This month's inflation numbers gave the SaaS pricing conversation a fresh data point. Software prices rose at a record pace in June, up more than 17% year over year, according to Bureau of Labor Statistics figures covered by the Washington Post and confirmed independently by CNN and other outlets tracking the same release.

Economists point to the AI build-out — memory and chip scarcity pushing up component costs, alongside genuinely new AI capability layered onto existing products.

Let's call it SaaS-flation — subscription costs climbing faster than most customers can point to a matching increase in value received.

This pattern has played out across the industry for a while now. Some price increases attach an AI label to what is really margin recovery, unrelated to any AI feature any given customer actually uses. An analysis of recent SaaS price hikes by Lynton Web found that AI has become a legitimate-sounding reason for a price increase, and enough customers go along with it that the practice keeps paying off for vendors who lean on it.

Many SaaS providers are, themselves, being squeezed by AI economics. Running inference at scale is genuinely expensive, and vendors that built AI features into their platforms are absorbing infrastructure costs that didn't exist three years ago. That's a real reason so many vendors, from spend-management players like Coupa to much of the broader market, are shifting away from flat per-seat subscriptions — rendered obsolete due to AI taking over some employees' responsibilities — toward consumption or outcome-based pricing tied to agent activity. Often, that shift is the only way the economics of embedding frontier models into a product actually work.

The industry's ongoing tokenomics debate captures why this is harder than it looks. Token consumption doesn't scale predictably with headcount the way seats did. An agent that needs three retries and a long chain of self-checks to complete a task ends up costing more than one that finishes cleanly, even though the customer receives the identical outcome. Under old seat pricing, that inefficiency was quietly the vendor's problem to absorb. Under consumption pricing, it becomes a line item the customer sees, and has to explain to their own Finance team.

Any vendor in this conversation needs to make sure a price increase reads as value delivered, not value leveraged, at a moment when it's genuinely harder for buyers to tell one platform's offering apart from a competitor's.

It's worth considering exactly what's funded by a price increase. If it's frontier-model access, real agent capability, and a measurable improvement in outcomes, that's a case worth making directly to customers, in plain language, ideally backed by usage data showing where the value landed. If it's funding margin recovery dressed up as an AI line item, customers tend to find out, and the retention cost ends up higher than the incremental revenue was worth.

There's a second, quieter pressure worth naming. The layer that used to be safely considered 'SaaS' is now genuinely contestable, in ways that change what differentiation even means. If a customer increasingly reaches a vendor's capability through someone else's AI platform — like how earlier this month, Claude became available through Google Workspace — the vendor may keep the subscription revenue but loses a real amount of control over the experience the customer actually has. Something similar happens if a separate provider builds a harmonized context layer that stitches real-time data across applications for AI agents to draw on. Either development means the vendor's own interface, the thing it spent years building, stops being the primary place value gets delivered.

What holds up against both pressures? Compliance depth and vertical specialization are the obvious answers, and they show up in nearly every current analysis of where SaaS retains real 'moats', since regulated industries need a vendor that owns the audit trail, not a general model bolted onto a spreadsheet. Backing this up with a more nuanced read than the SaaSpocalypse headlines suggest, diginomica's own network pulse research finds that CIOs view AI as a mixed picture across the enterprise stack rather than a wholesale replacement of existing platforms.

Two more moats exist that are worth inserting into this argument. End-to-end process ownership matters more than point-solution depth — a platform that owns a full business process, start to finish, is harder to disintermediate than a narrow tool, which increasingly looks like laying the groundwork for an agent rather than something a customer needs to buy on its own. Plus, a healthy ecosystem — where partners and customers are actively building extensions, integrations, and agents on top of a platform — is its own form of lock-in because it represents value the vendor didn't have to build alone and a switching cost competitors can't easily replicate.

Enterprise customers don't have to wait for vendors to lead this shift. By applying a few concrete steps during renewal and vendor evaluation cycles, they can shift leverage back to the buying side of the table.

1. Evaluate vendor AI ecosystems — look at how many partners and customers are actively building agents, integrations, and extensions on top of a platform, not just what the vendor itself has announced on its roadmap.
2. Ask vendors directly about frontier-model pricing exposure — which models they run, how they manage or rightsize token consumption on the back end, and whether customers get any say in which model handles a given workload.
3. Get clarity on consumption and outcome-based pricing before signing — how usage is metered, what counts as an outcome, and what recourse exists if actual consumption diverges from projections.

Watch the inputs, not the excuses. Memory and chip prices easing as new fabrication capacity comes online, inference costs continuing their historical decline per unit of intelligence delivered, and genuine competitive pressure from vendors willing to compete on transparent, outcome-tied pricing rather than opaque bundling — all of that would need to align before the broader trend slows.

The vendors that come out ahead of this moment won't be the ones that raise prices fastest. They'll be the ones that can demonstrate to their customers exactly what that increase bought them.