The World Runs on Semiconductors.
Own the Picks and Shovels.

The Picks-and-Shovels Principle

In the California Gold Rush of 1849, the most reliable fortunes were made not by the prospectors but by the men who sold them pickaxes, shovels, and denim trousers. Levi Strauss became wealthy. Most gold miners did not.

The same logic applies to the AI revolution. Predicting which company will produce the dominant large language model in 2030 is genuinely difficult — even the people building these systems cannot tell you with confidence. But every possible winning scenario requires the same underlying infrastructure: semiconductors, data centres, storage, power, and increasingly, quantum computing capabilities. Own the infrastructure, and you win regardless of who wins the model race.

This is not a new insight — infrastructure investing is as old as capitalism. What's new is the scale and speed of demand growth. AI data centres require orders of magnitude more computing power per workload than traditional cloud computing. A single large AI training run can consume more electricity than a small city for weeks. The buildout of infrastructure to support AI is the largest capital expenditure cycle in the history of the technology industry.

Layer 1: The Chips

NVIDIA is the obvious name, and it is obvious for a reason — their CUDA software ecosystem creates a moat that is more durable than the chips themselves. The AI industry has been trained on CUDA for fifteen years. Every researcher knows it, every model is optimised for it, every data scientist has built their workflow around it. Switching to a different chip architecture means retraining the humans, not just the models. That is an extremely sticky competitive advantage.

But ASML is the more interesting structural monopoly. ASML makes the lithography machines that all advanced chip fabs use to manufacture semiconductors — including NVIDIA's. There is literally no alternative to ASML for the most advanced chips. Their extreme ultraviolet (EUV) machines cost approximately €200 million each and there is a multi-year waiting list. ASML has a monopoly on the machines that make the machines that run AI. If you had to hold one semiconductor company for the next decade and could only look at it once a year, the case for ASML is very strong.

Beyond these two, the memory and storage supply chain (SK Hynix, Micron, Samsung) is experiencing its own AI-driven demand surge as large language models require massive amounts of high-bandwidth memory to operate efficiently.

Layer 2: The Real Estate (Data Centres)

Data centres are the physical homes of the cloud — and AI is making them build faster than the commercial real estate industry can keep up. Power availability is the binding constraint: connecting a large new data centre to the grid now requires years of permitting, infrastructure buildout, and negotiation with utilities. This creates a durable competitive advantage for companies that already have land, power, and connectivity in the right places.

Equinix and Digital Realty are the two dominant data centre REITs. They own and operate the facilities where the internet literally lives — the interconnection hubs where internet service providers, cloud companies, and enterprises meet. Their real estate is arguably more strategically important than any office building in any city.

Both companies operate an asset-heavy model (they own the buildings and power infrastructure) but have REIT-like dividend characteristics combined with technology growth rates. The AI buildout is structurally positive for both — their existing facilities are in exactly the places (major metropolitan interconnection hubs) where new AI workloads need to run.

Layer 3: The Raw Materials

Every semiconductor requires rare earth elements. Every battery requires lithium, cobalt, and nickel. Every magnet in an electric motor requires neodymium. The physical inputs to the digital revolution are mined from the ground, and the supply chains for many of them run through a remarkably small number of countries — most of which have been reducing Western access to these materials.

MP Materials is the only significant rare earth mining and processing company in the United States. Their Mountain Pass facility in California is the primary Western source of the materials needed for permanent magnets — which go into electric vehicles, wind turbines, military hardware, and robotics. MP Materials is not a pure-play AI company, but it is the closest thing to a strategic raw materials monopoly in the Western hemisphere.

The geopolitical dimension of this trade is substantial. Trade restrictions on advanced chips and rare earth materials have become a primary tool of US-China competition. Owning the Western rare earth supply chain has optionality value that goes well beyond normal equity considerations.

Layer 4: The Next Frontier (Quantum)

Quantum computing is not today's trade. It is next decade's trade, and the way to own it today is carefully and in small position sizes. IonQ is the most credible pure-play public quantum computing company — they are building on trapped-ion technology, which has better error rates than competing approaches and genuine potential to achieve commercial utility in the 5–7 year timeframe.

The risk here is real: quantum computing has been "5 years away" for 30 years. But if you believe it eventually arrives, the moment of first commercial utility will be a step-function change in valuations for the companies who got there. A small, patient allocation makes sense as an asymmetric option.