What I'm Seeing in AI Startup Funding

I still read funding flows out of habit, a leftover from years of raising money myself. That instinct has made the AI wave more interesting to watch than anything has been in a long time. What I am seeing is a familiar pattern at a scale I have not seen before. The money is pouring in, and it is pouring into a few places while running right past others.

Where It's Crowding

The capital is concentrating where it always does when a platform shift hits, in the layer everyone can see. Enormous sums are going into the foundational models and the infrastructure under them, which makes sense given how much those cost to build. What I find more telling is the flood into the thin layer on top, the steady stream of applications that are, underneath, a light wrapper around someone else's model.

I recognize this pattern because I have lived inside versions of it. When a new capability arrives, money rushes toward the most obvious expression of it, and you get dozens of companies chasing the same handful of ideas. Most of them are betting that being early and well funded is enough. From experience, it usually is not.

What's Getting Starved

The flip side is what the money runs past. The unglamorous, hard, specific work tends to go underfunded, the kind that requires deep understanding of a particular industry, real integration into how work actually happens, and the boring reliability and safety layers that no demo ever shows. It is hard to fund because it is slow, it does not photograph well in a pitch, and it does not ride the hype.

That is usually where the durable businesses turn out to be. The pattern I trust is that the lasting companies solve a hard, specific problem deeply, while the crowded ones solve an easy, general problem shallowly. The money does not always know the difference at the time. It tends to figure it out later.

From the Buyer's Side

What is new for me is that I now see this from the other side too. Sitting where products get pitched to a bank, I can tell quickly which AI offerings were built by people who understand the regulated reality and which are a clever model with a banking logo bolted on. The market is funding a lot of the second kind.

The gap is obvious from the buyer's seat. A bank does not need another general assistant. It needs tools that understand its constraints, its data sensitivities, and the fact that being confidently wrong is not an option in much of what it does. The products that get my attention are the ones whose builders clearly grasp that, and those are not always the ones with the biggest rounds behind them.

What Lasts

Funding froth always sorts itself out. It did in every cycle I lived through, and it will in this one. The companies that last will be the ones that solved something hard and specific, not the ones that raised the most on the thinnest idea. From the founding side that was always my bet, and from the buyer's side it is now the thing I am actually looking for. Both vantages point the same direction, which makes me trust it more.