TechCrunch reported on November 6, 2026 that Ramp data show AI token spending grew just 1.5% across August and September 2026, versus 42% month-over-month growth in July, even as token volumes kept rising. The divergence puts a basic assumption under pressure: heavier business AI use does not automatically produce equivalent growth in model-company revenue.
The most recent Ramp AI Index shows an October rebound, but not one that closes that gap. Total token spending rose 8.9% in the week ending October 23, 2026 from the comparable week ending September 25, 2026, while token volume climbed 37.1%. Consumption is therefore expanding much faster than expenditure, even after spending returned to growth.
Falling Prices and Concentrated Demand
Between those two weeks, the blended price per million tokens across OpenAI and Anthropic fell from $0.76 to $0.56. The index lists that as a 26.8% decline, alongside drops of 42.4% for OpenAI and 15.7% for Anthropic. Those reductions help explain why a considerable increase in usage can generate a much smaller increase in spending.
Spending is also concentrated enough for a small customer cohort to shape the broader trend. Ramp’s October 14, 2026 analysis said 1% of firms drove 80% of spending on OpenAI and Anthropic and identified two consecutive months of virtually flat token expenditure. In September, median monthly AI spending per employee was $9,568 within the top 1% of firms paying for AI, compared with $813 in the top 10% and $15.70 for the median firm.
Paid adoption is nevertheless broadening: the index reports that it reached 55.9% of Ramp businesses in September, with 43.8% paying Anthropic and 40.6% paying OpenAI. Anthropic’s share of token spending rose from 51.3% in the week ending September 25 to 60.5% in the week ending October 23, while OpenAI’s moved from 44.2% to 35.1%. A provider can gain share within a spending pool shaped by customer efficiency gains, so share movement alone does not resolve the question of overall spending growth.
Large Customers Are Controlling Costs
Ramp lead economist Ara Kharazian told TechCrunch that the slowdown is not a demand problem. He pointed instead to model price reductions and competition that have created a buyer’s market. Advanced customers are also becoming more disciplined: setting defaults that keep expensive models away from long-running tasks, imposing employee budgets, and using routers to move work from standard or light models to cheaper light models.
Kharazian’s assessment is that growth among median and top-decile firms is insufficient to offset weaker spending by the biggest users. He questioned whether medical offices or car dealerships will approach the top cohort’s spending intensity, even with demand for coding agents in software and finance. For buyers, the result is more AI use for less money; for model companies, it complicates the relationship between consumption and revenue.
Dataset Boundaries and the Next Growth Test
The index draws on more than 85,000 American businesses and billions of dollars in corporate spending through Ramp’s corporate-card and bill-pay platform. It identifies adoption through observed paid transactions rather than surveys. Ramp says this approach likely understates activity because free tools generate no payment and employees may use personal accounts. The token metrics cover a narrower subset using Ramp Token Spend Management with connected provider usage and billing data. That subset is not representative of Ramp’s total AI spend volume, although its results are directionally similar.
The concentration figures also carry a qualification. Ramp’s October analysis says the top 1% is a small group whose spending estimates are more volatile than those for the median and top 10%, and that it revised estimates upward after adding new AI vendors. That makes the exact size of the top cohort less certain, but it does not alter the separate token-volume and token-spend measurements behind the central divergence.
Kharazian said future products could drive another phase of growth, but the commercially advanced products available now show spending slowing. He is skeptical that personal agents will match coding agents’ economic productivity. The open question is whether another commercially valuable application can broaden demand beyond today’s concentrated buyers quickly enough to outrun falling prices and better cost management, rather than simply delivering more tokens at a lower cost.
Credit: Source link
