You use it every day. You have no idea where it runs, who owns it, or what it costs.
Literal meaning: AI as a utility describes the framing of artificial intelligence as infrastructure — like electricity, water, or internet connectivity — that is always available, consumed on demand, and whose production costs are not visible at the point of use.
Origin: The utility framing emerged from cloud computing discourse and was actively promoted by major AI providers from 2022 onward, as large language models became accessible via API. Amazon, Microsoft, and Google explicitly positioned their AI services as utility infrastructure. The framing draws on economic theory: utility goods are necessities whose provision benefits from scale and standardisation.
AI delivered as infrastructure — always available, apparently free, with the real costs and dependencies hidden from the people who use it.
The Appeal: The utility model genuinely delivers value. Ubiquitous, low-friction access to powerful AI tools democratises capability — tasks that once required specialists are available to anyone. For organisations and individuals, the infrastructure metaphor is accurate in one sense: AI is becoming as integrated into workflows as email or search.
The Friction: Utilities have regulators, accountability structures, and transparency requirements. AI as utility does not — yet. A concrete case shows the asymmetry: in June 2026, the US Commerce Department ordered Anthropic to suspend access to its most advanced models, Fable 5 and Mythos 5, for all foreign nationals — an export control directive citing an undisclosed national security concern about a potential jailbreak that could aid the discovery of software vulnerabilities. Anthropic disputed the justification but complied. The switch runs both ways: months earlier, Anthropic itself had refused to let the US military use its models for autonomous weapons systems, prompting the Pentagon to blacklist the company — until the Pentagon’s CIO put it plainly: “Some things are simply more important than revenue… and pre-IPO valuation. America First. Always.” Electricity has a regulator with due process; this exchange had an export-control order on one side and a corporate policy decision on the other, neither accountable to the people who depend on the infrastructure. The consumer stake has the same shape at smaller scale: access is revocable without warning, and whatever was built on top of it — institutional workflows, an accumulated way of working with a specific model — can disappear the moment access does, with no equivalent to a utility regulator’s continuity-of-service obligation. Externalized Costs — environmental burdens transferred to communities — are invisible in the utility model: the user sees a query box, not the data centre drawing power from a stressed grid. AI Dependency — structural reliance without critical reflection — is what the utility model systematically produces: infrastructure that recedes into the background is infrastructure you stop thinking about, evaluating, or being able to replace. Cognitive Offloading becomes structural when the offloading infrastructure is as invisible as electricity.
Why This Matters: When AI becomes a utility, it stops being a choice. Once something is infrastructure, opting out carries the same cost as opting out of electricity. The moment to ask questions about AI is before it becomes invisible — not after.
Related terms: AI Dependency · Externalized Costs · Cognitive Offloading · Deskilling · AI Literacy · Attention Economy
Read more:
- The Alignment Problem — Christian, B. (2020). W.W. Norton
- Power and Progress
- Sam Altman wants to turn AI into a utility - and meter “intelligence” like electricity
- Anthropic to disable its most advanced AI models after US order limiting foreign access — Reuters, in The Guardian (2026, 13 juni)
- AI as Public Infrastructure: A Critical Review of the Transition from Tool to Societal Necessity — Ahmed, R. et al. (2026). ResearchGate preprin— Warren, T. (2026). _