Satya Nadella’s Counterintuitive Take on DeepSeek

  • Post author:
  • Post category:
    guide

Satya Nadella, the Gandalf of cloud software, tweeted about the Jevons paradox to frame DeepSeek’s recent mic-drop moment. DeepThink R3’s performance has been surprising considering how little it cost to train and run. From the frenzy, it would seem NVIDIA and OpenAI’s dominance is in danger. Or is it?

The more shocking the news, the stronger the desire for certainty, however misplaced it may be.

https://twitter.com/satyanadella/status/1883753899255046301

DeepSeek’s progress is not a zero-sum game. It doesn’t mean OpenAI, NVIDIA, and Silicon Valley are done for, nor that everyone at these companies should pack up, cash in their options while they still can, and go fishing.

Jevons and Braess’s Paradoxes: When More Is Less

Satya said, “Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of.”

In that sentence, he unpacked all the conditions necessary for the Jevons paradox to be observed.

The Jevons paradox (1865) is the observation that the more efficient a system becomes at consuming a resource, the overall consumption of the resource counterintuitively increases. The efficiency increases the problem that it tries to solve. The core conditions that need to be true are:

  1. Efficiency improvements must be substantial. 
  2. This efficiency should lower costs, making the resource more accessible
  3. Increased affordability will increase usage
  4. This growth fuels the economic conditions, further driving demand

A good example of this paradox is MPG efficiency in cars. While vehicles have become more efficient, the consumption of gas over time has only increased. 

Braess’s paradox, while more recent (1968), also documents a similar counterintuitive observation. Addressing networks, it states that adding more network capacity or a new route only makes the network slower. 

The 405 in Southern California is a good example. It is the local freeway that runs from north Los Angeles to south Orange County. For over two decades, I’ve watched Braess’s paradox play out in real time. No matter how many lanes are added, rush-hour congestion remains unchanged. Just like in the Jevons paradox, population growth ensures the 405 stays a parking lot during rush hour, no matter how many lanes are added.

Is it all About Compute?

The efficiency improvements of DeepSeek V3 are eye-popping. Across the board, it looks like DeepSeek trains faster, requires fewer tokens, and is cheaper to run. 

Metric Comparison: DeepSeek V3 vs. Llama 3.1

These numbers challenge the prevailing Silicon Valley mindset of throwing money at nuclear-powered data centers and compute, while wooing Jensen Huang over sushi to get more GPUs. Then the industry saw these numbers and had a moment of panic, wondering if it had overspent on NVIDIA and OpenAI.

Satya disagrees.

We Are Selfish Automatons

What I think Satya was saying (we haven’t had the chance to personally chat) is that this breakthrough will expand AI usage to wider audiences and become cheaper, and that will be good for all and for the simple reason of the “selfishness” of the user. 

Both the Jevons and Braess’s paradoxes assume that participants in these systems—economic for Jevons, network for Braess—act independently and in their own self-interest.

Just like everyone wants their own car instead of carpooling, everyone wants their own access to GenAI. Efficiency gains like these will drive down costs and make AI more accessible, which will only increase usage. In the long term, everyone wins.

At least, that’s what I think Satya is saying.

And Finally, What About China?

If you’ve read Kai-Fu Lee (review), this advancement shouldn’t surprise you. Geopolitical considerations aside, this is technology moving forward thanks to constraints-driven creativity.

A small group of researchers worked within compute and cost restrictions and came up with a better model, leading to spectacular improvements. Silicon Valley quickly realized that it didn’t have the market cornered on go-to-market innovation and velocity.

That’s a good thing. 

To quote Charlie Munger, the grand wizard of all Gandalfs, competition has a way of preserving value.