After DeepSeek, AI developers are wrong that The Jevons Paradox will bail them out

In 1865 British economist William Stanley Jevons explained to the public that increased efficiencies in the use of resources per unit of production do not generally lead to lower consumption of those resources. Rather, these efficiencies lead to higher consumption as many more people can now afford the more efficiently produced goods which carry a lower price tag. Jevons was referring to coal, the cost of which was falling and demand for which was rising due to increased efficiencies in production. His idea became known as The Jevons Paradox.

When the Chinese-based artificial intelligence (AI) upstart DeepSeek demonstrated last week that complex and powerful AI can be delivered for a tiny fraction of the cost and resources of current AI tools, DeepSeek’s competitors cited The Jevons Paradox and told investors not to worry. Demand for AI would now grow even more rapidly in response to greater efficiencies and thus lower costs.

What those competitors failed to mention is that DeepSeek’s breakthrough is great news for buyers of AI tools, but very bad news for current developers who are sellers of those tools. DeepSeek is giving away free or at only 3 percent of competitors’ prices (for those needing application programming interface services) something comparable to the very expensive products of its competitors. This suggests that the hundreds of billions of dollars spent developing those expensive tools may have just gone up in smoke. That investment may never be recouped.

Moreover, DeepSeek has shown that its powerful AI tool can run on a laptop, so the need for vast cloud computing resources is not necessary in many cases. In addition, DeepSeek’s AI tool is open source and can be freely distributed. This means anyone can see the code, customize it, perhaps improve upon it AND make money off the improved or customized version. And, because anyone can see the code, anyone can see how DeepSeek achieved such efficiencies and design their own AI tool to match or exceed those efficiencies.

The one thing the big AI developers are right about is that at these new prices (free or nearly free) the demand for AI is likely to grow much more rapidly as it is applied to situations where AI was previously too expensive to justify—just as The Jevons Paradox suggests. And that means it is probably wrong to think that these vast new efficiencies will eliminate the need for large expansions of electric generating capacity. The demand for additional generating capacity will still be there. It may just rise at a slower rate than previously forecast.

This is NOT an endorsement of what is about to happen. In fact, the more rapid spread and even wider use of AI is likely to create problems at a faster rate. More efficient and broader use of AI means that the human sources of information will be driven from the marketplace even sooner—the very ones that are essential if AI is to have real information from informed experts and writers. What comes next is AI feeding on AI-generated information, a kind of digital cannibalism that will not end well.

As I wrote back in September:

It’s worth noting that expertise does not actually reside on the page. It resides in the minds of a community of interacting experts who are constantly debating and renewing their expertise by evaluating new information, insights and data from experiments and real-world situations.

When the information generated by this kind of expertise is gone from the web or at least crippled, what kind of nonsense will AI tools spew out then? One thing is almost certain: The nonsense will now come more quickly and from more and more of the systems we rely on. That’s hardly a comforting thought.