Why falling AI costs spark investor panic but drive new revenue

This article summarizes an episode of 20VC’s video series featuring Roman Chernin, co-founder of Nebius.

Roman Chernin, co-founder of Nebius / Photo credit: Roman Chernin

Roman Chernin, co-founder and chief business officer of Nebius, an AI cloud company that builds large-scale GPU clusters and full-stack infrastructure for developers, believes the AI industry misunderstands the relationship between cost and demand.

People often assume that lowering the cost of AI will lead to less spending on servers and chips. Chernin argues the opposite: cheaper AI encourages wider adoption, increasing demand for computing power.

Why cheaper AI leads to more spending

Business leaders treat falling computing costs as a sign of a shrinking market. Chernin argues that assumption overlooks the number of business problems that remain too expensive to solve with AI today.

Many companies have thousands of internal tasks that could benefit from computers doing the work, but they are currently too expensive to change.

The concept of Jevons paradox helps explain the shift. As a resource becomes more efficient and affordable, customers do not keep the savings. They spend their budgets to use the technology on more tasks.

“Every time we get [the same unit of intelligence] cheaper, we are not reducing consumption; we are increasing consumption because we can solve more complex tasks with the same budget,” Chernin notes, adding that lower costs also make previously uneconomical AI applications viable.

Computer suppliers benefit when cheaper systems make new products possible. The danger for business leaders is acting out of fear based on the false idea that lower prices lower demand over time.

Looking at free software differently

This change alters how businesses should look at free software. Financial analysts view free AI models as a threat to the profits of large cloud companies. The real competition is moving toward specific tasks inside companies.

A new system is forming where private models test new tasks first. Once proven, companies move them to free models to save money as they grow.

The release of the DeepSeek model showed this well. The fast drop in computing costs caused panic among technology investors, sending Nebius to experience a 40% drop in its stock price during that week.

Chernin notes, “The same exact week [Nebius stock went down 40%], we probably had the best week in sales… because so many people figured out that they can run inference in their production workloads with DeepSeek and the economics will work.”

DeepSeek proved that customers want more computing power. While banks prepared for a market crash, corporate engineering teams realized they could afford the computing power needed to grow their internal tools.

Building easier software tools

Helping businesses use the technology