Key Takeaways

  • The chatter about an AI infrastructure bubble misses the point: the market is just beginning, with Nebius co-founder Roman Chernin expecting “tens or hundreds times more” compute to be needed.
  • Cheaper AI intelligence doesn't reduce consumption; it increases it. Chernin points to Jevons' Paradox and the 'DeepSeek moment,' where lower unit costs unlock entirely new, more complex tasks.
  • Despite market jitters about falling AI prices, Nebius experienced its "best commercial week" during periods of concern, signaling robust, real-world demand.
  • While compute demand is elastic, it's not limitless. Chernin emphasizes that customer product economics ultimately dictate the viability and scale of AI adoption, setting a thoughtful boundary for pricing.

The Counterintuitive Logic of AI Compute Demand

Talk to enough founders and investors, and you'll hear the whisper: “Is AI compute a bubble?” Roman Chernin, co-founder of AI infrastructure giant Nebius, doesn't just dismiss the idea; he flips it on its head. For Chernin, the industry is barely out of the starting blocks. He sees a future where we'll need “tens or hundreds times more” compute than today, suggesting we're just at the beginning of an “amazing moment.”

The core of Chernin’s argument against the bubble theory is beautifully counterintuitive: cheaper AI doesn’t mean less demand; it means more. He cites Jevons' Paradox, an economic principle that states increasing the efficiency with which a resource is used leads to an increase, not a decrease, in its consumption. In AI, this translates to something Chernin calls the 'DeepSeek moment.' When a unit of intelligence gets cheaper, we don't scale back our usage. Instead, we find new, more complex problems to throw at it. “Every time we got intelligence cheaper, the same unit of intelligence cheaper, we are not reducing the consumption but we increasing the consumption because we can just solve more complex tasks with the same budget,” Chernin explains. This isn't just theory; it's what happens when new possibilities open up.

Why Lower Prices Mean Higher Consumption

The fear of a bubble often stems from the idea that as AI models become more efficient and compute costs drop, demand for infrastructure will fall. But Chernin's real-world experience at Nebius tells a different story. He recalls a period when “people on the market were concerned that market is going down and like infrastructure companies like Nebius not needed because okay if AI is such so much cheaper maybe it's bubble.” Yet, during that exact period of market anxiety, Nebius logged its "best commercial week." This isn't a fluke; it's a testament to the latent demand that cheaper, more accessible AI unlocks.

This isn't to say prices are infinitely flexible. Chernin acknowledges that compute demand is elastic “for some extent.” The price point must still make economic sense for customers. They need to integrate AI into their products profitably. This means providers like Nebius have to be “meaningful and thoughtful kind of what our customers need.” It’s about finding the sweet spot where lower costs spur innovation without torpedoing the customer's business model. The real growth isn't just from existing applications getting cheaper; it’s from entirely new use cases becoming viable as the cost barrier falls. This dynamic ensures that infrastructure companies like Nebius continue to be essential players, fueling a growing ecosystem rather than shrinking within a contained bubble.

What to Do With This

If you're building an AI-driven product, challenge any internal models that assume a linear relationship between compute cost reduction and your overall revenue or market size. Instead, project your total addressable market based on the new use cases that become economically feasible as compute prices fall. Model not just lower unit costs, but the massive expansion of demand those lower costs will inevitably unleash.