form.

Established The binding physical constraint is a single toolset with a single supplier and no second source anywhere. Extreme-ultraviolet lithography comes from one company, using a light source and an optics lineage each held by one further company. There is no partial substitute at the leading edge: a jurisdiction without those tools is not slowed, it is stopped at a node. No state programme has funded a second source, because the development cost and learning curve are multi-decadal with no second customer base to amortise against.

Established The second constraint is advanced packaging, which the first round of programmes under-funded relative to its concentration. Leading-edge logic increasingly ships as a package rather than a die, and interposer and stacking capacity for high-performance parts sits in a small number of facilities in one jurisdiction. A country can hold wafer-start capacity and still be unable to deliver a finished accelerator.

Established Specialty chemicals supply the best natural experiment in this subject and everybody forgets it. In 2019 one state restricted exports of three photoresist and etchant precursors to another, both allies of the same great power. The restricted party did not lose its industry; it substituted, requalified and partly localised over several years at a documented cost in schedule and yield. That is the closest thing on record to a controlled trial of a targeted input cut-off against a sophisticated economy, and it has never been written up as a sovereignty evaluation with full cost accounting.

Established Outside semiconductors the chokepoint is separation and refining, not mining. Ore deposits are geographically diverse; heavy-rare-earth separation and sintered-magnet manufacture are not, and the concentration ratios are the highest in any industrial supply chain in use. Building separation capacity is a permitting and waste-handling problem measured in years and litigation, not a capital problem.

Established This brief waits on one scientific result and two accounting decisions, and only the first is a research problem. The scientific gap is a credible causal estimate of what an export control does to the target’s capability. Everything needed exists in principle — dated announcements, product-level customs data, publication series, capacity trackers — and the difficulty is finding an untreated comparison group. Until it is done, the restriction leg runs on anecdote.

Established The first accounting decision is value-added trade accounts at annual frequency, and it belongs to another brief. Future Trade Systems establishes that nearly two-thirds of global trade happens within value chains, that every reallocation figure quoted anywhere rests on gross flows that double-count intermediates, and that no jurisdiction publishes value-added accounts fast enough to score a policy. This brief inherits that gap rather than claiming it.

Frontier The second is a capability inventory in physical units. “How many months of output at what specification can this jurisdiction sustain if supplier X stops shipping?” is answerable from data that fab operators, tool vendors and defence ministries already hold. It is published nowhere, for any technology. The absence is a choice, and it is the choice that keeps the subject unscoreable.

Established The decisive test in this subject is already scheduled and nobody scheduled it as a test: the European Chips Act’s 20% of global production value by 2030. It is the only technological-sovereignty target anywhere that is numeric, dated, externally measurable and already scored by an independent audit body that has projected a miss. What the 2030 measurement settles is not whether Europe makes enough chips; it settles whether a published sovereignty target functions as an instrument or as decoration. If the target is missed and nothing follows — no re-scoping, no reallocation, no revised objective — the unit-of-account problem is confirmed. If the miss forces public re-scoping, numeric targets bind and every other jurisdiction should publish one.

Established The second experiment costs nothing and is a memorandum: the cut-off drill. Take one named technology and one named foreign supplier and publish, in advance, what output can be sustained, at what specification and for how many months, if that supplier stops shipping. The data sit with firms and ministries today, and publishing the answer converts sovereignty from a slogan into a quantity that can be wrong. No government has published such a statement for any technology.

Frontier The third is a regression discontinuity the subsidy programmes have already generated and are sitting on. Both the United States and European semiconductor programmes scored applicants and funded above a threshold. The near-miss applicants are the comparison group economists spend careers constructing, and assignment and outcomes both sit in the agencies’ records. Innovation Ecosystems documents that this design has produced the strongest causal results in innovation policy. Running it requires releasing scores, not inventing a method.

Frontier The fourth is a retrospective that should already exist: a full cost accounting of the 2019 photoresist-precursor restriction between two allied economies. It is the only targeted input cut-off against a sophisticated economy with a clear start date and a multi-year window, and it would give the first empirical estimate of the substitution elasticity every sovereignty programme assumes.

