Cheaper AI tokens are driving more demand, and that’s Jensen Huang’s best-case scenario

The business move

Recent data from a16z reveals a classic economic quirk playing out in the AI market. Token prices, the cost per unit of AI compute on popular models, continue to fall. Meanwhile, rental prices for Nvidia’s high-end H100 GPUs either remain stable or increase. Demand for AI compute is growing faster than the cost improvements from chip advances or cloud efficiency.

Why it matters

This is an example of the Jevons paradox: making AI compute cheaper per token does not lower total spending. Instead, users consume more tokens, driving overall demand and keeping GPU rental prices from dropping. For Nvidia and cloud providers, this means the rising tide of AI adoption supports chip and infrastructure pricing power. If token costs drop, AI workloads scale up, absorbing more compute capacity.

However, if demand were to flatten, the chain from chip makers to cloud vendors would face shrinking margins. This creates a high-stakes balancing act—AI token prices must fall to attract broader usage but not so fast that demand plateaus. Nvidia CEO Jensen Huang benefits from rising compute volume rather than price-wars, so this dynamic matches his best-case scenario of growing, not shrinking, GPU utilization even as token costs fall.

Who gains and who gets squeezed

Nvidia and cloud providers gain pricing leverage from sustained GPU demand despite token price drops. On the flip side, AI developers and enterprises face rising infrastructure costs unless they significantly optimize token use. Smaller builders relying on cloud GPU rentals might find their cost advantages limited by steady or rising hardware prices.

Investors watching AI compute economics should track whether demand growth keeps pace with or outstrips falling token prices. A stall in demand would squeeze the chip-to-cloud chain, pressuring margins in a hardware-centric AI economy.

What to watch next

Monitor GPU rental pricing closely as new AI models and workloads emerge. Token pricing trends will drive how much compute users consume, but hardware capacity and pricing will set a floor on AI infrastructure costs. Look for shifts in demand elasticity—if AI token economics encourage bulk usage without proportional infrastructure growth, the market could face capacity or cost bottlenecks. Nvidia’s pricing moves and cloud vendors’ capacity expansions will be leading indicators of where AI’s compute economics head next.

AI Quick Briefs Editorial Desk