What does AI as a utility actually mean?
AI as a utility refers to a model where artificial intelligence is billed by usage — like electricity or water — rather than through flat-rate subscriptions. Speaking at the BlackRock Infrastructure Summit in Washington, DC on March 11, 2026, OpenAI CEO Sam Altman argued that this shift is not a distant aspiration but the inevitable direction of the industry.
“We see a future where intelligence is a utility like electricity or water and people buy it from us on a meter and use it for whatever they want to use it for,” Altman said. The unit of exchange he envisions is the token — the basic input/output unit AI systems use to process text, code, and data. “Fundamentally our business and I think the business of every other model provider is going to look like selling tokens,” he added. That is a striking admission: Altman is not just predicting OpenAI’s future, he is predicting the entire industry’s.
From subscriptions to metered billing: why the shift matters
Today, most consumers and businesses pay flat monthly fees for AI access — a familiar SaaS model. The move to metered billing would fundamentally change the economics of AI adoption. Heavy users would pay more; casual users would pay almost nothing. This mirrors how cloud computing evolved from server leases to pay-per-use pricing, and it has the same disruptive potential for every sector that relies on knowledge work.
The practical implication is significant. Altman described a near-future where AI handles not just quick queries but multi-day and multi-week projects, operating proactively like a senior employee rather than a tool waiting to be prompted. Workplaces are already shifting, he argued — employees increasingly guide AI rather than perform technical tasks themselves. He pointed to startups in India building what he called “zero person” companies, where AI handles software development, legal work, and customer support without a human workforce. Whether that vision is prescient or promotional depends heavily on infrastructure that does not yet exist at the required scale.
The infrastructure bottleneck threatening Sam Altman’s AI utility vision
Altman’s ambition runs directly into a physical constraint: power. AI data centers consume electricity at the scale of small cities, and the US grid is already under strain from transformer shortages and slow permitting processes. OpenAI is investing in gigawatt-scale data center campuses and has formed partnerships with Amazon, Nvidia, SoftBank, and North American building trades unions to accelerate construction.
The goal, as Altman framed it, is to make intelligence “too cheap to meter” — to flood the world with AI capacity so abundant that scarcity becomes irrelevant. That is an extraordinary claim, and it deserves scrutiny. Elon Musk, speaking on the Moonshots with Peter Diamandis podcast in January 2026, argued that electricity generation itself is the binding constraint on AI scaling, and predicted that China could outpace the US in AI compute simply by building out energy infrastructure faster. Whether the US can resolve grid bottlenecks and permitting delays quickly enough to sustain Altman’s vision is an open question — and arguably the most important one in the AI industry right now.
AI as a utility and the labor question nobody is answering honestly
Altman acknowledged the social friction his vision creates. “Data centers are getting blamed for electricity price hikes. Almost every company that does layoffs is blaming AI, whether or not it really is about AI,” he said. That candor is notable. The term “AI washing” — companies attributing workforce reductions to AI adoption to signal modernity rather than because AI actually drove the decision — is a real phenomenon, and Altman is right to flag it.
But the broader disruption to the labor-capital balance is genuine, not manufactured. If AI as a utility becomes reality, the nature of skilled work changes permanently. The question is not whether that happens, but how fast, and whether the infrastructure and regulatory frameworks can keep pace. Other AI model providers will face the same pressure to shift toward token-based selling, as Altman himself predicted. The companies that build the most reliable, lowest-cost token infrastructure will define the market — much as Amazon Web Services defined cloud computing by making compute so cheap and accessible that it became invisible.
Will AI really become as ubiquitous as electricity?
Altman’s vision is that future generations will view AI the way people today view electricity — a background utility so fundamental it is never questioned. That is a compelling framing, but it papers over the differences. Electricity is a physical commodity with well-understood physics. AI inference depends on model quality, data freshness, and hardware that evolves rapidly and unpredictably. Metering intelligence is not as straightforward as metering kilowatt-hours.
What is a token in AI billing?
A token is the basic unit AI systems use to process input and output data — roughly equivalent to a word fragment or a few characters of text. In a metered AI billing model, users would be charged based on how many tokens their queries consume, making costs directly proportional to usage rather than a flat subscription fee.
How does OpenAI’s utility model compare to other AI providers?
Altman stated that all major AI model providers will eventually shift to token-based selling, not just OpenAI. The competitive dynamic will therefore center on price per token and infrastructure reliability rather than feature differentiation — a commoditization race that favors whoever can build the most efficient, large-scale data center infrastructure fastest.
Altman’s metered intelligence vision is either the most accurate forecast of AI’s economic future or the most ambitious piece of infrastructure salesmanship in recent tech history — possibly both. The physical constraints are real, the grid challenges are documented, and the labor disruption is already underway. What remains unproven is whether OpenAI, or anyone else, can actually build the gigawatt-scale infrastructure required to make AI as a utility a lived reality rather than a summit keynote abstraction.
Edited by the All Things Geek team.
Source: Windows Central
