On physical AI

Exactly as adding a layer of AI to your very outdated software suite won’t make it more appealing to the public, using AI as a selling point for your physical product won’t work.

This post was written without the reassuring supervision of my friend Chris. Be forgiving if the style isn’t as sharp as always.

In this understanding of tech dominance, the driver of generational change is a paradigm shift: from mainframes to personal computers, from desktop applications to the web, first on personal computers, and then on mobile. Each shift brought a new company to dominance, and when the next shift arrives, so will new companies rise to prominence

The end of the beginning

Many in VC are claiming AI has killed software, and that’s why it’s only worth investing in hardware today.

I don’t agree with this claim.

I believe AI is a paradigm shift for the world of bits: the way we interact with software will change because of AI, and so all the infrastructure of the web will have to change accordingly.

Let’s take the internet as an example. Human attention is the main value driver on the web - catching and selling human attention through ads is the only business model that has worked at scale, and it’s what has allowed Google and Meta to become what they are today.

All the big paradigm shifts Ben Thompson mentions in the quote above - “from mainframes to personal computers, from desktop applications to the web, first on personal computers, and then on mobile” - had to do with the way we interact with software (and the time we spend doing so).

This shift will create disruption and will force business models to change: some won’t work anymore, and some new ones will emerge.

If we start interacting with the digital world via other means that are not a screen, what happens to Apple’s app ecosystem?

If humans don’t browse websites anymore, delegating that to agents, what happens to Google’s ad revenue?

I think the concept of agentic web is fascinating, and there is a lot of value to be created. In software, certain technological innovations become paradigm shifts not because they are hard to copy, but because they enable new business models and market dynamics that compound and help create stickiness - i.e., create a moat

Many in VC are also claiming that AI is a paradigm shift for the physical world.

I don’t agree with this claim either.

In the physical world, AI is an enabler - it has the potential to make everything more efficient, more intelligent, more useful, but generally doesn’t change the incentives to buy or use a product.

For this reason, I believe it is an error to treat physical AI like a paradigm shift.

If we apply AI to existing markets, using it as a differentiating factor, we are implicitly taking a market risk: we are assuming that something that didn’t sell before AI, will start selling when we add AI.

There are some edge cases in which this might work, maybe if we are able to build a humanoid robot that does chores for us someone will want it, but I hate market risk, so I don’t buy into this vision.

Does it mean we shouldn’t be bullish on AI applied to the physical world?

Absolutely not! We should be, and I’ll explain why.

Widespread adoption of transformer-based AI models (what everyone calls “AI” these days) is a productivity shift that has come at the same time as a real paradigm shift for the physical world: the end of the Pax Americana with the geopolitical order it created, and the beginning of a transition period that has brought extreme geopolitical uncertainty.

The latest chapter in the US-China cold war is the export ban on rare earth metals. That China has a near monopoly in this is by now very clear, but why is that?

It’s not only a matter of raw materials; in fact, China imports many of the raw materials that it then processes. It’s a matter of technology, know-how, and willingness to industrialize the combination of the two. It’s not a case that technology and machinery are on the export ban list..

This has enabled China to offer a product that buyers need at a price that makes it the most wanted on the market. In this case, the product is also a commodity, and in a normal state of the world price is all that matters for commodities - easily explained the monopoly.

Will AI change this? Will transformers make rare earth metals refined in Italy, Texas, Ohio, or Germany more appealing to the market? No, unless their cost suddenly matches the willingness to pay of the customers. This brings us back to the main point: all of a sudden, certain buyers have a higher willingness to pay, and we can use the latest technologies (including AI), to supply a product that matches that demand.

This geopolitical shift changes incentives in the biggest industries out there, providing a new market and customers that have strong intentions to buy - the famous urgent buyers. This will make some business models outdated, creating disruption in otherwise crystallized industries, and allowing new entrants to undermine incumbents’ domination in these sectors.

New markets are way stronger moat than new technologies for young startups. As the world evolves and needs change, it’s easier for startups to tackle these changing needs than it is for large incumbents:

Why do incumbents immediately copy new technology when they see it is working but don’t immediately enter new markets when they see they are working? […] The answer is that to adapt they have to feel comfortable with how uncertain the new technology or new market is. Incumbents strongly dislike uncertainty so they wait for it to be mitigated. But startups can build moats in new markets while they are still uncertain where they usually can’t with new technologies.

https://reactionwheel.net/2020/11/productive-uncertainty.html

I think physical AI has the potential to change the world, but we shouldn't fall into the trap of throwing money at anything that has some intelligence in it - that would burn money and momentum, and we need both of them right now!