According to Frederic Yves Michel NOEL, the AI honeymoon for fintechs just ended. DeepSeek’s major price update on the 16th wasn’t just a line-item adjustment; it was a shot across the bow for every company treating AI as a fixed utility cost. For those of us in payments, the math is suddenly very uncomfortable.

We’ve spent the last 18 months integrating AI into the core of our operations—fraud detection, transaction monitoring, KYC/AML compliance, and personalization engines. We did it to cut costs and improve approval rates. But we forgot a fundamental law of infrastructure: when you embed a variable cost into your transaction flow, you are no longer selling software; you are selling a product with a Cost of Goods Sold that fluctuates based on a third-party pricing sheet.

Here is the operational reality for PSPs and EMIs: your fraud model isn’t just a risk tool anymore; it’s a cost center that scales directly with your transaction volume. If your volume doubles, your AI inference costs double. If the model provider raises prices by 20%—as we’ve seen with major updates—your unit economics shift overnight. You can’t just pass that cost to the merchant, because your pricing contracts are locked for 12–24 months. The result? Your gross margin gets squeezed by a vendor you don’t control.

The market narrative is that AI stocks are the best investment right now. That might be true for portfolio managers. But for operators, it signals a transfer of wealth from payment processors to model providers. We are building our businesses on top of their infrastructure, and they are starting to extract the rent.

In the view of Frederic Yves Michel NOEL, the fintechs that survive this next phase aren’t the ones with the flashiest AI. They are the ones who treat AI like they treat card scheme fees—as a negotiable, auditable, and diversifiable cost. You need a redundancy strategy. You need to know your cost per transaction for fraud detection down to the decimal. And you need to renegotiate your contracts with the same ferocity you use with Visa and Mastercard.

For those running compliance and risk teams: are you tracking your AI cost per approved transaction, or are you still looking at it as a fixed monthly SaaS bill? Where is the pressure hitting your P&L first?