Speculative The experiment nobody can run is the one that matters most. The state of the world these programmes hedge — sustained refusal of supply across a bloc — cannot be simulated and has no modern precedent at the relevant technological depth. Everything above is a proxy, and the central claim of sovereignty policy is not testable before the event it is designed for. That is an argument for pricing it explicitly, not for exempting it.

Established A leading-edge fab is a schedule problem before it is a money problem. Site and permitting, shell, cleanroom fit-out, tool install, process qualification, yield learning, volume. The last two cannot be accelerated with capital: yield learning is information accumulation whose rate is set by wafer starts. Programmes budget the first four stages and quote the end of stage four as the arrival of capability.

Established The tool fleet is the sovereignty question in physical form. A modern fab runs hundreds of distinct tool types from dozens of vendors, many single-source for the specific process step, nearly all requiring vendor field service, spare parts and software under contracts licensable by the vendor’s home government. A jurisdiction owning the building and the workforce but not the service contracts has bought a hostage, not a capability, and the difference becomes visible only when someone withholds service.

Frontier Second-sourcing is an engineering discipline that industrial policy treats as a procurement preference. Qualifying a second supplier means re-running process qualification against a different impurity profile, particle distribution and failure-mode set, at a cost in wafers and calendar time often comparable to a node transition. A mandate to do so has a price that could be estimated and has not been.

Frontier The requirement that would change the picture is a designed-for-substitution product architecture. Specify parts against interface envelopes rather than named vendor devices and the cost of an input cut-off falls sharply with no reshoring at all. It is standard in a few safety-critical sectors and rare elsewhere because it costs performance. Nobody has published the trade curve, and it is the cheapest sovereignty instrument on this list.

Established The three neighbouring briefs each own one leg of this argument. Economic Resilience owns shock behaviour, the reshoring-raises-volatility result, and the finding that no jurisdiction has published an ex-ante price for the resilience it claims to buy. Future Trade Systems owns the flows-versus-rules distinction, the collapse of binding dispute settlement, tariff incidence and the value-added accounting gap. Innovation Ecosystems owns the causal evidence on subsidy, clusters and place-based policy.

Frontier The cooperative alternative is a live design question elsewhere on this map. Global Cooperation Models owns the machinery by which states could hold a capability jointly — shared stocks, mutual-supply guarantees, allied licensing. The engineering case is strong because fixed costs dominate; the institutional case is weak because a guarantee is worth what the guarantor’s interest is on the day.

Established Restriction is exercised by licensing authorities that are national, and the technology is not. Leading-edge controls require at least three jurisdictions to act, because the toolchain is split across them, and each acts through its own statute, list, officers and industrial exposure. The observed result is an announcement in one capital, a slower implementation in another, and a window between in which orders are pulled forward. There is no allied licensing authority and no serious proposal for one, because creating one means conceding a veto over a state’s own firms.

Frontier Research security is the leg where the cost is most clearly borne by individuals and least clearly measured. Disclosure requirements, foreign-gift reporting and nationality-sensitive access rules have tightened across the same jurisdictions running the subsidy programmes, and survey work reports substantial chilling effects among researchers of the targeted national origin. The counterfactual — how much sensitive transfer was prevented — is published in no checkable form, and the one named enforcement programme of the last decade was wound up in early 2022 after criticism of its hit rate.

Frontier The rules layer that used to arbitrate industrial subsidy no longer functions. Binding multilateral dispute settlement has been non-operational since 2020, a majority of members lack an appellate stage, and the interim substitute has issued a handful of awards in six years; Future Trade Systems owns the record. The consequence is rarely stated: the subsidy race runs with no functioning referee, so discipline on it is fiscal and political rather than legal, and it stops when budgets stop rather than when a panel rules.

Established The measured incidence of fragmentation falls hardest on countries that are not parties to the dispute. Estimates put low-income countries’ losses at up to four times the global average, and the reallocation record shows gains concentrating in a few middle-income connector economies rather than spreading. A sovereignty programme is a rich state buying an option whose premium is partly paid by states with no vote and no equivalent programme available to them.

