My question to you today is if DeepSeek R1 means models will get rapidly better and cheaper, will that spike adoption and lead to more GPU sales for Nvidia, or not?

Are more huge and lavish AI Infrastructure datacenter campuses necessary or the energy projects that they will require? Are Trillions needed or a U.S. capital oversight? Does the AI stock bubble need to keep going higher?

The relationship between Nvidia, DeepSeek, and the Jevons Paradox is a fascinating intersection of technology, economics, and market dynamics.

The "Jevons Paradox", named after the economist William Stanley Jevons, posits that as technological advancements make the use of a resource more efficient, the overall consumption of that resource may actually increase rather than decrease. This paradox is particularly relevant in the context of AI and computing resources.

So will more efficient models and Open-source availability of them to use locally really require that much more compute? Please share your opinions and back it up with data if you can.

DeepSeek, a Chinese AI startup, has recently made headlines for its ability to achieve significant computational efficiency for less cost and likely less bleeding edge hardware.

At its core, the Jevons Paradox highlights a counterintuitive outcome: improvements in efficiency can lead to greater overall consumption. But is that even true in the Generative AI use case?

ChatGPT has a lot of paid subscribers, but consumers are quick to change their behaviors. Why would they use a $200 dollar a month model if cheaper and better models will soon be open-source?

Is more "consumption" really going to require more huge Datacenters for the likes of Meta, Microsoft, Google, OpenAI, xAI and friends? Perhaps we are over-estimating the importance of AI as some force of the future.

Morgan Stanley and Satya Nadella are using the term now a lot the past week. Jevons Paradox is typically used in "energy" e.g. historical example: His example was coal: After the Watt steam engine was introduced — which burned much less coal than its predecessors — demand for coal rose substantially, as Watt's engines were built into a wide range of applications.

The truth is we don't know how long transformer based LLMs will be around, a better architecture could be just around the corner that is more compute efficient. Efficiency in LLMs and SLMs themselves could require a lot less compute and power than is being projected into the future.

Morgan Stanley energy analysts concurred in a note last Tuesday, saying "there are multiple indications" the Jevons Paradox applies to the cost of compute in AI applications. But is that even true?

Nvidia's stock is only down 13% in 2025, even with the DeepSeek going viral and raising to the top of many app download charts. Wall Street doesn't sound very concerned.

Jevons Paradox is one of the main arguments they are making.

"The cost of compute has been one of the primary constraints in the development of AI. If it comes down, then it is possible that many more companies and research departments will get into the space."

Who Wins the AI Agent Battle?

20+ best free AI tools | Zapier