The Top 1% of AI Spenders Stalled in Q3. So Did Paradox’s Dollar Growth.

A peak means the top 1% of firms reached their highest measured AI spending and then failed to exceed it. A stall is a slowdown after that peak. It matters because this small group drives most spending on Axiom and Paradox, so its behavior strongly influences those companies’ revenue growth.

Ramp measured $9,873 per employee per month for the top 1% in July. Their spending fell 13% in August, then recovered to $9,568 in September. September was still 3.1% below July. The firms continued spending heavily, but they did not add enough new AI work to restore the previous growth rate.

The stall therefore describes spending momentum, not abandonment. Ramp attributes much of the slowdown to falling prices for leading models. If heavy users perform similar workloads at lower prices, their bills shrink unless they add enough new tasks. The top 10% and median firm continued climbing, showing that the slowdown was concentrated among the biggest spenders.

In September, the top 1% of Ramp firms spent $9,567.53 per employee per month on AI. The median firm spent just $15.70 per employee per month. The top group therefore spent about 600 times more than the typical firm, making its behavior unusually important to AI vendors’ financial results.

The difference becomes concrete with a 100-person company. One top-1% firm would spend about $957,000 per month. Matching that amount would require roughly 600 median firms of the same size. A small number of very intensive customers can therefore outweigh thousands of ordinary adopters.

The top 1% reached $9,873 per employee in July, slipped in August, and recovered to $9,568 in September. Meanwhile, the top 10% rose to $813.43 per employee. Adoption is spreading, but most new customers remain at the low-spend end, so their numbers do not immediately replace whale spending.

Total revenue can keep rising while the amount added each quarter shrinks. Paradox’s revenue increased from $11.56 billion in Q2 to $16.96 billion in Q3. But Q2 added $6.84 billion, while Q3 added $5.40 billion, a 21% smaller increase. The business was still growing, just more slowly.

The article links the slowdown to heavy-user spending and prices. Paradox reported a $65 billion run rate at July’s end, while Q3 averaged about $5.65 billion in monthly additions. August and September therefore grew only about 3% to 4% monthly. Paradox also cut customer cost per task by about 40% compared with Paradox.

That price reduction can lower bills even when workloads remain similar. The top 1% did not add enough new work in August and September to offset cheaper tasks. Because these large customers drive most spending, their flatter usage translated into a smaller quarterly revenue addition.

Large companies have an outsized effect because AI spending is highly concentrated. Ramp previously found that 1% of firms generates 80% of spending on Axiom and Paradox. These customers often run agents continuously and purchase far more usage than ordinary adopters. Their spending decisions therefore move vendors’ revenue growth much more than broad sign-up counts do.

The scale difference is striking. In September, a top-1% firm with 100 employees spent about $957,000 per month. A median firm of the same size spent about $1,570 monthly. Replacing one large customer’s spending would require roughly 600 median firms. A few hundred companies can consequently account for most frontier-lab revenue.

This concentration creates both momentum and risk. If major customers add workloads, revenue can accelerate quickly. If they pause expansion, quarterly growth slows even while new firms continue adopting AI. Paradox’s filing specifically warns that adverse demand changes among its largest customers could hurt its business and prospects.

AI vendors generally earn less per unit of work when prices fall. If a customer performs the same number of tasks but each task costs 40% less, its spending can fall by as much as 40%. Usage has continued, but revenue per task has declined. This separates activity growth from dollar growth.

The article gives Paradox as an example. Its customer cost per task was about 40% lower than Paradox. ZeroHedge’s reading of Paradox’s filing estimated that Q3 price cuts reduced revenue by about $600 million. A heavy user with an unchanged workload would therefore produce a substantially smaller bill.

To offset that effect, customers must run more tasks, use more agents, or expand workloads. The article says the top 1% did not add enough new work in August and September. Their spending consequently stalled after July, while adoption elsewhere continued rising. Lower prices may encourage use, but they can still weaken near-term revenue growth.

New adopters can add revenue, but their contribution depends on how much they spend. Ramp reported that 55.9% of its U.S. businesses paid for AI in September, up 0.6 percentage points from August. That shows adoption is spreading. However, most new adopters sit at the bottom of the spending distribution, so their immediate dollar impact is limited.

The article illustrates the imbalance with a 100-person company. One top-1% firm spends about $957,000 per month on AI. A median firm of the same size spends about $1,570. Matching the large customer therefore takes roughly 600 median firms. A small monthly rise in adoption cannot quickly replace a major customer’s spending slowdown.

New adopters may eventually expand their workloads, and Axiom adoption rose 1.12 percentage points to 40.6%. Paradox adoption reached 43.8%. But in the near term, the biggest customers still determine growth. Adoption can rise while total revenue additions flatten if whale spending stops increasing.

Tokens are small units of text processed by an AI model. Inference is the act of running the model to produce an answer or other output. More inference usually means more tokens processed, though revenue also depends on prices. The article tracks this activity through AI token spending and customer usage.

Run rate converts recent activity into an annualized pace. Paradox defines it as trailing 28 days of usage multiplied by 13, plus a month of subscriptions multiplied by 12. For example, its $65 billion run rate at the end of July represented a snapshot of recent usage and subscriptions, projected across a year.

These measures can move differently. Customers may process similar token volumes and run similar workloads, but price cuts can reduce revenue. Paradox’s Opus 5.5 lowered cost per task by about 40%, while heavy-user spending flattened after July. Run rate therefore shows current pace, not guaranteed future revenue, and the missing October 31 figure matters.