Frontier Research-security rules allocate suspicion by nationality, and that is a cost with a distribution. Whatever the security benefit, the burden falls on identifiable people by origin rather than conduct, and the justification is classified while the burden is not. The minimum defensible standard is that a regime imposing a nationality-correlated burden should publish a periodic, externally audited estimate of what it prevented, and no such publication exists.

Speculative There is a defensible egalitarian case for technological sovereignty and it is almost never made. A state that cannot make anything critical has its population’s welfare hostage to decisions taken elsewhere, and the people most exposed are the least mobile. The case fails as usually advanced because it is made in the language of national prestige rather than exposure, and because the programmes it justifies send money to capital-intensive plants employing comparatively few people.

Frontier The largest civilizational cost of a sovereignty regime is duplicated fixed cost in technologies whose fixed costs are the whole cost. Leading-edge semiconductor development has negligible marginal cost, as increasingly do frontier model training, large-scale biological instrumentation and fusion engineering. Duplicating those fixed costs across four blocs produces roughly the same capability at four times the cost, minus the spillovers from a single global learning curve. Nobody has estimated the lost learning curve, and it is plausibly the largest number in this brief.

Speculative The opposing consideration is that one global learning curve is a single point of failure for a civilization. If one region holds the only capability to produce the substrate of computation, an earthquake, a blockade or a bad decade of governance there is a global capability loss. On that reading duplication is redundancy, and the right comparison is cost against expected loss from a correlated failure — never made quantitatively, because the failure probability is unknown to within orders of magnitude.

Handwave The strongest version of the sovereignty case rests on a step that is pure assertion. It runs: the world is entering an era in which technological capability determines political autonomy, therefore any cost is justified. The first clause is a generalisation with no measurement behind it, the therefore does not follow even if it is true, and the conclusion is unbounded. When a policy argument reaches “any cost”, it has stopped being an argument about the world.

These horizons track when the evidence about technological sovereignty changes, not when a plant opens. Every entry is a date on which someone could be shown to have been wrong.

• 10 yr: Frontier The 2030 European target is measured and either forces a public re-scoping or does not; the first cohort of subsidised advanced fabs completes ramp and its per-unit cost gap becomes observable in customer pricing; and the first credible causal estimate of an export control’s effect on a target is either published or still missing, which after a decade of the policy would itself be a finding.
• 25 yr: Speculative Either second sources emerge for at least one current single-supplier chokepoint — most plausibly advanced packaging and rare-earth separation, least plausibly extreme-ultraviolet lithography — or the structure survives three decades of funded attack, which would be strong evidence that concentration at that depth is an equilibrium rather than an accident. The subsidy race ends on fiscal grounds well before this horizon.
• 50 yr: Speculative Whether blocs converge on incompatible technical stacks is decided in this window, and the mechanism would be standards rather than trade: divergent specifications, certification and tooling compound until interoperating costs more than not. Nothing in the current data forces this, and the counter-evidence is that flows never fragmented as much as the rhetoric.
• 100 / 250+ yr: Handwave Claims at this range are claims about whether states remain the relevant holders of industrial capability at all. The candidates — firms larger than most states, supranational consortia, distributed production making the question moot — are each coherent and none supported. Any number attached to this row is decoration.

