Everyone's debating whether AI will kill SaaS. They're asking the wrong question.

There is a popular narrative out right now that generative AI will make traditional SaaS irrelevant. The argument goes that companies will stop buying software and simply generate whatever they need with tools like Claude or ChatGPT.

I think that view is overblown. Enterprise software is sticky, complex systems cannot be replaced with a few prompts, and most businesses will still prefer reliable products over bespoke scripts. Lastly, incumbents are not sitting still. Every SaaS product under the sun is scrambling to add AI functionality.

The real disruption may be different than what everyone thinks. AI is not going to kill SaaS demand but it might break SaaS economics.

As companies embed AI into their products, they inherit a fundamentally different cost structure. That shift will reshape gross margins, pricing models, and what it takes to build a profitable software company.

The assumption that made SaaS work

For more than a decade, SaaS has been built on one powerful economic assumption:

Once the platform is built, adding another customer costs almost nothing.

High fixed costs, minimal variable costs. That dynamic produced 70 to 90 percent gross margins, with best in class companies often above 80 percent. It is also what made models like freemium viable, where a small percentage of paying customers subsidized a much larger free user base.

When Shopify or Salesforce signs a new customer, their infrastructure costs barely move. Revenue scales much faster than costs, which is why subscription software compounds so effectively over time.

Why GenAI changes this

GenAI native products do not behave this way. Their costs rise with usage, not just headcount.

Many AI first companies operate closer to 50 to 60 percent gross margins, not 80 percent. They still pay for cloud compute, storage, and databases like any other SaaS company, but they also pay for every token, inference call, and GPU cycle consumed by their users.

In Frugal’s case, we spend more on AI than on traditional cloud costs. That is not unusual for modern AI products.

From a CFO’s perspective, this flips the classic SaaS flywheel. Growth no longer automatically improves profitability. In some cases, rapid adoption can actually make the business worse if heavy users drive costs faster than revenue.

A single enterprise customer running large scale AI workloads can be far more expensive than a dozen light users. That is a very different risk profile than traditional SaaS.

What this means for pricing

Because costs are variable, pricing has to change. Many AI companies have already moved away from pure seat based models toward one of three patterns:

1. Seat plus usage pricing
2. Pure consumption based pricing
3. Hard caps on AI features with paid upgrades

None of these restore classic SaaS margins on their own, but they do protect the downside and align revenue more closely with cost.

Freemium, in its traditional form, is likely broken for most GenAI products. Free trials or limited credits make far more sense than unlimited free usage.

Can efficiency save the day?

Most SaaS companies were not profitable with 80% margins. What happens when you make this 20% harder?

It is possible that AI will reduce R&D, sales, and marketing costs enough to offset lower gross margins. If that happens, some companies may still reach strong profitability even with 60 percent margins.

But that is not guaranteed. Teams should plan as if margins will remain structurally lower, not assume AI magically fixes everything below the line.

It’s also likely that inference costs will continue to drop. Maybe this mitigates some of the pain but there are also reasons to doubt. The pattern in the upgrade cycle has been to keep pricing the same for the latest model and discount the older one. If you can get by with the old one then there’s savings but many will want to chase the latest functionality and keep paying.

Why FinOps and cost engineering matter more than ever

FinOps was already growing before the AI wave. Now it is mission critical. Variable AI costs make uncontrolled growth dangerous, even for fast growing companies.

Engineers have not traditionally focused on costs. In most cases they don’t know how their code changes impact costs. That is no longer optional. Every design decision, API call, and model choice now has a direct impact on the P&L. Moving forward, cost engineering will be as important as quality, security, performance and other non-functional requirements.

Bottom line

Classic SaaS was built on the idea that growth fixes everything. GenAI breaks that assumption.

Growth at any cost made sense when margins were 85 percent and every incremental customer was almost pure profit. At 55 percent margins with variable compute costs, that playbook can destroy you.

In the AI era, profitable growth requires deliberate cost engineering, not blind scale.

For a technical deep dive, see post on cost engineering, by Frugal CTO, Craig.