The gold rush metaphor is overused, but it keeps coming back for a reason. In AI, the shovels keep changing hands - and we're watching it happen in real time.
The pattern isn't new. Infrastructure companies often win early (Cisco during the internet boom), but long-term profits go to those offering tools, not pipes (Microsoft, Amazon).
AI has already been through multiple iterations of where the profit is believed to lie. The first phase was OpenAI's ChatGPT, which quickly grew a previously smaller company into a household name. Despite its popularity, it remains unprofitable - now reportedly looking into integrating advertisements, which Altman had previously called a last resort.
The second phase is happening now. All the major players are racing to secure electricity, data center capacity, and the infrastructure to power them. Companies like Oracle are raising massive debt ($45-50B) to build data centers - often for the same AI companies that then lease capacity back to them as revenue. It's circular financing dressed up as growth.
So the AI industry has been through two eras in its understanding of where the money will come from. First, that service providers like OpenAI would profit hugely. Then, that it would be companies building infrastructure. Neither is generating profit margins that justify the capital intensity.
We're entering a third era. My bet is that the winners won't be the companies building models or data centers. They'll be the vertical AI agents charging per outcome - like how Fin.ai is charging a fixed price per resolved support ticket - and the companies solving the physical bottlenecks around energy, cooling, and grid capacity.
The shovel sellers have shifted again. This time, they're selling outcomes, not access.