AI is rewriting geopolitics, not just policy

The Atlantic Council’s “Eight ways AI will shape geopolitics in 2026” is a solid map for diplomats and defense types. But yeah, no, the map mostly stops at the waterline of the nation-state. The piece treats AI as a set of levers that presidents and generals will pull—without asking who actually owns the levers, maintains them, and decides who else gets a turn.

That omission matters, because in this story the vendors are half the battlefield.

The vendor is the new battleground

The article sketches rivalry in terms of capitals, doctrines, and alliances. Necessary, sure. But control over compute, model architectures, training data, and cloud infrastructure now shapes power as much as anything decided in a situation room.

Cloud providers, chip manufacturers, and model labs are the practical gatekeepers. They determine who gets access to frontier models, how quickly updates spread across borders, and which sanctions or export controls quietly fizzle. If you’ve watched firms like Microsoft or Amazon structure their cloud regions and data residency promises, you know these aren’t just IT decisions—they’re foreign policy by other means.

This isn’t hypothetical. Policies get blunted when private incentives run in the opposite direction. Contracts, jurisdictions, and developer ecosystems can race ahead of government guidance, then quietly box policymakers into a corner. That yields a geopolitical posture where influence is projected through commercial dependencies—data flows, platform lock-in, MLOps pipelines—rather than only embassies and treaties.

It starts to look less like classic statecraft and more like corporate diplomacy with better branding. William Gibson’s Neuromancer didn’t foresee today’s model weights, but he absolutely nailed the vibe: the owners of cyberspace set the price of admission, and the price is rarely just money.

Export controls on chips or advanced models might slow some programs, but they don’t magically stop determined actors who can rent capacity in friendlier jurisdictions or reconstruct capabilities from widely available model weights. So if you’re talking about AI and geopolitics and only tracking who leads in R&D, you’re missing the harder question: who runs the rails, and under which terms.

Small states and proxies as spoilers

Funny thing is, the Atlantic Council piece leans, understandably, into great-power rivalry. But the architecture it describes is being rented, not just owned—and that gives small states and non-state actors a way to punch above their weight.

Commercial AI platforms flatten capability differences. A small state with cash and competent procurement can assemble intelligence analysis pipelines, influence tooling, and automated decision-support that would once have required a permanent bureaucracy and a national lab. They don’t need to build the stack; they just need to negotiate the right set of service agreements and integration partners.

Non-state actors—criminal networks, mercenary tech outfits, political consultancies—can assemble similar stacks by renting compute and contracting specialized firms. The work is splintered across cloud providers, model vendors, and data-labeling shops spread across multiple countries. That diffusion buys deniability, makes attribution fuzzier, and stretches traditional concepts of escalation. You’re not just worrying about missiles and troop movements; you’re worrying about a synthetic media shop quietly plugged into an ordinary enterprise cloud account.

Policymakers who still design doctrine as if the main adversary must be a flag-waving state actor are going to misjudge both risk and response.

Yes, states still matter—but not like they think

Now, the obvious rebuttal: states still hold the big sticks. They command militaries and intelligence agencies, write laws, control borders, and can subpoena, sanction, or, in extreme cases, nationalize critical infrastructure. If a government decides a capability is existential, it can drag it into the state apparatus.

Sure, but that response glosses over two hard constraints. Economies are tightly linked, so overt nationalization is economically and politically expensive. And tech talent is mobile; push your firms too hard into a state-centric mold and your best engineers quietly accept offers elsewhere. States can behave like landlords, but they’re rarely the sole inventors anymore.

Look at something like OpenAI’s partnership with Microsoft or Google’s positioning of its AI stack: these are global platforms threading needles between different regulatory environments and national strategies. No single capital fully dictates their moves; it’s a negotiation between sovereign interests and shareholder value. In a lot of day-to-day decisions, corporate incentives win the contested inch.

The blind spots that actually bite

Three blind spots deserve more space than they typically get in geopolitics talk.

First, data sovereignty quietly shapes capability. Whoever can legally and technically aggregate, clean, and retain certain classes of data ends up with a durable edge in domain-specific models—urban logistics, financial crime patterns, regional language nuance. AI without rich data is just an expensive calculator.

Second, interoperability and standards will define whether alliances are functional or ceremonial. You can sign a security pact, but if your partners’ AI systems can’t safely share models, logs, or threat assessments—because their vendors don’t talk to each other or their legal regimes clash—coordination frays right when it matters most. NATO without compatible AI tooling is just a very large book club.

Third, deterrence theory hasn’t really caught up to non-kinetic AI tools. Retaliation logic is murky when the hostile act is a tailored disinformation campaign run on rented infrastructure, or subtle interference in logistics systems. Traditional red lines don’t map cleanly onto adversaries who can rent, outsource, and deny.

Contracts as quiet doctrine

The Atlantic Council is right to flag AI as a major axis of competition and governance. Look, if you don’t center contracts, cloud regions, chip supply chains, and the brokered relationships between states and tech firms, you’re missing the plumbing that decides which strategies survive contact with reality.

By 2026, the sharpest geopolitical fights over AI may not be framed as “AI policy” at all—they’ll show up as procurement clauses, licensing terms, and cross-border data arrangements that quietly set the rules of power.