Three Scenarios for 2050 — and the Signals That Distinguish Them

Why scenarios rather than forecasts

A forecast asserts what will happen. A scenario describes a coherent way things could go, states what would have to be true for it, and identifies what you would observe early if it were occurring.

The second is more useful over a twenty-five year horizon, because the honest confidence interval on a single forecast that far out is wide enough to be useless. What can be done well is bounding the space and identifying the signals that distinguish the branches.

Scenario one: fragmentation

Technology stacks separate along geopolitical lines. Incompatible standards, restricted data flows, and duplicated infrastructure become permanent rather than transitional. Multinational firms maintain parallel operations by jurisdiction.

This requires that the security logic driving separation continues to outweigh the efficiency losses — plausible, since the costs fall on diffuse consumers while the benefits accrue to concentrated and organised interests.

Early indicators: divergence rather than convergence in technical standards bodies; data-localisation requirements extending from sensitive to ordinary commercial data; firms building genuinely duplicated stacks rather than compliance wrappers over a shared one. That last is the strongest signal, because duplication is expensive and only undertaken when fragmentation is believed permanent.

Scenario two: federation

Blocs maintain distinct governance but agree interoperability at defined interfaces — a settlement resembling international aviation or telecommunications, where systems differ but connect through negotiated protocols.

This requires functioning multilateral technical institutions and enough commercial pressure to make interoperability worth the sovereignty concession. It is the outcome most participants would prefer and the hardest to reach, since it demands sustained cooperation among parties who do not trust one another.

Early indicators: mutual recognition agreements for AI evaluation and certification; cross-border data frameworks that survive a change of government on either side; standards bodies producing adopted text rather than deadlocking. Watch particularly for agreements that hold through a diplomatic incident — that is the test of a genuine federation.

Scenario three: consolidation

Capability concentrates in a small number of providers — possibly corporate rather than national — whose infrastructure becomes the substrate everyone else builds on. States regulate but do not meaningfully substitute.

This requires returns to scale in AI to remain steep enough that no entrant can reach the frontier. That is currently true and is a genuinely open question over a decade.

Early indicators: sustained widening between frontier and open-weight capability; national AI programmes quietly reducing scope from building to procuring; concentration of the specialised talent pool. The most telling signal is the ambition of national programmes, since it reveals what governments privately believe about their prospects.

What to do given uncertainty

Since the branch is undetermined, the defensible strategy is to invest in what holds across all three.

Adaptation and evaluation capability is valuable in every scenario — it is how a state or firm extracts domestic value from technology it did not build, whether that technology comes from a bloc partner, a treaty counterparty or a dominant provider. Data governance capability likewise. Institutional credibility likewise.

Conversely, the investments that only pay off in one branch — frontier pretraining ambitions, or deep dependence on a single external provider — should be sized as bets rather than as plans.