Cheaper AI tokens might sound like a sign that computing is becoming less valuable. But the source describes a different pattern: token prices are dropping while H100 GPU rental prices hold steady or climb. That combination suggests falling costs can encourage enough new use to keep demand for computing high.
Lower prices can expand AI use
The source connects this pattern to Jevons' paradox, where greater efficiency or lower costs can lead to more total consumption. In the AI market described, cheaper tokens may make it practical to use AI agents, automate tasks, and build new applications. Those uses can add demand even as each token costs less.
The key question is how quickly usage grows. If people and organizations use substantially more AI as prices fall, the increase in volume can outweigh the reduction in per-token cost. In that scenario, demand for the hardware needed to provide AI remains strong.
Agents add uncertainty to the picture
The source says it is unclear how much demand comes from human users and how much comes from AI systems themselves. Agentic AI can consume tokens at a high rate, so systems performing work autonomously may contribute significantly to usage.
That makes raw token demand harder to interpret. A rise in usage might reflect broader human adoption, more tasks being handed to agents, or both. The source also raises the possibility that agent-driven consumption could inflate apparent demand. Even modest growth in human use could translate into outsized hardware needs if automated systems use tokens intensively.
Hardware prices depend on continued growth
The data cited in the source comes from Ornn, Silicon Data, and Bloomberg, as of August 2026. It shows falling token prices alongside steady or rising rental prices for H100 GPUs. The source says a16z describes this as a textbook case of Jevons' paradox.
That interpretation rests on a central assumption: AI usage must grow quickly enough to offset falling prices per token. If that happens, computing capacity remains in demand, and the hardware behind AI can stay scarce and expensive. Lower prices for users would then coexist with strong prices for the resources required to serve them.
The pattern is not guaranteed to continue. Should demand flatten, the effects could spread beyond AI services. The source identifies chip makers, memory suppliers, energy providers, and cloud companies as parts of the chain that could be affected.
A market sensitive to demand signals
The source points to a market reaction as a sign of that sensitivity: US stocks dropped after reports that OpenAI's annualized revenue might be lower than previously reported. The episode illustrates how expectations about AI demand and revenue can influence investors' views of the wider ecosystem.
For now, falling token prices and firm H100 rental prices point in the direction Jevons' paradox predicts. But they do not settle how much usage is durable, who or what is generating it, or whether growth can keep pace with lower prices. The answer matters because the AI market's hardware and infrastructure depend on continued demand.