Five Games, One Board

I — One Word, Five Games

Every major economy now has an AI strategy on paper. Almost none of them states which game it is playing. The documents share a vocabulary — leadership, sovereignty, competitiveness, the race — and the vocabulary is the problem: "winning the AI race" treats what is actually a five-board tournament as a single sprint toward a single finish line.

The boards differ in everything that matters to a strategist: the price of entry, the shape of the payoff, how fast an advantage decays, and what failure looks like. Parts I, II and III of this series described the physics of the board — where the chokepoints sit, how access is hardening into blocs, what law can and cannot move. This part is about the play. Five games, in descending order of glamour and, not coincidentally, ascending order of accessibility.

II — Game 1 — Full-Stack

The full-stack game means fielding all of it: frontier labs, hyperscale infrastructure, leading-edge silicon access, and the capital markets to fund the compounding. Two players hold tickets — the United States and China, which between them account for roughly ninety percent of the world's AI compute performance. (Pilz et al., 2025) The entry price is measured in trillions of dollars and decades of accumulated ecosystem, which is why the honest version of every other country's strategy begins by admitting this game is closed.

What the sprint framing misses is that even the two ticket-holders face constraints they cannot legislate away. The American constraint is social license: at least 69 local governments have enacted data-center bans or moratoriums, blocking well over a hundred billion dollars of projects in a single quarter — a ceiling imposed not by adversaries but by neighbors, and invisible in every federal strategy document. (Gilbert, 2026) The Chinese constraint is delivery: the Eastern Data, Western Computing program built capacity far from demand, and latency has left many facilities running at a fraction of utilization — building is not the same as delivering. Silicon access compounds the gap at the leading edge.

The conclusion is not that the poles are weak. It is that neither "wins" in the sprint sense. They anchor blocs, set the terms everyone else lives under, and grind against their own physical and political limits. For everyone else, Game 1 exists mainly as a temptation — and as the most expensive failure mode on the board: national-champion cosplay, Game 1 ambition funded with Game 5 resources.

III — Game 2 — Chokepoint Tenancy

The second game is to hold one node the system cannot route around. The Netherlands holds lithography. Taiwan holds the leading-edge fab. South Korea and Japan hold memory, materials, and precision equipment. Measured per capita, these are the most leveraged positions on the board — a country of eighteen million people holds veto-adjacent influence over the entire trajectory of the technology.

A chokepoint is a tenancy. The rent is real; the lease is written elsewhere.

But the tenure is the point, and the word matters: this is a tenancy, not ownership. The leverage exists only inside an alliance structure that could reassign it; it depreciates as rivals replicate the node; and it can be traded away over the tenant's head in a grand bargain between the poles. The correct play, therefore, is to price compliance rather than donate it — which is exactly what the Dutch objection to Washington's proposed allied export-control alignment represents. (US Senate, April 2026) That episode is routinely reported as friction. It is better read as a chokepoint tenant negotiating rent.

The metric that matters in this game is not market share but years-to-replicate: how long would it take the rest of the system to route around your node? When that number starts falling, the rent is expiring, and the strategy question becomes what to convert the remaining leverage into.

IV — Game 3 — Physical-Inputs Arbitrage

The third game converts endowments into capacity: firm power, land, cold climate, water, and patient capital, assembled into compute that the capability owners need somewhere to run. The Gulf states are playing the aggressive version — courting both blocs simultaneously and selling position as well as power. Canada is playing the defensive version: its recently announced nationally-owned compute capacity is motivated less by ambition than by jurisdiction — capacity that no foreign statute can reach. (Gilbert, 2026) The Nordics play the quiet version, selling reliability and cool air.

The underrated feature of this game is that the product is only partly megawatts. What the buyers increasingly pay for is jurisdictional quality: legal predictability, neutrality between blocs, enforceable contracts, acceptable data law. A rack in a stable, neutral, rule-of-law jurisdiction is a different asset from the same rack somewhere contested — which is why this game rewards small, credible states out of proportion to their size.

Its ceiling is equally clear. Capacity without models or demand is real estate. The landlord earns rent, sometimes excellent rent, but rent is not power: the tenant can leave, the technology can shift, and the pricing is set by people playing other games. Game 3 is a sound position and a poor identity.

