DeepSeek’s energy efficiency counterintuitively bolsters AI energy companies’ outlook

We’ve all read about DeepSeek and the remarkable competitive edge it presents to existing mega-cap AI companies. It's more efficient, costs less to operate, and has decent output.

However, the energy efficiency that DeepSeek achieves evidences that AI can operate on less energy than the dominant AI players are running on. Thus, the market is taking an axe to AI energy providers.

However, this energy efficiency may have unintended consequences encapsulated in a 160-year-old economics paradox, Jevons Paradox, which has come to light again in AI circles. It theorises that efforts to reduce the demand of a commodity, say energy, through energy efficiency initiatives may actually increase the use of it because the resultant lower prices for the commodity spur significantly higher demand.

DeepSeek's performance is in line with the latest OpenAI results (see chart below), but it is less energy intensive and apparently significantly cheaper. Therefore, it's cheaper, opening the door to more applications being taken up in the everyday world. This increase in penetration will create a net positive demand and actually increase AI’s total energy consumption.

When computers and mobile phones were first invented and trialled, people thought they would be used by a fraction of the actual Total Addressable Market (TAM), which is why their unit prices in the beginning were so high. Then, because pricing came down, the TAM opened dramatically. You had prices come down X and then another X, but volumes go up 10X, taking up the total net demand of the market.

I’m not too sure how this plays out in the semiconductor chips space (e.g. Nvidia) in terms of their pricing and margins; they will sell a lot more chips/units, but their margins are likely to come down with an unknown result on profit growth vs current estimates, in my opinion. But with energy and probably the volume in computing (e.g. data centres), I believe the effect of DeepSeek on AI energy and infrastructure companies will be a net positive, the roll out of which I believe DeepSeek has likely reaccelerated more than, say, the chip makers and the designers of these.