Everyone Is Building AI Apps. The Real Money Is in Infrastructure

Everyone is building AI apps right now. But when I looked at our technology spending at Roar Global over the past year, something interesting stood out. We are not spending more on SaaS tools. We are spending more on infrastructure.

Supabase, Vercel, Railway, Claude APIs, Gemini APIs, Cloudflare, AWS, Redis, Sentry, Stripe. The tools that let us build. And that bill keeps going up.

This reminded me of a pattern that shows up in almost every major technology boom.

During the California Gold Rush, thousands of people rushed west hoping to strike gold. Most of them didn’t. The people who made consistent money sold the equipment. Picks, shovels, tools. Levi Strauss didn’t mine gold. He sold jeans to the miners.

The same pattern appeared during the dot-com era. Everyone was building websites, but the companies that captured the most value were the ones powering the internet itself. Cisco sold the networking equipment. Sun Microsystems built the servers. The infrastructure layer captured more value than many of the applications built on top of it.

AI is following the same structure.

While hundreds of startups are racing to build AI products, companies like Nvidia are supplying the chips that power all of them. They don’t need to pick winners. They sell to everyone. Cloud providers play a similar role. AWS, Azure and others win because every team building AI needs compute, storage and deployment infrastructure.

When technology makes it easier to build applications, more people build applications. That naturally increases demand for the layer underneath.

Infrastructure wins by volume.

There is one important catch though. Infrastructure markets tend to consolidate quickly. They are expensive to build and expensive to maintain. Switching costs are high. Once a company builds deeply on a particular stack, moving away becomes slow and painful. As a result, the value often concentrates among a small number of companies that become defaults for the ecosystem.

For anyone building software today, this creates a simple reality. If your product sits purely at the application layer with no proprietary data, no deep workflow integration and no meaningful switching costs, you are effectively building on land owned by someone else.

The AI gold rush is happening. Most people are digging for gold. The companies selling the shovels are doing very well.