Tokens Are Only 10 to 25 Percent of AI Spend. Price the Rest Before You Pitch

Tokens make up only about 10 to 25 percent of total AI spend, according to Forrester principal analyst Kevin Ogunsua, who argues that total cost of AI (TCA) may replace total cost of ownership as the planning lens. Infrastructure, energy, capital, software, licensing and labor account for the rest. He reported this from the October 8 Linux Foundation Analyst Virtual Summit, citing the State of Tokenomics 2026 study of 472 respondents representing $4.6 trillion in revenue: 96 percent use frontier models, and 86 percent are evaluating or using model routers.

Sellers should care because finance gains a way to price what you sell. Ogunsua says FOCUS, the FinOps Open Cost and Usage Specification, is emerging as the data layer that maps AI costs to business units, products and the general ledger. A business case that stops at seat price can be rebuilt by the buyer’s finance team with numbers you did not supply.

The vendor side shows the pressure. In GitLab’s Q2 FY27 results, subscription cost of revenue rose 76% against 21% revenue growth, and AI inference costs compressed gross margins by roughly 400 basis points. GitLab’s pricing response was Flex, a consumption model that drew over $20 million in commitments from more than 130 customers within six weeks of launch.

The stronger deal puts the TCA on the table before the buyer asks: what the customer will spend on tokens, what they will spend on integration, data work and people, which business outcome pays for it, and who owns that metric. Name the baseline, and name the cost controls, such as model routing.

Ogunsua’s point is that cost justification must extend beyond license fees and token prices to full-stack economics and measurable outcomes. If your AI-in-sales pitch only covers the license line, you are leaving the harder half of the buyer’s math to someone else.