The median user of AI models through Databricks’ Unity Gateway spent $2.49 in the week of 2 November, up from $0.40 in the week of 20 July, according to the Databricks AI Index, which Databricks published on 16 November. Our number: that is 6.2x in 15 weeks, by our arithmetic.
The heaviest users moved far less. Weekly spend at the 99th percentile went from $511.44 to $699.74, about 1.4x, and at the 90th percentile from $49.47 to $95.96, about 1.9x. A top-1% user now spends about 281x the median, down from about 1,279x, by our arithmetic.
Business Pill · Usage-based pricing explained
An evergreen Business Engineer explainer on usage-based pricing, paying for work rather than seats. It teaches the general idea only and does not discuss the Databricks AI Index.
The key insight: As we read the Databricks AI Index, the spread between light and heavy users is closing from the bottom: by our arithmetic the median user’s weekly spend rose 6.2x in 15 weeks while P99 spend rose about 1.4x, so a top-1% user went from about 1,279x the median to about 281x.
How Databricks Measures Spend
The index page describes the chart’s method: “Each user’s input, cached and output tokens priced at each provider’s published list price; human users only.” Databricks co-founder Patrick Wendell wrote on X that the index data “covers >10,000 companies with broad representation (Digital Natives, Fortune 500, Global 2000, etc…)”.
The page shows three lines, the median, the 90th percentile (P90) and the 99th percentile (P99), weekly from the week of 20 July to the week of 2 November. Rebased to 100 in the first week, the median reached 622, P90 194 and P99 137, by our arithmetic.

What Was Posted on X
Patrick Wendell, a Databricks co-founder, wrote in the fourth post of his 16 November thread that most AI spend comes from power users, with P99 spend at $700 a week against $2.50 at the median, and that power-user spend appears to be holding mostly flat, likely because of budgeting. His figures round the index chart’s $699.74 and $2.49 for the week of 2 November, which this piece uses.
What Wendell Says Is Happening
Wendell wrote: “Most spend is incurred by power users (P99 spend is $700/week vs P50 is $2.50/week).” He added that “Power user spend appears to be holding mostly flat, likely as a result of budgeting, while the median user spend is quickly growing from a small base.” His $700 and $2.50 are the chart’s $699.74 and $2.49, rounded.
The page lists Databricks posts on the same theme, including “Managing AI Coding Costs at Scale” of 7 August and “The AI Bill is coming” of 22 October, described as “Why AI coding costs rise with usage, and how engineering leaders control them with gateways, routing and visibility.”

The Median Rose Every Week Since August
The median rose in every week from the week of 10 August, when it was $0.36, to the week of 2 November, the page’s figures show. P99 peaked at $781.48 in the week of 21 September and was $699.74 in the week of 2 November. P90 was $95.66 in the week of 21 September and $95.96 in the latest week.
As a scale check only, $2.49 a week held for 52 weeks would be about $129 a year, by our arithmetic. The same index puts open-weight models at 30.9% of tokens on Databricks-hosted models, and Wendell links them to cost: our pieces cover Anthropic’s 70% of closed-model spend and the open-weight rise.
Our earlier pieces covered Menlo Ventures’ enterprise AI spend figures and a Berkeley paper on agent harness cost.
The Structural Read
As we read the series, the median’s rise is steady rather than a one-week jump: it went up in every week from the week of 10 August, when it was $0.36, to $2.49 in the week of 2 November.
As we read Wendell’s thread, Databricks connects the flat top end to budgeting, and he says so with the word likely; the chart shows P99 at $781.48 in the week of 21 September and $699.74 in the week of 2 November.
Patrick Wendell, Databricks co-founder, on X
“Most spend is incurred by power users (P99 spend is $700/week vs P50 is $2.50/week).”
Three Implications
FOR FINANCE TEAMS At $2.49 a week the median user is about $129 a year at list prices by our arithmetic, a scale check rather than a forecast.
FOR PLATFORM OWNERS The page lists Databricks posts on per-user budgets and on controlling AI coding costs with gateways, routing and visibility.
RELATED Our first piece on the index covers which labs get the spend.
The Business Engineer Lens
This story maps onto the Business Engineer framework The CFO’s Guide to the Token Economy.
The framework says of AI: “In 2026 it became a CFO problem, because the economics finally turned legible”.
As we read it, the index makes that spend legible per person: the median user at $2.49 a week, the top 1% at $699.74, both at list prices. The framework predates the index and does not discuss it.
What Is Not Established
Wendell offers budgeting as the likely reason power-user spend held flat; the chart shows spend levels, not the budgets. The figures cover human users of companies that route AI through Unity Gateway, at list prices.
We read the index page in full, including the values behind each chart, and Wendell’s six-post thread. We did not contact Databricks for this piece.
The Bottom Line
The median Unity Gateway user’s weekly AI spend rose from $0.40 to $2.49 between the weeks of 20 July and 2 November, 6.2x by our arithmetic, while P99 spend rose from $511.44 to $699.74, about 1.4x.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. All figures come from the Databricks AI Index page at developers.databricks.com/ai-index, which we read in full on 16 November 2026, including the exact values its charts plot, and from Databricks co-founder Patrick Wendell’s six-post thread on X of the same day. The page marks the spend chart as updated on 2 November 2026; it prices each human user’s tokens at providers’ list prices.
Multiples, ratios between percentiles, the rebased index and the annual scale check are our arithmetic. Databricks was not contacted for this piece.
Sources: Databricks AI Index (page updated 2 Nov 2026, published 16 Nov 2026) · Patrick Wendell on X, 16 Nov 2026 (post 1 of 6) · Patrick Wendell on X, 16 Nov 2026 (post 4 of 6)









