Governments are not failing to invest in AI-native infrastructure for lack of capital or political will. They are failing because the instruments they use were designed for a different asset class. InvestEU, TEN-T funding, standard procurement criteria — all were built for infrastructure that depreciates. Applied to learning systems, they systematically produce the wrong allocation decisions.
The wrong instruments for the right assets
Cost-benefit analysis discounts future performance. Procurement rewards lowest upfront cost. Budget cycles run on three-to-five year horizons.
These rules were correct for assets that depreciate. Applied to learning infrastructure — systems whose performance improves with operational experience — they penalise precisely the long curves that compound into strategic advantage.
Learning first, leverage later
A viable capital architecture for AI-native infrastructure is sequenced capital — each tier matched to a specific layer of risk and learning horizon. The design principle: without patient capital in the early stages, intelligence defaults upward to platforms; without governance protecting learning, it fragments.
-
Public balance sheets
Absorb early deployment risk. De-risk learning curves. Signal strategic commitment that crowds in private capital.
Sequence · First in -
Development banks
Patient capital, first-loss absorption, cross-border standardisation, auditability enforcement. The EIB as archetype.
Sequence · Early stage -
Sovereign & pension capital
Scale once learning curves stabilise. Natural alignment with multi-decade horizons and compounding resilience.
Sequence · Mid-stage -
Disciplined private capital
Participates where governance protects learning from premature extraction and preserves execution autonomy.
Sequence · Last in
Three things that must change before the architecture hardens
The digital economy offers a precise warning: governance is nearly impossible to impose after platform architectures are established. Procurement frameworks and financing instruments are being designed right now. Three changes are required before they crystallise.
-
Redesign procurement criteria.
Availability-based contracts that reward long-term performance improvement — not lowest upfront cost. Frameworks that evaluate an operator’s learning potential alongside engineering specifications.
-
Adapt existing instruments.
InvestEU and TEN-T were designed for depreciating infrastructure. They require blended finance where public capital absorbs first-loss risk, longer performance horizons, and explicit learning-curve metrics as evaluation criteria.
-
Establish data governance before the intelligence layer forms.
Operational data generated by public infrastructure must remain under public governance. Auditability standards, data portability requirements, vendor contract structures — in place before procurement locks in the intelligence layer.
Governments are not short of capital for the velocity economy. They are short of instruments designed to deploy it correctly — and that difference determines whether public investment accelerates or impedes the transition.