Sam Altman recently declared that intelligence will become a utility, with token consumption growing exponentially. The truth is: this is a business narrative dressed in technological clothing. Logic doesn't care about narratives. It only cares about constraints—cost curves, infrastructure limits, and incentive misalignments.
Context: The Utility Pitch
The statement, reported by Crypto Briefing, frames OpenAI's pay-per-token model as the future of universal intelligence. Altman's vision is seductive: smart electricity, metered by the token. But the comparison to electricity is structurally flawed. Electricity became a utility because its unit cost dropped by orders of magnitude over decades, enabling mass adoption. Token costs, while declining, have not followed a sustainable, exponential-downward curve. The article lacks any data on token price elasticity, base usage, or time horizon. Without these, “exponential growth” is a marketing term, not a forecast.
Core: The Hidden Arithmetic
Let me be precise. The transformer architecture ties token consumption directly to inference compute. Exponential token usage means exponential inference demand. During my 2017 Ethereum testnet triage, I learned that every exponential growth claim must be backed by measurable efficiency improvements. For Altman's vision to hold, the cost per token must drop faster than usage grows. Otherwise, enterprise clients face runaway AI bills—a problem the article's author correctly identifies as needing new cost management strategies.
The cost curve is the elephant in the room.
OpenAI has lowered API prices, but the reductions are incremental, not orders of magnitude. Based on my experience auditing Compound's interest rate model, I know that compounding growth assumptions can hide implementation fragility. If token volume grows 10x while price drops only 2x, the total cost to users grows 5x. That is not utility; it is a cost crisis.
The business model tension
Altman's narrative repositions OpenAI from a lab to an infrastructure company. That is a valuation play. But utility services face natural monopolies, public oversight, and price regulation. If OpenAI becomes the “grid,” it loses pricing freedom. The article brushes past this, but the tension is real: utility status invites regulatory intervention, which caps profit margins. Greed is the feature; the bug is just the trigger.
Competition and commoditization
If intelligence is a utility, it becomes a commodity. The winners are low-cost producers, not the loudest brand. Open-source models are closing the gap. Google, Anthropic, Meta—all are racing to lower inference costs. Altman's exponential narrative may be a preemptive strike to maintain premium pricing before the market forces unit costs down. I don't trust exponentials without cost curves. The market does not reward hype; it rewards the lowest cost at sufficient quality.
Infrastructure: the real bottleneck
Exponential token consumption requires exponential compute and energy. During the Axie Infinity exploit, I saw how design flaws ignored real-world constraints. Here, the constraint is physical: data centers, power grids, chip supply. The article does not address this. If token demand grows even 50% year-over-year, the current infrastructure will buckle. The real utility opportunity lies not in AI tokens but in the hardware and energy that power them. You didn't ask the right question: who owns the grid?
Contrarian: What the bulls got right
To be fair, Altman's utility framing has a kernel of truth. The trend toward pay-per-use AI is real. Enterprises are already struggling with AI cost governance. The emergence of FinOps for AI confirms that token consumption is a new cost center. The article's author is correct that new consumption and cost management strategies are needed. And Altman's long-standing interest in UBI aligns with the redistributive implications of a universal intelligence service. The mistake is assuming the growth will be smooth, unregulated, and profitable.
Takeaway: The accountability call
Altman's exponential token forecast is not a prediction—it is a fundraising pitch. The real test will come when enterprise CFOs demand to see the unit economics. Until then, treat every exponential claim as a hypothesis to be falsified, not a fact to be funded. The exploit wasn't in the code; it was in the assumption. Logic doesn't care about narratives. It only cares about the math. And the math on token utility doesn't add up—yet.