AI cannot replace SaaS under current cost structures

By Alan Jacobson, AI Economics Strategist

The hare is fast, but every step burns energy.
The tortoise is steady, and keeps moving forward.

The problem isn’t speed. It’s cost per step.

The core assumption behind SaaS is simple:
build once, sell many times, and margins expand as revenue scales.

AI breaks that assumption.

AI introduces a variable cost at the unit level — every prompt, every inference, every interaction consumes compute. That means cost scales with usage, not just revenue.

This isn’t theoretical. It’s already visible in reported financials.

The evidence is already in the numbers

Across a cohort of public cloud and AI-exposed companies, cost of revenue is growing faster than revenue — compressing margins instead of expanding them.

Cloudflare:

Revenue growth: +30.7%
Cost of revenue growth: +52.5%
Spread: +2,179 basis points
Gross margin: declining

Cloudflare is seeing strong demand from AI-driven workloads, with management explicitly pointing to increased infrastructure intensity and network cost tied to AI traffic.

This is the opposite of SaaS operating leverage.
Costs are accelerating faster than revenue.

C3.ai:

Revenue growth: –29.1%
Cost of revenue growth: +18.4%
Spread: +4,750 basis points
Gross margin: sharply declining

In this case, the divergence is even more extreme:
costs are rising while revenue is falling.

Even allowing for restructuring effects, the core signal is clear —
cost to serve is not decoupled from usage.

This is a structural difference, not a temporary phase

SaaS economics depend on decoupling cost from usage:

the software is built once
incremental users add minimal cost
margins expand over time

AI does the opposite:

every interaction incurs compute cost
higher usage increases cost directly
margins compress unless pricing rises faster than compute

That creates a fundamental constraint:

If cost per inference does not fall faster than revenue per inference rises, AI cannot achieve SaaS-like margins.

The market is still pricing AI like software

Narratives around AI are dominated by:

model capability
benchmarks
adoption
revenue growth

But the underlying business is still subject to a basic requirement:

it must generate profit.

Right now, the data shows:

cost of revenue rising faster than revenue
gross margins declining
no clear evidence of operating leverage

This is not a model problem.
This is a business model problem.

The implication

AI will absolutely generate revenue.
It will reshape products, workflows, and industries.

But under current cost structures:

AI cannot replace SaaS economics.

Not because the technology isn’t powerful —
but because the unit economics don’t scale the same way.

The bottom line

This is already showing up in public filings.

Across multiple companies:

cost scales with usage
margins compress before revenue catches up
operating leverage breaks

The companies that win AI won’t be the ones with the best models.

They will be the ones that solve one problem: cost per inference.

– Published Friday, March 20, 2026

Where are SaaS assumptions breaking under AI workload economics?

Send signal: signal@revenuemodel.ai

I read every signal.