Beneath the baroque facade, the ledger bleeds.
In the week that Sam Altman reportedly told a gathering that intelligence would become a utility and that token consumption would grow exponentially, the crypto Twitter machine did what it always does: it translated the statement into a ticker symbol. Somewhere, a chartist drew a projection of an AI token hitting escape velocity. Somewhere else, a retail trader remembered that the word token has a history of making people rich. But the real discovery is not in the sentence, and not in the chart. It is in the pricing page of OpenAI’s API, where every prompt, every completion, every reasoning call is counted, metered, and billed in units of a token. That metering architecture is the quiet code beneath the noise.
Let me be clear about what is known and what is not. The article in question is thin. It contains an assertion attributed to Altman, an inference about enterprise cost-management strategies, and little else. No specific data, no date, no counterargument, no third-party verification. Crypto Briefing is a publication that sits at the intersection of two communities that frequently confuse each other: crypto traders and AI observers. In crypto, a token is an asset, a bearer instrument, a claim on a protocol or a status. In AI, a token is a fragment of text, a statistical atom in a language model. These are not the same thing. The sentence token usage will grow exponentially is grammatically banal in the AI context, but in the crypto context it vibrates with speculative electricity. That semantic mismatch is not a bug; it is the mechanism by which opinion pieces become investment narratives.
As an analyst who has spent two decades observing the intersection of monetary infrastructure and technological narrative, I have learned to treat the absence of numbers as itself a number. The absence of data in the original piece tells me that the writer was not interested in verification. The writer was interested in atmosphere. That is fine for a meme. It is not fine for a market. Let me reconstruct the structure underneath the prose.
The Meter Is the Message
Altman’s utility framing is not a forecast. It is a retroactive rationalization of a pricing model. OpenAI has been charging by token since the API era began. To make that pricing model feel inevitable, you need a story. What story is more magnificent than electricity? The electric meter did not create electricity; it standardized it. It made power negotiable at the margin, purchasable by the hundredth of a kilowatt-hour. In the same way, the token meter creates a unit of exchange for a statistical process.
But there is a crucial difference. A kilowatt-hour is a physical quantity with an energy standard. A token is a moving target. The same token that means apple in a fruit-picking context becomes a legal liability in a securities filing. Value is context-dependent, and context changes from user to user, from prompt to prompt. The unit is not yet a standard. It is a billing convenience.
I have spent years reading whitepapers at a crypto investment bank, and I have learned to look for the moment when a protocol confuses its own unit of account with its unit of value. The utility narrative performs exactly this confusion. It asks the market to believe that a token, because it is countable, is therefore meaningful. But countability is not value. The difference between a token and a watt is the difference between a statistical guess and a physical law. A watt does not hallucinate. A token does.
The Recursion Flaw in Altman’s Architecture
Ten years ago, I sat in an apartment in Le Marais and manually audited 42 Ethereum whitepapers. The habit taught me to look for the failure mode inside the architecture. Altman’s architecture has a recursion flaw. It assumes token demand can grow exponentially while token price remains elastic enough for adoption. It assumes a unit of measurement can be both a cost and a value. But in my audits, the moment a protocol began using its own token as both the fuel and the measure of value, the books stopped adding up.
The same applies here. If OpenAI’s revenue is denominated in tokens, and if the value of intelligence is also denominated in tokens, then any increase in token volume can be interpreted as either more intelligence or more cost. The statement selects the optimistic reading. It ignores the possibility that token volume will grow because agents are inefficient, because models repeat themselves, because content bloat has replaced careful reasoning, or because poor prompt design burns tokens the way a leaky boiler burns gas.
The phrase token usage will grow exponentially is not a model. It is a hope. A model requires a denominator: price per token, cost per token, energy per token, revenue per token. None of those appears in the original article. Without a denominator, exponential growth is a Rorschach test. The AI community sees scale. The crypto community sees a bull run. The CFO sees an invoice. The energy trader sees a power purchase agreement.
The Missing Denominator
Let me put the financial model on a whiteboard. Revenue equals token volume times price per token. Beneath that sits net income: subtract inference cost, energy cost, capital depreciation, and human labor. Altman says the volume grows exponentially. Fine. But what about the price? Since the market is competitive and open models exist, the price is under structural pressure to decline. If volume grows one hundred times and price falls ninety times, revenue grows only ten times. If inference cost declines only fifty times, margins still get squeezed. The exponential growth of token consumption does not automatically become exponential revenue growth. It becomes exponential compute and energy procurement. The narrative is a demand-side argument with the supply side missing.
