The Token Tax: How AI Economics Are Splitting Software Into Winners and Renters

The Token Economy Has Two Sides

Per-token AI inference costs are dropping. That sounds like a tailwind for software companies. It isn't, uniformly.

As model costs fall, enterprises aren't pocketing the difference. They're running more agents. Agentic workloads expand to fill the budget. Net AI spend stays flat or climbs. The cost efficiency gain disappears at scale.

That's the token tax. And it's splitting software companies into two camps: those who control the billing layer, and those who absorb the cost.

How Consumption Billing Changes the Math

Salesforce (CRM) and Adobe (ADBE) moved early on this. Both are transitioning away from pure seat-based licensing toward consumption-based or hybrid billing models.

The mechanic is simple. When a customer uses more AI, they pay more. Token costs pass through. Gross margin holds.

Under the old model, a vendor sold 1,000 seats and owned whatever AI infrastructure cost followed. Under consumption billing, AI usage becomes a revenue line, not a cost line. That's a structurally different business.

Gross Margin Is the Signal

For SaaS, gross margin above 70% is healthy. Below 60%, you start asking questions.

A company absorbing token costs at scale without a pass-through mechanism is running a structural risk. It doesn't show up in one quarter. It shows up when agentic usage compounds over 2-3 earnings cycles and the cost base has quietly expanded.

Consumption-based billing provides a floor. The margin doesn't compress as AI usage scales because revenue scales with it.

Adobe's Model vs. the Renter Class

Adobe (ADBE) has been explicit. Firefly usage generates incremental revenue. More AI creation means more billing. The model scales with the customer's consumption.

Companies still on flat-fee AI licensing took the opposite bet. They pre-committed to token costs at a fixed price point. If usage spikes, the margin compresses. Revenue doesn't move.

That's owning the downside and renting the upside.

The divide isn't about which companies use AI. It's about which companies get paid when their customers use it.

What This Means for Traders

• Gross margin trends over the next 2-3 quarters are the tell for SaaS names with deep AI integration. Compression signals a billing model problem, not just a cost problem.
• CRM and ADBE have consumption-based buffer built into their revenue architecture. Pure seat-license SaaS players pricing AI as a feature add-on don't.
• ChartOdds margin tracking across SaaS earnings shows which names are holding the line and which are quietly compressing. Run the screen before the next cycle.