US companies spend 2.5 times what they planned on AI

A Menlo Ventures report shows AI spending far above budget at American companies, driven by a few heavy users. What to do before your own bill goes up.

American companies planned to nearly double their AI spending in 2026 and, on average, ended up at 4.7 times the previous year’s level, about 2.5 times what they expected, according to the annual report from Menlo Ventures published on Tuesday (November 3). The investment firm estimates the market went from US$33 billion in 2025 to US$176.8 billion in September 2026. Yahoo Finance also covered the data.

How to read this number

The survey heard from almost 400 decision-makers at US companies. The market size is Menlo’s own estimate, and the methodology states that revenue is shown at an annualized run rate. In addition, Menlo invests in Anthropic and OpenRouter, both cited in the study.

The most useful detail is in the median: 2.2 times the previous year, against a planned 1.8. According to the report, a small group of very heavy users accounts for a disproportionate share of the spending.

Adoption rose only a little, from 55% to 65% of companies. The jump came from spending by those who were already users, and Menlo attributes it to agents doing longer tasks. The signal matches what Amazon showed in its October earnings.

What this changes for your company

1. Treat AI like a utility bill. A per-user license has a predictable price. Agents usually charge by usage, and that volume grows as they take on more tasks. Set a limit per department, with an alert before the ceiling.

2. Find out who concentrates the spending. If a few people consume a lot, the question isn’t just how much it cost, but what those people delivered. Sometimes it’s the company’s best investment. Sometimes it’s a badly designed process running nonstop.

3. Don’t count on software cuts to pay the bill. In the study, 55% of the additional spending came from new budgets and 22% from existing software budgets. Build your 2027 budget expecting AI to add to costs, not replace them.

In 29 years of programming, I’ve never seen usage-based costs control themselves: whoever designs the process has to set the limit before the invoice arrives.


Originally published on IWA7 Insights, IWA7 Digital’s blog, which is now part of Escovando Bytes.

Keep readingSee all →