This is not a trivial accounting problem. It is the core of the entire story. OpenAI is not simply an AI laboratory; it is a capital-intensive infrastructure company. Its marginal cost is not zero. Every token carries a physical cost: electrons, silicon, cooling water, network bandwidth, and the amortized capital embedded in a data center. As token usage grows, OpenAI must procure more computing power, more electricity, and more warehouses. The narrative of intelligence as a utility hides the fact that utility companies are capital-heavy businesses with steady but modest returns. Tesla, not OpenAI, is the better analogy for the tension between narrative and capital.
I have modeled institutional capital flows into Bitcoin ETFs and the crypto market, and I have repeatedly seen a pattern: the asset that is most necessary to a future economy is bid up first, not the asset that promises to be the future economy. If token consumption grows exponentially, the immediate beneficiaries are the providers of upstream resources: the chip fabricators, the power utilities, the environmental compliance consultants, the cooling system manufacturers. They do not need the AI token narrative. They need the energy meter. The macro does not whisper; it screams in silence.
Utility, Monopoly, and the Political Meter
There is a deeper contradiction in the utility analogy. Real utilities are not free-market dreams. They are regulated monopolies. They have obligations to serve, price caps, grid codes, and public oversight. A utility cannot refuse to sell power to a hospital because the hospital is unprofitable. A utility cannot suddenly raise rates during a snowstorm without facing political revolt. If intelligence becomes a utility, OpenAI will lose its pricing freedom. It will be treated like the electricity company, not like a luxury software subscription.
Altman seems to understand this. His association with Worldcoin is not an accident. A global token of personhood, an iris-scanning orb, a universal basic income protocol: those are the building blocks of a utility world. If intelligence is a basic service, someone must decide who gets it, at what price, and under what conditions. That is not a technical problem. That is a political economy problem.
But the regulatory path cuts against crypto’s instincts. Regulated utilities settle in regulated money. They do not settle in anonymous stablecoins. They are audited by state bodies. They are forced to report outages, discrimination complaints, and price adjustment petitions. The more OpenAI becomes a utility, the less it resembles a decentralized protocol. It becomes a counterparty. The market should not conflate Altman’s vision of abundance with the crypto anarchist dream of trustless exchange. They are opposed in practice.
Agents, FinOps, and the New Governance Layer
Now let’s talk about the most plausible part of Altman’s prediction. Agents will multiply token consumption. A human who asks a chatbot a question consumes hundreds of tokens. An agent that reads an email, checks a calendar, searches a database, compares legal clauses, and writes a draft can consume hundreds of thousands of tokens. The move from interactive AI to agentic AI is not a linear increase; it is a multiplier. That is the strongest part of the story.
But it cuts in two directions. For enterprise buyers, the multiplication of token consumption means the multiplication of cost. The new consumption and cost management strategies that the original article mentioned are not an afterthought. They are the market. We are entering the era of AI FinOps. Every large company will need an audit function for token spend.
Which model should handle which task? When should a task be cached? What is the maximum response length? Should a local small model handle routing and reserve the frontier model for high-value reasoning? Should the company build a central AI gateway to enforce token budgets across departments? These are not toy questions. They will be answered by a new layer of infrastructure that sits between the model and the money.
The existence of this governance layer creates an adversarial tension. The more effectively the middle layer optimizes token usage, the less raw token volume flows to the model provider. The utility provider wants volume; the meter reader wants efficiency. This is not a win-win. It is a fee battle. I saw the same dynamic in DeFi during the liquidity wars of 2020. Tooling providers promised to reduce risks, but their real business was extracting fees from the flow. We trade in shadows cast by invisible hands. The hands are the middleware companies that control the visibility of the token ledger.
The Energy Denominator
Every token has a physical cost. The macro does not whisper; it screams in silence. In 2023 and 2024, we watched hyperscalers triple their capital expenditures, acquire nuclear power plants, and sign long-term electricity purchase agreements. That is not an AI story. That is a commodities story. The largest winners of the intelligence utility may not be OpenAI, or even the model layer, but the holders of power generation assets and the companies that build the high-voltage grid.
Let me make this concrete. A single frontier-scale query can consume a small amount of electricity, but an enterprise agent ecosystem processing millions of queries per day consumes the energy of a small town. If token usage grows exponentially, energy procurement becomes the binding constraint. The cost of intelligence will be correlated with the cost of electrons. In the long run, the market will price intelligence not by the token but by the joule behind the token.
This is where the crypto macro lens becomes useful. In a high interest rate environment, capital-intensive utility projects struggle. They require long repayment periods. They are sensitive to the cost of capital. If Altman’s utility narrative persuades investors that AI infrastructure is safe, regulated, and bond-like, then the cost of capital for datacenter and energy projects falls, and more capital flows into the upstream supply chain. That is a macro event. It will affect commodity markets, electricity markets, and even municipal bond portfolios long before it affects the price of an AI token.
