Cheaper AI tokens are driving more demand, Jensen Huang's best-case scenario
The Decoder argues that the AI compute market is behaving like a textbook Jevons paradox, the idea that making a resource cheaper to use increases total consumption of it. The piece opens with the line "Jevons' paradox is Jensen's best friend" and frames the trend as the best-case scenario for Jensen Huang.
The evidence is data from Ornn, Silicon Data and Bloomberg, as of August 2026. According to the article, it shows that token prices keep dropping while H100 GPU rental prices hold steady or climb. The article attributes the "textbook Jevons paradox" label to a16z.
The mechanism, as the article describes it: cheaper tokens unlock AI agents, automation and new applications, so total volume grows faster than per-unit costs fall. The result is that falling prices do not ease pressure on hardware.
The article flags a caveat about where the demand comes from. How much of it comes from humans versus the systems themselves is not clear, since agentic AI burns through tokens at a staggering rate. Compute demand could therefore be artificially inflated, and even modest human usage growth could trigger outsized hardware needs.
The whole system, the piece says, rests on one assumption: AI usage has to grow fast enough to offset falling token prices. As long as it does, hardware stays scarce and expensive. If demand flattens, the chain from chip makers and memory suppliers to energy providers and cloud companies takes a hit, and markets could spiral from there. The article points to how sensitive markets already are: US stocks dropped on reports that OpenAI's annualized revenue might be lower than previously reported.
Key facts
- Data from Ornn, Silicon Data and Bloomberg (as of August 2026) shows token prices falling while H100 GPU rental prices hold steady or climb.
- a16z calls the pattern a textbook Jevons paradox: cheaper tokens unlock agents, automation and new applications, so volume grows faster than per-unit costs fall.
- It is unclear how much demand comes from humans versus the systems themselves, since agentic AI burns through tokens at a staggering rate.
- The system rests on one assumption: usage must grow fast enough to offset falling token prices; if demand flattens, chip makers, memory suppliers, energy providers and cloud companies take a hit.
- Markets are already sensitive: US stocks dropped on reports that OpenAI's annualized revenue might be lower than previously reported.
Why it matters
Cheaper tokens might be expected to make compute cheaper and easier to get. The article says the opposite is happening at the hardware layer: token prices keep dropping, yet H100 rental prices hold steady or climb, because volume grows faster than per-unit costs fall. That is the Jevons paradox applied to AI, and it explains why the demand for chips, memory, energy and cloud capacity stays strong even as the price of using a model falls. It is also the reason the piece frames the trend as the best-case scenario for Jensen Huang.
Who it affects
The article names the whole chain: chip makers and memory suppliers, energy providers and cloud companies. They benefit while demand keeps growing and take a hit if it flattens. Markets are affected too, as the reaction to the reports about OpenAI's annualized revenue showed. Anyone who buys tokens or rents H100 capacity is on the other side of the same trend, with cheaper tokens but no relief in GPU rental prices.
How to use it
This is an analysis rather than a product, so there is nothing to adopt. The practical reading is the article's own test: watch whether AI usage keeps growing fast enough to offset falling token prices. As long as it does, hardware stays scarce and expensive. A flattening of demand is the signal that the chain described above would come under pressure.
How solid is it
It is a short analysis piece. The data comes from Ornn, Silicon Data and Bloomberg as of August 2026, and the Jevons paradox label is attributed to a16z. The article gives no actual token prices, H100 rental prices or percentage changes, so the direction of the trend is stated but not measured in the text. It also says openly that the split between human and machine-generated demand is unclear. The statements about what happens if demand flattens are conditional, not predictions.
Risks and caveats
The article's central risk is that compute demand could be artificially inflated by agentic systems burning through tokens, so that even modest human usage growth could trigger outsized hardware needs. The whole structure depends on usage growing fast enough to offset falling token prices; if demand flattens, the supply chain takes a hit and markets could spiral from there. The OpenAI revenue reports it cites are described only as reports that revenue might be lower than previously reported, and the article does not give a figure.
“The whole system rests on one assumption: AI usage has to grow fast enough to offset falling token prices.”
— The Decoder