What remains scarce when intelligence becomes abundant?

Artificial intelligence is becoming both more capable and dramatically cheaper. According to a new Deutsche Bank report, on a like-for-like basis, the cost of AI capabilities has almost halved every quarter over the past three years.

If this trend continues, it could have profound consequences for the economy. Cognitive work that is expensive today because it requires highly skilled people, such as software development, engineering, scientific research and financial analysis, could become much cheaper.

This raises an important investment question: what happens to the rest of the economy if intelligence becomes cheap and abundant? What would remain scarce in such a world?

The Deutsche Bank report highlights several categories: proprietary data, regulatory permission to operate, sticky customer relationships and physical assets such as land and machinery.

That points to investment opportunities well beyond the technology sector:

1. Physical assets may become relatively more valuable. AI can design a power plant, but it cannot instantly build one. It can optimise a transport network, but it cannot create land, transmission lines, airports or roads out of thin air. Permits, electricity, raw materials and construction capacity could therefore become increasingly important bottlenecks.
2. Regulatory barriers may also become more valuable. A highly capable AI system cannot automatically acquire a banking licence, obtain approval for a new medicine or receive the right to operate regulated infrastructure. Sectors such as banking, regulated utilities and pharmaceuticals could therefore be relatively insulated from disruption because existing licences, contracts and accumulated infrastructure create barriers to entry.

There are thus two ways for investors to benefit from AI: 1) investing in intelligence creation itself and 2) owning scarce assets in a world where intelligence is cheap.

The first approach has dominated markets in recent years. Investors have focused on semiconductors, data centres, cloud providers and power-related infrastructure.

The second has received much less attention.

Yet it could favour some very traditional parts of the economy: energy, property, financial institutions, transportation infrastructure and pharmaceutical companies.

This is quite relevant to the way I am currently positioned.

My portfolio has little exposure to the companies producing AI itself. Instead, a significant part is invested in financials, infrastructure, property and energy: sectors where value is often tied to licences, physical assets, land, networks or other forms of scarcity that AI cannot easily reproduce.

These positions were not selected primarily because of an AI thesis. They are the result of fundamental and macroeconomic considerations. But it is interesting that many of them could also benefit from a world in which intelligence becomes abundant while physical assets and regulatory licences remain scarce.