Menlo Says Average Enterprise AI Spend Grew 4.7x. The Median Grew 2.2x
Menlo Ventures, an Anthropic investor, puts enterprise AI at a $176.8 billion run rate. Average spend per enterprise hit 4.7x last year's level, while the median reached 2.2x.
TL;DR
- Menlo Ventures estimates enterprise AI deployment at a $176.8 billion run rate, with adoption up from 55% to 65%.
- Spend per enterprise averaged 4.7x last year's against a 2.2x median, so most of the expansion comes from heavy users, where budgets break.
- Ramp saw token volumes rise in August and September while token spend stayed flat as prices fell, so watch its next index.
Menlo Ventures published its annual enterprise AI report on November 3, estimating that enterprise AI deployment spending reached a $176.8 billion annualized run rate as of September 2026. The figure comes from Menlo's own bottom-up model, informed by a Morning Consult survey of about 400 US decision-makers fielded September 8 to 26. Like any run rate, it annualizes a pace instead of counting money already spent. Menlo puts the three frontier labs at 69% of the total and Anthropic's share of API spend at 55%, up from 40%. The lab shares rest on respondents' self-reported workloads, and Menlo, an Anthropic investor, discloses that stake in the report.
The share of enterprises paying for AI rose from 55% to 65%. Spend per enterprise nearly quintupled on average, but the median enterprise spent 2.2x last year's level, so power users account for most of the growth. Enterprise AI revenue now grows mostly through heavier use at a small group of buyers, which is also where budgets break.
Usage inside existing accounts is outrunning the plan
Menlo's explanation is that "spending scales with work rather than headcount." An agent that answers tickets or drafts outreach bills per task, so a team that adds volume adds cost without adding a seat. The function numbers show where that's happening. Menlo sizes customer success applications at $2.7 billion, up 4.3x, sales at $1.4 billion, up 3.6x (Menlo also backs Monaco, one of the sales companies it names), and marketing at $1.2 billion, up 1.8x. Menlo says customer success grew fastest of the three because support playbooks document the steps and a resolution defines success, so an agent's work is easy to check. Together those categories come to $5.3 billion, about 3% of the $176.8 billion, so most of this money isn't going to GTM tools. Menlo also says 55% of incremental spend came from new budgets and 22% from software budgets, most of which stayed with incumbents.
The overrun shows the same split. Menlo puts average spend at 4.7x against a planned 1.9x, about 2.5 times over plan. The median enterprise came in at 2.2x against a planned 1.8x, so most of the blowout comes from heavy users and the typical company overshot by a modest margin. Menlo also restated 2025 from $37 billion to $33 billion under a new taxonomy, so don't compare this year's figure with last year's report.
Ramp's October AI Index, built on Ramp's own spend data, complicates the momentum read. Ramp sells AI spend-management tools, and its customers skew smaller than Menlo's enterprise sample. It agrees that spend is concentrated, with 1% of firms driving 80% of OpenAI and Anthropic spend. But Ramp lead economist Ara Kharazian reports "two straight months of virtually no growth in token spend" in August and September, even as token volumes rose, and he puts that mainly on falling frontier prices. Spend per employee at the top 1% of firms rose 11.7% in September but sits 3.1% below its July high, and AI spend hit a new peak in early October, 1.5% above August's, though growth has slowed. Ramp also puts business adoption at 55.9%, with new-customer growth slowing. That supports the expansion read, since heavier use is the main growth left when new buyers dry up. It also exposes a soft spot. Vendors' revenue can stall on price cuts even while usage keeps climbing. Menlo's run rate is pegged to September, the same month Ramp saw flat token spend, though the two measure different populations and should be compared on direction only.
What this means for GTM leaders setting next year's AI budgets
Find out who drove your overrun before you raise the line. Both datasets show spend concentrated in a few heavy buyers, and the same pattern is worth checking inside your own company. If two or three teams or agents account for most of this year's overage, budget those workloads on their own and set the rest closer to plan.
Price usage-based agents per unit of work. Before you sign, ask the vendor to forecast cost against your own volumes, such as tickets per month, accounts worked and calls logged. Then split the contract into a committed line and a usage line, with alerts on each agent. Meters are spreading to systems you already pay for, as we covered when Salesforce started charging agents to read CRM data.
At your next frontier-lab renewal, ask for the discount. Ramp's Kharazian says that, from what he's heard, lab enterprise contracts favor the customer, and OpenAI has been discounting to win Anthropic's top accounts. Before you commit to a volume, check whether your own flat line reflects lower prices or lower usage, and tie any price break to usage you already expect to hit.
Ramp's next index is the one to read. If token spend stays flat for a third month while volumes rise, the labs are cutting price to keep volume growing, and your renewal is the time to claim a share of those cuts.
Enterprise AI revenue now grows mostly through expansion at a small group of heavy buyers, and that's where budgets break.
Pull your last two quarters of AI invoices and find the two or three teams or agents behind most of the overage. Then ask each vendor to forecast cost at your own volumes before you renew.
