The best AI stocks may be the toll collectors

If your thesis is that AI spending keeps expanding over the next 10 years, the best investments may not be the flashiest consumer AI products.

They may be the companies with the deepest moats in the infrastructure stack.

That starts with ASML. If advanced chips need to be made, ASML sits at the most difficult bottleneck in semiconductor production. Its tools are extraordinarily hard to replace, which makes it one of the strongest strategic positions in global tech.

Next is TSMC. Even if multiple chip designers win, they still need world-class manufacturing. TSMC’s scale, process leadership, and trust with top customers make it one of the clearest toll collectors in AI.

NVIDIA remains central because its moat is broader than hardware. CUDA, networking, software tooling, and developer adoption make it the default compute platform for much of the AI ecosystem.

Microsoft deserves a place near the top because enterprise AI adoption runs through distribution. Azure, Microsoft 365, GitHub, and its enterprise relationships give it a durable advantage as AI moves from experimentation to standard workflow.

Broadcom is another important beneficiary. Custom AI chips and networking are becoming more critical as hyperscalers optimize their own infrastructure. That plays directly into Broadcom’s strengths.

Amazon rounds out the list. AWS, Bedrock, and Trainium give it meaningful exposure to the buildout, even if the stock is less pure than some suppliers.

The main lesson is simple: when a platform shift triggers huge capex, the biggest winners are often the companies with the strongest bottlenecks.

In AI, that is where the moats are.