I trace the shadow before it casts. Altman’s latest prediction—that intelligence, measured in tokens, will grow exponentially and become a public utility—landed in the crypto press with the weight of a founding myth. The article itself was thin: four bullet points, no data, no timestamp. The real story isn’t the prediction. It’s the silence around the cost curve that makes it hold.
Context: The Utility Narrative and Its Hidden Anchors
Altman’s framing is elegant. Token usage, he claims, will follow an exponential curve, and intelligence will become as ubiquitous as electricity. For anyone who has watched the DeFi summer unfold, the pattern is familiar: a narrative of infinite demand paired with a business model that charges per unit. OpenAI’s API has always been a metered service. Calling it a utility isn’t prediction—it’s branding. The article, published on Crypto Briefing, reinforces the semantic overlap between AI tokens and crypto tokens, blurring the line between a unit of computation and a speculative asset.
But the article’s core inference—that “new consumption and cost management strategies” are needed—is the only honest admission in the piece. It acknowledges that exponential token usage, without a corresponding drop in unit cost, is not a utility. It’s a cost explosion waiting to happen.
Core: The Cost Curve That Defines the Narrative
I’ve spent years auditing DeFi protocols where exponential growth assumptions hid structural fragility. The same pattern appears here. Token consumption in LLMs is tied directly to inference compute. Every token burned requires a measurable amount of electricity, GPU cycles, and memory bandwidth. To call token usage exponential is to call for an exponential increase in compute infrastructure. That’s not a forecast—it’s a supply chain demand.
Logic blooms where silence meets code. The silence in Altman’s statement is the absence of a cost-per-token trajectory. Between 2022 and 2024, OpenAI dropped prices on GPT-3.5 and GPT-4 by roughly 50% to 80% per token, depending on the model. That’s impressive, but it’s not a sustained exponential decline. The history of semiconductors (Moore’s Law) and electricity (Swanson’s Law) shows that true utility requires a cost decline of at least 10x per decade. For AI tokens to behave like a utility, the cost per token must fall faster than the growth in usage. Otherwise, the total cost of intelligence becomes a burden, not a liberation.
Finding the pulse in the static. The static here is the unspoken assumption that token growth is value growth. In my work modeling tokenomics for DeFi projects, I learned to separate usage from value. A token used for low-quality SEO spam or repetitive agent tasks inflates the count without creating proportional economic output. The article’s call for “cost management strategies” implies that enterprises are already seeing AI bills rise faster than ROI. That’s the real signal: the market is already building tools to cap token consumption, not just to accelerate it.
I’ve seen this before. In 2020, during the Curve Finance deep dive, I simulated 10,000 attacks on the stable swap invariant. The flaw wasn’t in the math—it was in the assumption that liquidity would always flow in and out at the same rate. The same assumption plagues the utility narrative: that token usage will grow without bound, and that the cost will somehow take care of itself. The missing piece is a unit economic model. Without knowing the marginal cost per token and the elasticity of demand, the exponential claim is a hope, not a thesis.
Contrarian: The Blind Spot of Exponential Cost
Here’s the counter-intuitive angle: if Altman is right about exponential token usage, he may be wrong about it being a utility. A utility is defined by affordability and reliability. If token costs do not decline at a rate that outpaces usage growth, intelligence becomes a luxury good consumed by the wealthy or by institutions with deep pockets. The article’s inference about cost management strategies is, in itself, an admission that the cost is a problem. The blind spot is the assumption that “more usage” is always better. In my experience auditing protocol designs, the most dangerous narratives are the ones that ignore the denominator.
Consider the security implications. If intelligence becomes a public utility, the failure modes scale. A single hallucination in a utility-grade AI could trigger cascading errors in healthcare, finance, or logistics. The current model of API reliability—with outages, rate limits, and prompt injection vulnerabilities—is far from the five-nines standard of electricity or water. The article avoids this entirely. The ethical dimension is also missing: if token usage is a measure of economic access, then exponential growth without redistributive mechanisms deepens inequality. Altman’s interest in UBI aligns with this, but the article doesn’t connect the dots.
Takeaway: The Unanswered Question
Vulnerability is just a question unasked. The question Altman’s prediction leaves hanging is simple: what is the cost per unit of intelligence, and how fast will it fall? Without that data, the narrative is a beautiful shadow—but shadows are cast by objects. The real object is the infrastructure of compute, energy, and carbon that must support exponential growth. The market will eventually demand a cost curve, and when it does, the projects that build the cost management layer—the “AI FinOps”—will be the ones that profit. The ones that bet on infinite growth without cost discipline will find themselves in the same position as Terra’s LUNA: a beautiful model that collapsed under its own weight.
In the void, the bytes whisper truth. The truth is that intelligence as a utility is possible, but only if the cost decline is as steep as the usage growth. The crypto ecosystem should watch this space—not for the token price, but for the cost curve. That’s where the real security lies.
