Last year, when the Deepseek moment hit, analysts panicked. Nebius, a key AI infrastructure provider, saw its stock plummet 40% in a week. Harry Stebbings, host of 20VC, challenged Roman Chernin, Nebius's co-founder: won't open-source AI models, cheaper and locally hosted, eat into the business of frontier providers like OpenAI and even infrastructure plays like Nebius?
Chernin’s answer flips conventional wisdom. He argues that the market's fear was unfounded. That very week, as Nebius’s stock crashed, the company logged its best commercial week ever. The core insight? Cheaper intelligence doesn't reduce consumption; it ignites it. This is Jevons Paradox in action: when a resource becomes more efficient or affordable, demand for it expands, often dramatically.
Cheaper Intelligence, Exploding Demand
Harry Stebbings articulated the common fear: a shift to open-source models, driven by cost, would harm everyone from OpenAI to Nebius. Roman Chernin disagreed, citing the fundamental principle of Jevons Paradox. He explained, “Every time we got 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 or we can finally economically viably solve the tasks that we already kind of knew that was solvable but economics didn't work and we could not scale.”
Think about it: before large language models, many complex tasks were simply too expensive or impossible to automate with AI. Now, even with proprietary models, new use cases emerge daily. When open-source models drop the barrier to entry even lower, it doesn't just cut costs on existing tasks. It makes a whole new class of tasks — those previously deemed too niche or costly — suddenly feasible. The net effect is not a redistribution of existing demand but an expansion of the total pie.
The Power of Tunable Models
The real game-changer with open-source models, Chernin explains, isn't just their price tag. It's their adaptability. “The most important kind of cause of those models is not just they open source but they are tunable they are trainable,” he said. “So you can take them and you can do something you you can postrain them and you can create the specialized model that in your particular case may work better.”
This tunability unlocks a critical lever: specialization. A general-purpose frontier model might be overkill or too expensive for a specific internal task like triaging hyper-specific customer support tickets or analyzing proprietary manufacturing data. But an open-source model, fine-tuned with your own data, becomes incredibly efficient and accurate for that narrow domain. This allows companies to build agentic applications and specialized tools that were previously economically out of reach. Each new, specialized application adds to the overall compute demand across the industry.
What to Do With This
Stop viewing open-source AI as a threat to your budget or your existing AI providers. Instead, identify niche, high-value problems within your business that were previously too expensive to solve with frontier models. Then, experiment with fine-tuning open-source models to build specialized AI agents and applications for those specific use cases this quarter. You'll likely find new opportunities for growth that were invisible before.