Sam Altman recently painted a vision where intelligence becomes a utility, consumed in exponentially growing token volumes. The narrative is seductive: as AI becomes as ubiquitous as electricity, token usage will rocket, and OpenAI will be the metering company. But as a researcher who reverse-engineered whitepapers during the 2017 ICO madness and watched the 2022 Terra collapse unfold, I see a familiar pattern: a narrative wrapped in technological inevitability, obscuring a structural fragility that few are willing to audit.
The Hook: A Prediction Without a Base Case
In an interview with Crypto Briefing, Altman stated that the consumption of AI tokens will grow exponentially, turning intelligence into a commodity like electricity. The quote is light on specifics—no base year, no growth rate, no price assumptions. It's a textbook example of a non-falsifiable narrative: if usage grows, he was right; if it doesn't, the technology wasn't ready. This is the same rhetorical strategy used by ICO projects in 2017, who promised exponential adoption of their 'utility tokens' without a single unit of demand.
safe
But the market is already pricing in this future. OpenAI's valuation, reportedly north of $80 billion, relies on the assumption that token revenue will compound at a rate that justifies massive infrastructure spending. The question is not whether usage will grow—it will. The question is whether the cost of generating those tokens will fall fast enough to keep the economics viable. And that's where the macro watcher in me starts to see a liquidity trap of a different kind.
Context: The Global Liquidity Map of Intelligence
Think of token generation as a function of compute, energy, and data center capacity. In 2023, global AI inference compute grew roughly 2-3x year-over-year, driven by LLM adoption. But the cost per token has only fallen by about 30-40% annually, according to publicly available API pricing. To sustain exponential token usage, the cost per token must fall faster than usage grows, otherwise total spending on AI tokens becomes a burden on enterprise budgets, not a utility.
During the 2024 spot Bitcoin ETF inflow study, I observed a similar phenomenon: institutional inflows did not immediately translate to price rallies because custody lag created a 'absorption phase.' Here, the absorption phase is the enterprise AI budget cycle. CFOs are already seeing AI bills expand 5-10x year-over-year, and they are starting to ask for ROI metrics. The 'exponential usage' narrative assumes that companies will keep paying regardless of unit economics. That assumption is untested.
Core: The Forensic Analysis of Token Economics
Let's break down the token-based business model. OpenAI charges per token, which is a unit of text. But 'intelligence' is not a homogeneous commodity. A token used for a simple chatbot query costs roughly the same to generate as a token used for a complex code generation? No, the latter requires more reasoning steps, but the pricing is flat. This creates an adverse selection problem: power users will gravitate toward the cheapest tasks, while high-value tasks may be priced out.
From my 2017 due diligence audit of Stratis, I learned to look for hidden assumptions in the whitepaper. Here, the hidden assumption is that token demand is elastic and that OpenAI can maintain pricing power. But the AI market is becoming commoditized. Google's Gemini, Anthropic's Claude, and open-source models like Llama 3 are all offering competitive pricing, sometimes at 10-20% of OpenAI's rates. If intelligence becomes a utility, it will be a commodity utility, like electricity, where the lowest-cost producer wins. OpenAI's unit costs are not publicly known, but the company's massive infrastructure spending suggests they are not the lowest.
safe
Moreover, the 'exponential' claim ignores the real-world constraints of compute supply. The global chip shortage in 2022-2023 showed that semiconductor production is not infinitely elastic. To sustain 10x annual token growth, we would need 10x more GPUs, which would require 10x more energy and 10x more data centers. The energy grid is already strained. The carbon footprint of AI is becoming a regulatory concern. The EU's digital euro pilot framework that I worked on in 2025 taught me that infrastructure-level scale requires public-private coordination, not just market forces.
Contrarian: The Decoupling Thesis
The contrarian view is that 'intelligence as utility' is a decoupling narrative—it decouples OpenAI's revenue growth from the underlying cost structure of the real economy. The market is currently pricing OpenAI as a growth stock, but the macro environment is tightening. Rising interest rates make capital-intensive infrastructure projects less attractive. The 2024 experience with Bitcoin ETF inflows showed that even when institutional money flows in, it doesn't automatically boost the underlying asset if the custody and settlement infrastructure is inefficient.
In the same way, even if token usage grows exponentially, the value of that growth may not flow to OpenAI. A parallel can be drawn with DeFi liquidity mining: projects subsidize TVL with token rewards, but when the rewards stop, the users leave. OpenAI's 'token utility' is subsidized by venture capital and cloud credits. Once the subsidies end, the true cost of token generation will be exposed. The 2022 Terra collapse taught me that stablecoins, like utility tokens, can unravel when the mechanism depends on perpetual growth. The same principle applies to AI tokens: if the growth rate falters, the narrative cracks.
Furthermore, the 'intelligence utility' framing centralizes power in a way that invites regulation. If AI becomes as critical as electricity, governments will demand price controls, service level agreements, and data sovereignty. OpenAI's profit margins, already thin, could be squeezed further. The company's own API pricing has been cut multiple times, indicating competitive pressure. The 'exponential usage' may be a mirage if the unit price keeps falling faster than the volume grows.
Takeaway: Positioning for the Cycle
We are in the early stages of a bear market for AI hype, just as we were for crypto in 2022. The narrative of exponential growth is a lagging indicator, not a leading one. The real leading indicators are: (1) the rate of decline in per-token cost, (2) the enterprise adoption rate of AI cost management tools, and (3) the regulatory response to AI as a critical infrastructure. As a macro watcher, I see the next 12-18 months as a period of 'cost governance' where the market will shift from 'how fast can we use tokens' to 'how valuable is each token.'
safe
Altman's prediction is a bet on the future, but it's also a marketing tool for his own company's fundraising. The smart money is on the infrastructure that enables low-cost token generation—energy, chips, and data centers—not on the token seller itself. The lessons from 2017, 2020, and 2022 all point to the same conclusion: when a narrative relies on exponential growth without a clear cost curve, the eventual correction is brutal. The question is not whether intelligence will be a utility, but whether the token-based model is the right architecture for that utility. My bet is that it will be replaced by a more efficient, decommoditized alternative before the exponential growth even materializes.
