AI Is Breaking the Zero Marginal Cost Assumption in Payments and Fintech

For a long time, the economic model of software was reassuringly simple. You built the product, absorbed the upfront cost, and then scale did the rest of the work for you. Each additional user improved your margin almost by default.

That model is now being quietly dismantled.

AI does not replace it entirely, but it introduces something that had largely disappeared from modern software economics, which is meaningful marginal cost at the point of use.

You can see this most clearly in how AI is actually priced. Both OpenAI and Anthropic ultimately charge for usage in tokens. Every prompt, every response, every intermediate step carries a cost. It may be fractions of a cent, but it is real, and it scales directly with activity.

At the surface, this is easy to miss. Subscription products such as ChatGPT or Claude smooth this out. You pay a monthly fee, you get access, and the cost feels contained. But that is a packaging decision, not an economic one. Underneath, the system is still metering usage, and someone is still paying for it.

The moment you move into APIs or embed these models into your own product, the abstraction disappears. Cost becomes visible, and more importantly, it becomes sensitive to design.

A slightly longer prompt, a more verbose response, an extra model call in a workflow. Each of these feels like a small product decision. Taken together, they define your cost base.

The return of marginal cost, in disguise

What makes this shift easy to underestimate is that it does not arrive as a line item labelled “AI cost explosion”. It arrives incrementally, feature by feature.

A fraud model here. A customer support assistant there. A decisioning layer that calls out to a model before approving a transaction.

Individually, these look like improvements. Collectively, they start to behave like a variable cost engine sitting inside your product.

For payments and fintech businesses, this is slightly uncomfortable territory. These are sectors that have spent years optimising around infrastructure that behaves in a broadly fixed way. Once the platform is in place, volume is something to be encouraged, not something to be feared.

AI complicates that assumption.

If each incremental interaction carries cost, then growth is no longer purely accretive to margin. In some cases, cost can scale in line with revenue. In less well-controlled cases, it can scale faster.

What developer tools are already revealing

You can see a more concentrated version of this dynamic in developer-facing tools.

Anyone who has spent time with Claude Code from Anthropic will recognise the pattern. It is an extraordinarily capable tool, particularly because it ingests large volumes of code context in order to be useful. But that capability has a cost profile.

It is entirely possible to burn through a session allowance in minutes.

That is not a flaw in the product. It is the economic model. The tool is doing exactly what you want it to do, which is to read, reason, and respond across a large context. The cost follows naturally from that behaviour.

Translate that into a customer-facing product and the implication becomes clearer. If a single internal user can generate that level of cost through normal usage, then a scaled user base interacting with AI-powered features can create a cost curve that is both steep and, if unmanaged, unpredictable.

Product decisions are now financial decisions

This is where the role of finance starts to shift.

Historically, finance could afford to sit slightly at a distance from product decisions. The connection between a feature choice and the cost base was indirect and often delayed. Infrastructure was pooled, costs were amortised, and the marginal impact of a single decision was small.

AI removes that distance.

The choice of model, the structure of a prompt, the number of calls in a workflow, even the tone of a response all have cost implications. These are not technical details in the background. They are economic levers.

Which means that someone needs to be able to answer questions like:

• What does this feature cost per use?

• How does that change with scale?

• At what point does margin begin to compress?

• Where can we trade accuracy or latency for cost?

These are not purely engineering questions, and they are not purely financial ones either. They sit in the space between.

The emergence of embedded strategic finance

This is why a different kind of finance role is starting to appear, particularly in companies that are building or heavily adopting AI.

Not the traditional reporting or control function, and not strategy in the abstract, but something more embedded. A role that sits close to product and engineering, and understands the mechanics well enough to translate them into economic outcomes.

In practical terms, that means:

• Working with product teams to model the cost impact of new features before they are built

• Understanding how usage patterns translate into token consumption and spend

• Helping define guardrails so that growth does not quietly erode margin

• Designing pricing models that reflect the underlying cost structure rather than ignoring it

The companies that are getting ahead of this are not necessarily the ones with the most advanced AI capabilities. They are the ones that have recognised that AI is not just a capability layer, but an economic one.

A shift that will not stay contained

It is tempting to see this as an issue for AI-native companies and assume it will remain there.

It will not.

The same dynamics are already moving into payments, fintech, and any sector that is starting to embed AI into core workflows. The shift is gradual enough to be missed in the short term, but structural enough to matter in the long term.

Software is not returning to a world of heavy marginal cost, but it is moving away from a model where marginal cost was low enough, and stable enough, to be largely ignored.

It is somewhere in between now, and that middle ground is less forgiving.

The companies that recognise this early will design for it. The ones that do not will discover it through their margins.