Finding · The 2026 State of B2B Software Monetization
How do B2B software companies handle the cost of AI inference?
According to Nalpeiron’s 2026 State of B2B Software Monetization study (n=243 B2B software companies with AI features, fielded June 2026), 44% of companies absorb the cost of AI inference entirely into existing pricing, and only 13% fully pass those costs through to customers.
Key numbers
| Response | Share of companies with AI features |
|---|---|
| Absorbing the cost entirely (built into existing pricing) | 44% |
| Partially passing it through (credit / token pricing) | 42% |
| Fully passing it through to customers | 13% |
| Have not formally addressed this yet | 1% |
| Not applicable to our business | 0% |
Question Q20. n = 243 (companies with AI features).
What this means
AI inference is a direct, usage-linked cost of goods sold: every model call has a price. How companies recover that cost is the clearest read on whether AI features are monetized or subsidized. In this study, 44% of companies absorb the cost entirely into existing pricing, 42% partially pass it through with credit or token pricing, and only 13% fully pass it through.
This is a monetization gap, not a measurement gap. The same study finds that most companies can already meter AI usage, and 79% can separate AI-agent usage from human usage (n=243). The capability to charge exists; the pricing decision to charge lags behind it.
Absorbing inference cost compresses margin quietly and scales with adoption. As customers use AI features more, the unrecovered cost grows, which is the mechanism behind AI revenue leakage. The size of that gap is company-specific and can be estimated directly rather than assumed.
Methodology
The 2026 State of B2B Software Monetization is based on n=255 product and engineering leaders at B2B software companies, screened for direct involvement in pricing, packaging, or monetization decisions (manager-level and above; 86% director-level or higher). It was fielded in June 2026 through an industry research panel; figures reflect 255 completes as of 19 June 2026. Two AI-inference questions use a reduced base of n=243.
This finding uses a reduced base of 243 responses, the companies to whom the question applied (survey Q20).
See the full methodology and every question on the report hub.
Read the full benchmark
This finding is one cut from The 2026 State of B2B Software Monetization. Get the complete report, all findings, and the full data appendix.
Read the full report