The promise software was built on

Software has always been sold on one economic promise above all others. Once the product is built, serving each additional customer costs very little, so gross margins are high and they stay high as the business scales. That promise is why software has commanded the valuations it has, and why the whole category has been easier to finance than more capital-intensive businesses.

Artificial intelligence is quietly testing that promise. For the first time in a long while, gross margin is a live question in software rather than a settled assumption. Understanding why matters for how these businesses grow, how they are valued, and how they can be funded.

Where the cost creeps in

The traditional software cost base is dominated by costs that do not scale linearly with every unit of usage. Hosting and infrastructure exist, but for a well-run product they are a small fraction of revenue, and their share of revenue often falls as the company grows.

AI-enabled features break that pattern. Every time a user invokes a model, there is a real, direct cost to serve that request. Inference is not free, and unlike traditional hosting it scales with usage rather than being spread thinly across a growing base. A product where users lean heavily on AI features can find that the cost of serving each customer rises with engagement, which is the opposite of how traditional software economics are supposed to work. The more successful the feature, the more it can cost to run.

For some businesses this pressure is modest and well managed. For others it is material enough that a margin that initially looked like classic software starts to behave more like a utility bill, rising with consumption. The category no longer has a single margin profile. It has a spectrum, and where a given company sits on it has become something you have to check rather than assume.

AI can also push margins the other way

The picture is not one-directional, and it would be a mistake to treat AI purely as a cost. For some software businesses, AI is improving gross margin rather than eroding it.

The clearest example is in the cost of delivering and supporting the product. AI is automating parts of customer support, onboarding and service that used to require people, and for businesses where those functions sat inside the cost of revenue, that automation lifts gross margin. Some companies are using AI to serve more customers with the same team, which is exactly the kind of leverage software is supposed to provide.

A second example sits earlier in the business, in how the product itself gets built. AI-assisted development is letting engineering teams ship features faster and with fewer people per feature, which can lower the cost of building and maintaining the product that ultimately feeds into the product’s gross margin. A smaller, more productive engineering team is not a line-item customers ever see, but can have a meaningful effect on operating costs and overall profitability.

A third, increasingly common example is direct monetisation. Rather than absorbing AI costs into an existing subscription, a growing number of software vendors, from large incumbents to smaller SaaS providers, now charge separately for AI usage, whether through consumption-based credits or a priced add-on tier. Done well, this does not just offset the cost of running the feature. It turns AI from a margin risk into a new revenue line in its own right.

Taken together with the cost pressure described earlier, the same technology is compressing margins in one place and expanding them in another. The determining factor is where AI sits in the business, and how deliberately a company manages that position. Left unpriced and unmanaged, AI usage presses on margin. Automated, monetised or used to build the product more efficiently, it can improve the economics of the business. Two companies can both describe themselves as AI-enabled and have gross margins moving in opposite directions.

A worked example

Picture a software product that has not yet repriced for AI, with a traditional gross margin of 85 percent. The company adds an AI feature that customers use heavily. Each active user now generates a real inference cost every month that did not exist before. If that cost amounts to, e.g., ten percent of the revenue generated by those users, the gross margin on the AI-heavy portion of the base falls from 85 percent to 75, and can fall further if the usage increases without a correstponding increase in pricing. 

The uncomfortable part is that the more successful the feature, the larger the potential drag. Traditional software generally becomes cheaper to serve as it scales. AI-enabled software can become more expensive to serve as usage grows, unless the company actively manages the economics.

The pricing and packaging response

The companies handling this well are not simply absorbing the cost. They are repricing and repackaging so that the customers who consume more AI also contribute more revenue.

That can mean usage-based elements in the pricing, higher tiers for AI-heavy features, or fair-use limits that prevent consumption from running too far ahead of revenue. Done thoughtfully, this restores the link between what a customer pays and what they cost to serve, which is what protects gross margin as adoption grows. Done carelessly, or not at all, the margin quietly erodes while the top line still looks healthy, and the problem only becomes visible when the cash does not stretch as far as the revenue implied.

Why this matters for growth and financing

Gross margin is not an accounting detail. It is the share of revenue left after the direct cost of delivering the product, before operating expenses and financing costs. When margin is high and stable, revenue is a reliable engine. When margin is uncertain or declining, the revenue is worth less than its headline, because more of each euro is consumed before it can do anything useful.

This has a direct effect on how much financing a business can support. Any capital repaid from revenue depends on the durability of the margin underneath that revenue. A business with strong, stable gross margins can comfortably support growth capital, because the cash is genuinely there. A business whose margins are quietly eroding under rising cost-to-serve is a different proposition, even if its top-line growth looks identical, because the revenue is less able to carry an obligation.

It also affects valuation and the Rule of 40. Rising AI costs can pull gross margins down even when the underlying investment is sound. If those costs also weigh on profitability, they can drag down a company’s Rule of 40 score in ways that do not always reflect the long-term health of the business. The companies that come through this well are the ones that understand their own cost-to-serve precisely and can show where their margin is heading, not just where it sits today.

Underwriting margin durability

The practical response, for founders and financiers alike, is to treat gross margin as something to assess over time rather than take at face value.

That means separating the cost of serving AI features from the rest of the cost base and watching how it moves with usage. It means asking whether AI in the business is a cost centre or an efficiency lever and being honest about the answer. It means projecting where gross margin is likely to land over the next two to three years under realistic assumptions about adoption and model costs, rather than assuming today’s margin holds.

The businesses thinking hard about this are building durable economics and will be better positioned to access financing. The ones hoping the question does not apply to them are accumulating a risk that will surface later. At Round2 Capital, gross margin and its direction are central to how we assess a software business, precisely because our financing is repaid from the revenue that ultimately supports repayment. Understanding your own cost-to-serve is one of the most valuable things you can do before raising any capital that depends on it.