The Semantic Conflation of Token
The word token is doing a tremendous amount of heavy lifting. In the crypto context, token is a security, a commodity, a collectible, or a redeemable promise. In the AI context, token is a positional fragment of text used to calculate probabilities. The original piece, published on a crypto site, chooses to use the word token without clarifying which token it means. That is a rhetorical choice. It is designed to make AI growth sound like crypto growth.
But the economics are inverted. A blockchain token is issued, distributed, and sometimes burned. It can be held in a wallet and appreciated. An AI token is consumed, erased, and gone forever. It cannot be held. It cannot appreciate. It is a running cost. The semantic overlap is a trap. The sentence token demand will grow exponentially sounds like a statement about a hot asset class. In reality, it is a statement about a growing expense line in corporate income statements.
The conflation becomes even more dangerous when crypto projects claim that AI agents will need crypto wallets to pay for AI tokens. This is possible in a narrow technical sense. A stablecoin rail may be used to settle machine-to-machine payments. But the regulated utility model that Altman describes does not naturally use an anonymous, volatile, decentralized settlement layer. OpenAI has every incentive to keep billing inside its own ledger, with central pricing, central risk, and central supervision. The utility will be the opposite of a crypto bull market. It will centralize and bureaucratize the exchange of value.
The Decoupling Thesis
Here is the contrarian thesis: the utility narrative is not bullish for crypto token markets. It is, if anything, a liquidity drain. Think about it carefully. A crypto token is a store of value, a claim, a speculation on future network adoption. An AI token is an expense. When an enterprise consumes AI tokens, it pays a fee to the model provider. That fee leaves the company treasury and enters the model provider treasury. It does not circulate in a public blockchain. It does not generate transfer fees for decentralized exchanges. It does not create yield for DeFi. It is a cost item, not a yield item.
If Altman is right, the next decade will see hundreds of billions of dollars spent on AI tokens. That is a massive absorption of global liquidity. The crypto market should not celebrate it; it should fear it. Every dollar spent on AI compute is a dollar not available for speculative asset allocation. The AI utility is a competitor for capital, not a source of it.
The decoupling thesis is simple. Crypto token prices will not rise because AI token usage rises. They will decouple. AI tokens are consumed; crypto tokens are held. Consumption is a subtraction from the balance sheet. Holding is an addition. The two are opposite in their effect on liquidity. The market’s desire to see the word token as bullish is a failure of pattern recognition. History repeats, but the code changes the rhythm. The rhythm here is billing, not building.
Volatility Is the Tax on Ignorance
Now let’s return to the investment implications. The original article invites its readers to imagine a world where intelligence is metered like electricity and where token usage grows forever. That world, if it arrives, will not be frictionless. It will be a world of price wars, regulatory battles, energy shortages, and political arguments over who gets access to the meter.
Volatility is the tax on ignorance. The ignorance here is the refusal to distinguish between a billing unit and a network asset. The next few years will punish everyone who invested in the meme of the token without understanding the accounting. A market that treats an AI token as if it were a crypto asset will be violently repriced when the CFO notices that the AI bill is not an asset. It is an expense.
The safest positioning in this cycle is not in the token myth. It is in the infrastructure of measurement. The FinOps layers, the model routing protocols, the energy traders, the chip fabricators, the regulated utility entities: these are the places where liquidity will actually flow. They are not glamorous. They do not issue tokens. But they are the structural counterparties of the metered mind.
The Ledger at the End of the Cycle
If intelligence is truly a utility, the most important ledger will not be a blockchain. It will be a cost ledger. Someone will have to account for every token, every joule, every instruction, every failure. In that world, trust will be measured by auditability, not by transparency. Liquidity evaporates when trust calcifies. The current trust in the exponential token narrative will calcify when the first major enterprise reports an AI cost overrun so large that it requires a revision of earnings.
I have been through this before. In 2020, I wrote a memo arguing that yield farming was a liquidity illusion, not a sustainable economic model. The market hated that memo until the mid-year correction proved it right. I am not predicting a correction in AI stocks. I am predicting a correction in the narrative. The token meter is real. The exponential growth may be real. But the value of that growth is not settled by the word exponential. It is settled by the price per token, the margin per token, and the economic output per token.
Beneath the baroque facade of Altman’s utility speech, the ledger bleeds. The bleed is not always visible in a tweet, but it is visible in the quarterly cost reports of every enterprise that adopts AI at scale. The macro does not whisper; it screams in silence. The silence is the gap between the predicted token boom and the absent data in the article that bears its name.
Let me end with a forward-looking question, not a summary. In five years, will OpenAI be more valuable than the electric grid that powers it? Or will the real utility trust be owned by the companies that never confused a meter with a coin? The answer will not be written in the price of a token. It will be written in the cost of a thought. As someone who has spent a career reading the shadows cast by invisible hands, I intend to read the bills.