A question runs through many of the conversations we have had this year about artificial intelligence companies. If the underlying model is available to everyone, where does the lasting value sit?

For a while, access to the best model looked like advantage enough. That looks less certain today. According to the Stanford AI Index (2025), the cost of running these systems has fallen a long way, to a small fraction of what it was two years ago, and the free and open versions have crept closer to the paid leaders. When a capability can be rented by the month, it feels less like a differentiator and more like a shared input.

Not everyone reads it this way, and there is a reasonable case that the leading models will keep pulling far enough ahead to matter. Even so, a good part of the market is moving toward the same view. Companies described as AI agents raised more than $1bn in the first half of 2026, roughly double a year earlier, and a portion of that money goes to products that are only a thin layer over someone else's technology.

If the model is becoming shared ground, the advantages that last may turn out to be the familiar ones, which is data that is hard for others to gather or copy. A place in a customer's daily routine that is awkward to unpick. A reputation for handling sensitive work with care. None of these sit inside the model, and each tends to take years rather than months to build.

The founders who hold our attention are generally less focused on the model they have chosen than on what they are steadily building around it.

#ArtificialIntelligence #CorporateFinance #MergersAndAcquisitions