Enterprise AI spending in 2026 grew far faster than corporate finance teams had planned, and a small group of heavy AI agent users was largely responsible for the overshoot, according to a new report from venture firm Menlo Ventures.

Companies had budgeted for AI spending to roughly double this year. It did considerably more than that. Menlo Ventures released its 2026 State of Enterprise AI report on November 3, 2026, and the findings suggest enterprises did not simply miss their AI budget targets — they missed them by a wide margin. A gap of that size — actual outlays at roughly two and a half times what was planned — suggests budgeting processes have not yet kept pace with how quickly AI consumption can scale inside a company.

A summary of the findings shared on X stated that a small group of heavy AI agent users drove enterprise AI spending to 4.7 times 2025 levels, while companies had planned for 1.9 times.

The numbers behind the overrun

The report's top-line deployment figure is striking. Enterprise AI deployment spending reached $176.8 billion by September 2026, up from $33 billion in 2025 — a 5.4x jump that also marks an acceleration, given that the prior year's growth rate was 3.9x.

Enterprises had forecast growth in the range of 1.9x to 2x. Instead, they ended up spending roughly 2.5 times their planned AI budgets. The gap between plan and reality was widest among the heaviest users, particularly organizations running autonomous AI agents.

Notably, the overrun was not driven by a broad wave of new adopters. The share of enterprises investing in AI rose only modestly, from 55% to 65%. That detail matters for how the numbers should be read: the total grew because existing AI buyers spent far more, not because a much larger base of companies started buying.

As for where the money came from, the report found that 55% of the increased spending came from new budgets, while only 9% was shifted from traditional SaaS tools to AI-native alternatives. In other words, the overrun was funded mostly by money allocated outside existing software budgets rather than by replacing incumbent tools — a sign that enterprise AI spending is, so far, behaving as additive spend rather than substitution within existing line items.

Agents are now the biggest customers

Autonomous AI agents now account for 80% of token usage on platforms built for model interaction, according to the report. Agent activity has grown 37 times since agents overtook human usage in early 2026. This is where the planning gap becomes clearest: token-metered consumption scales with workloads rather than headcount, so spending inside agent-heavy organizations can outrun budget assumptions set before that usage pattern took hold.

Where the money went

Spending on AI applications reached approximately $77 billion, compared with $19 billion the year before. Two categories dominated: coding tools took in $34 billion, while productivity tools accounted for $25 billion. Together, coding and productivity made up 76% of application spending. That concentration shows enterprises steering application budgets toward tools embedded in day-to-day work.

At the infrastructure layer, the leading frontier labs tightened their grip. Anthropic, OpenAI, and Google captured 69% of AI deployment spending in 2026, totaling $122.7 billion, up from a combined 50% share in 2025. More than two-thirds of enterprise deployment dollars now flow to three providers — a marked consolidation from the more even split of a year earlier.

Within large language model API spend specifically, Anthropic led with 55%, followed by OpenAI at 31% and Google at 9%.

How the survey was conducted

The report draws on a survey of approximately 400 US decision-makers, conducted from September 8 to September 26, 2026. The figures reflect a snapshot of the US enterprise market taken roughly five weeks before the report's November 3 release. With 2026 actuals landing well above plan, the data gives finance teams a concrete baseline for the coming budget cycle, where the open question is whether planning assumptions are reset to reflect agent-driven consumption.