V — Game 4 — Diffusion

The fourth game is the unglamorous one, and it is where most of the value has always been. Every general-purpose technology in the historical record — electricity, the internal combustion engine, information technology — paid out its returns not at the invention layer but at the adoption layer, over decades, to the economies that reorganized themselves around the new input fastest. (Bresnahan & Trajtenberg, 1995; David, 1990) There is no obvious reason AI breaks the pattern, and early evidence suggests it doesn't.

Concretely: a country that gets working AI into its hospitals, factories, ministries, courts, and small businesses two years faster than its peers gains more compounded GDP than a country hosting a frontier lab whose capability never leaves the API. Diffusion is won with instruments no one puts on a poster — skills at scale, data infrastructure that actually interoperates, procurement rules that let the public sector buy what works, regulatory clarity that lets firms deploy without heroic legal budgets, and cheap, reliable inference access treated as economic plumbing rather than strategic theater.

Every economy can play the adoption game, from any starting position — and almost none does so on purpose.

The political economy explains why the game is neglected: there is no ribbon-cutting for "we adopted faster," no gigafactory photo-op, no sovereign champion to name. Diffusion produces statistics, not monuments. But it is the only game on the board with no entry ticket and no ceiling — every economy can play it, from any starting position, and the returns scale with seriousness rather than with endowments. For most countries most of the time, it is not a component of AI strategy. It is the strategy, and everything else is hedging.

VI — Game 5 — Terms-Setting

The fifth game belongs to whoever controls access to demand rather than supply. The European Union cannot field a frontier stack, but it can price entry to four hundred and fifty million wealthy customers and export the world's most-copied regulatory template. The AI Act, CADA's assurance levels, procurement conditionality: assembled, they are buyer power dressed in the language of sovereignty — and buyer power is real power on the commercial layer. (European Commission, 2026)

Part III of this series traced where that power stops: at the national-security layer, terms-setting hits a wall, because the asset that would convert regulatory sovereignty into operational sovereignty — control of the frontier capability itself — is the one asset no provider will surrender and no procurement rule can compel. The graded honesty of CADA's own projections concedes the point.

The coherent version of the European position, then, is not frontier parity — Game 1 played with Game 5 resources — but a deliberate portfolio: terms-setting on the commercial layer, an all-in commitment to Game 4, and two or three vertical bets where Europe's industrial data and domain depth are genuine edges rather than aspirations — industrial AI, health systems, energy optimization. That is a smaller story than "sovereignty." It is also one Europe could actually win.

VII — The Portfolio Test

Across all five games, three properties predict success, and they can be read off any national strategy document in an afternoon.

First: hold at least one node the system cannot route around, or out-diffuse your peers — ideally both, because the combination is self-insuring. A country with neither is a price-taker in every scenario, whatever its documents claim.

Second: keep claimed sovereignty equal to actual control. The gap between the two — sovereignty on slides, dependence in operations — is not merely embarrassing; it is strategic self-deception with a budget line, because resources flow to defending the claim instead of managing the exposure. Part III's argument in one sentence: the countries that name their dependencies govern them; the countries that deny them are governed by them.

Third: hedge the capability curve, because the games reprice against each other depending on how frontier progress unfolds. If capability keeps steepening, the poles pull away, chokepoint rents rise, and access becomes the decisive asset. If it flattens, open weights commoditize capability from below, chokepoints depreciate, and the diffusion players quietly collect the prize while the race's spectators are still watching the frontier. A robust national strategy is one that does not need to know which scenario arrives — which, again, points every non-pole country toward diffusion as the scenario-proof core.

The failure modes compress to a sentence: champion cosplay, blocking without building, and sovereignty that never gets priced.

— Five Finish Lines

So read your own country's strategy document with the board in view. Which of the five games does it think it is playing? Does it say so? Can it afford that game's ticket — and if not, which game could it actually win with the endowments it has, rather than the ones its rhetoric assumes?

Most documents will fail the test in the same direction: too much Game 1 language, too little Game 4 plumbing, and a sovereignty chapter that never meets a price tag. That is the quiet diagnosis running through this whole series, and it ends where strategy always ends — not with the race as metaphor, but with the board as fact.

The race has five finish lines. Most nations are sprinting toward the one they cannot reach, past the one nobody guards.