• Depends on Nothing on this map produces a result this brief waits on. What the subject lacks is a causal estimate, a published inventory and a released set of application scores, none of which is a discovery. It inherits three findings rather than deriving them: reshoring-raises-volatility and unpriced resilience from Economic Resilience, the flows-versus-rules asymmetry from Future Trade Systems, and subsidy-works-where-an-ecosystem-exists from Innovation Ecosystems. Inheritance is not dependency and no typed depends-on edge is claimed.
• Requires (not on this map) Five constraints, in the order of the tokens. First, a published cut-off test: one jurisdiction stating in advance how many months of output at what specification it can sustain if a named foreign supplier stops shipping. The data sit with firms and ministries now; the missing item is a decision to publish, and until it exists the subject has no unit of account. Second, a causal estimate of what an export control does to the target rather than the imposer. The imposer’s cost is measured — about −2.5% cumulative abnormal returns and about $130bn aggregate — and the target’s side is empty, which is why “the controls worked” and “the controls failed” remain equally sayable about the same 7-nanometre-class part. Third, a second source for extreme-ultraviolet-class lithography: one company makes the tool, one its light source, one its optics, and no programme has funded an alternative because the development cost is multi-decadal with no second customer base. Fourth, domestic advanced packaging matched to domestic wafer starts: leading-edge logic ships as a package, stacking capacity is concentrated in one jurisdiction, and a country with wafer starts but no packaging cannot deliver a finished part. Fifth, market-shaped rather than industrial: a buyer that pays the sovereignty premium without a mandate. Domestic production runs about 25–50% above Asian cost of ownership on the industry’s own estimate, and self-sufficiency is put at $1tn-plus upfront with prices 35–65% higher; if no customer pays that voluntarily, the capacity is a permanent subsidy line rather than an industry.
• Enables Every programme on this map assuming an advanced semiconductor, a permanent magnet or a specialty chemical arrives from somewhere inherits this brief’s uncertainty. No typed enabling edge is claimed, because what would enable a downstream claim is a published capability inventory rather than a result this brief could supply, and because the tier-one statistics such programmes would use do not measure the exposure they are taken to measure.
• Adjacent Within this map: Economic Resilience, Future Trade Systems, Innovation Ecosystems, Autonomous Supply Chains, Infrastructure Resilience and Global Cooperation Models. Outside it: gravity-model trade econometrics, input-output economics, the economics of defence procurement, semiconductor process engineering and the history of technology diffusion.

Established “Reshoring makes a supply chain safer.” The best available evidence says the opposite. A review of the supply-chain literature finds chains attenuated past shocks, that foreign-supplier diversification lowers volatility and that reshoring raises it; an independent general-equilibrium model finds localisation cutting global real GDP by more than 5% while producing larger price swings under shock. The claim survives only if what is bought is availability under deliberate refusal — a different objective no programme states.

Established “Dependence on China has been substantially reduced.” The tier-one number moved and the exposure did not. China’s United States import share fell 21.6% to 16.5% between 2017 and 2022 while the dollar value rose; Vietnam’s imports from China went from about 9% to about 40% and Mexico’s from about 1% to about 20%; the biggest gainers were the economies most integrated with China; and for over 70% of affected products a single supplier absorbed at least three-quarters of the lost share.

Established “Export controls are costless to the country imposing them.” The best-identified estimate is a cost at home. Affected firms show about −2.5% cumulative abnormal returns within twenty days, roughly $857m of market-capitalisation loss each and about $130bn aggregate, with revenue down about 8.6%. Whether the control achieved anything against its target is unestimated, so the honest summary is a measured cost against an unmeasured benefit.

Frontier “China has been cut off from advanced chips.” The evidence supports ‘slowed at a cost’ and no more. A domestically fabricated 7-nanometre-class processor shipped in a consumer handset in 2023 without extreme-ultraviolet tools, at an undisclosed yield and cost — precisely the variables deciding whether this is a workaround or a capability.

Frontier “Sovereignty means self-sufficiency.” No serious programme targets it. Stated targets are market shares, capacity figures and money mobilised. Semiconductor self-sufficiency is estimated by the industry itself at over $1tn upfront and 35–65% higher prices, which is why nobody targets it. The practical objective is a lead time under interruption and it is never expressed as one.

Speculative “Standards bodies are the new instrument of technological power.” The claim is coherent and the evidence is participation counts. Chairmanships and contribution volumes are countable; influence over a specification, and rents captured from it, are measured at national level nowhere in the public literature. Handwave The version treating a national bloc’s committee votes as directly convertible into industrial advantage asserts the conversion step outright.

Handwave “Technological capability is national power, so any cost is justified.” This sentence does the real work in most public advocacy and has three defects in one clause: an unmeasured premise, a non-sequitur and an unbounded conclusion. A programme whose justification cannot name a price it would decline to pay has not been justified.