Why AI Costs Keep Rising Despite Efficiency Gains

The efficiency paradox reshaping AI economics

A counterintuitive pattern is emerging across the AI industry: as models become faster and cheaper to run, total compute consumption is accelerating rather than declining. The explanation lies in a 160-year-old economic principle that Microsoft CEO Satya Nadella has repeatedly cited to explain AI's explosive growth trajectory.

Jevons Paradox, first identified by economist William Stanley Jevons in 1865 while studying coal consumption, holds that improving resource efficiency typically increases rather than decreases total demand. Lower costs remove barriers, unlock new use cases, and ultimately drive consumption higher. Three companies at different layers of the AI stack—hardware, infrastructure, and enterprise software—are now seeing this dynamic play out in real time.

Why it matters

This pattern fundamentally changes the economics for startups and small businesses adopting AI. Rather than simply reducing costs for existing workflows, efficiency gains are making entirely new categories of AI applications economically viable. The shift could democratize access to capabilities previously reserved for well-funded enterprises, but it also means compute demand will likely continue outpacing supply for the foreseeable future.

Hardware efficiency creates new consumption patterns

Tensordyne, which builds inference racks using 90% less power than alternatives, is watching efficiency gains drive expanded usage rather than cost savings. Cofounder Gilles Backhus points to the evolution of AI workloads as evidence: early ChatGPT interactions involved single inputs and outputs, while today's agentic systems perform hundreds of simultaneous tasks requiring detailed reasoning.

"To outrun Jevons Paradox, we need to make it much cheaper to run AI workloads, without compromising on what customers value: speed and model quality," Backhus explained. The company's Napier inference system runs 13 times higher throughput than Nvidia while using less energy and space.

For startups operating on limited capital, the implications are significant. If state-of-the-art AI models drop from $5 per session to $0.50 due to hardware improvements, Backhus predicts adoption could increase tenfold or more. "It could be rocket fuel for growth," he said, noting that lower costs would democratize AI research and development by making previously uneconomical use cases testable.

Professional services unlock impossible workflows

Orbital, a real estate AI platform supporting 200,000 annual transactions across the U.S. and U.K., exemplifies how productivity gains create new possibilities rather than simply reducing hours worked. CTO Andrew Thompson notes that developers using AI aren't working fewer hours—they're producing far more output in the same time.

"The single greatest predictor of a software team's success is the rate of shipping new things to customers," Thompson said. AI supercharges that discovery loop, letting small teams deliver what once required entire departments. The result: salary growth is spiking for roles that have effectively leveraged AI tools.

In commercial real estate, firms can now review every document for every property and produce comprehensive due diligence reports—work that didn't exist before because it wasn't economically viable. "This is a new offering that law firms can sell to clients," Thompson noted.

Infrastructure providers see exponential demand

Verda, an AI infrastructure provider that grew revenue twentyfold in two years to exceed a $100 million annual run rate, attributes its growth to the reinforcing cycle between hardware improvements and model capabilities. CTO Arturs Poli describes models moving from "almost-good-enough to genuinely production-ready" as a key driver.

"Rather than thinking of rapid growth purely as a by-product of easier access to compute, I'd see it as hardware and models improving together," Poli said. High-quality open-weight models paired with more accessible compute have opened doors to lower-cost agentic workflows, removing barriers that previously favored large enterprises.

Poli expects increased usage to accelerate as businesses find more efficient and diverse ways to consume AI capabilities. The company is also addressing sustainability concerns, exploring everything from producing clean energy to using excess heat from data centers to warm homes.

These details were first reported by Alison Coleman for Forbes.