Software is cheaper than it has been in a decade — but not uniformly

Why seats are vulnerable

A seat-based model assumes value scales with headcount. If a single operator with an AI assistant can do the work of three, headcount growth decelerates — and so does the revenue line. This is not a 2026 problem. It is a 2028 problem that the market is pricing today.

Usage-based models invert that: revenue scales with consumption, which AI increases. The trade-off is predictability. Buyers dislike variable bills, which is why hybrid pricing — a platform fee plus consumption — is emerging as the consensus structure.

The businesses with a consumption meter inside them are the ones that benefit from AI adoption rather than being cannibalised by it.

Where to look

• Infrastructure and platform names where AI workloads already appear as metered consumption.
• Data and distribution owners: proprietary data plus a distribution channel is the durable combination.
• Avoid pure seat-expansion stories where AI is a feature rather than a pricing mechanism.

The valuation trap

Cheap software is often cheap for a reason, and the forward estimates that make it look cheap are exactly the ones that get revised down first. We prefer businesses trading at a modest premium to the group where estimate revisions are positive. Revision direction leads multiple direction; the reverse is rarely true.

Our screen requires three things simultaneously: positive 90-day estimate revisions, expanding gross margin, and a trend score above 60. That filter removes most of the value traps and most of the momentum traps at the same time.