Enterprise AI spending hit $176.8 billion in 2026: agents now spend more than humans
Direct answer
On November 3, 2026, Menlo Ventures published its 2026: The State of Enterprise AI report: enterprise AI spending reached $176.8 billion, 5.4 times the $33 billion of 2025, accelerating from 3.9x growth the year before. The deeper shift: agents are now the biggest consumers of AI models. On OpenRouter, agents generate 80% of all tokens, about 7 times human-driven usage. Since agent traffic overtook human traffic in February, agent token consumption has grown 37x, against 5x for humans. And the money concentrates: Anthropic, OpenAI and Google capture 69% of enterprise AI spending, up from 50% in 2025.
The numbers that matter
- $176.8 billion, 5.4x more: enterprise AI spending rose from $33 billion in 2025 to $176.8 billion by September 2026, per the report.
- Anthropic 55%, OpenAI 31%, Google 9%: in enterprise model API spending, Anthropic rose from 40% to 55%, OpenAI from 27% to 31%, and Google fell from 21% to 9% in a year.
- 80% of tokens come from agents: on OpenRouter, agents generate 4 of every 5 tokens, about 7 times human usage.
- Open source weighs in usage, not in dollars: open models handle about 1 in 5 tokens, but only about 1 in 25 dollars of model spending.
- The study's base: a survey of nearly 400 US enterprise decision-makers overseeing more than $4 billion in AI spending, plus a bottom-up market analysis. It is the fourth annual edition of the report.
What changes for your business
- Budget by task, not by license: if agents consume 80% of tokens, your AI cost depends on how many tasks they run and what each costs, not on how many users you have.
- Pick the vendor by use case: with Anthropic at 55% of API spending and OpenAI at 31%, the practical question is which model performs best on your concrete tasks, tested on 30 to 50 real tasks before signing.
- Open source enters through price: 1 in 5 tokens already runs on open models. For routine, cost-sensitive tasks, testing an open model before paying frontier prices is worth it.
- Concentration raises lock-in risk: three labs capture 69% of spending. A price or terms change by one of them moves your whole cost, so design agents that can switch models without a rewrite.
Note: this article summarizes Menlo Ventures' November 3, 2026 release and coverage of the report. Menlo Ventures is an investor in Anthropic and OpenRouter, two of the companies the report measures, a fact its own release discloses. The spending and share figures come from the report and describe the US enterprise market.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- AI ROI: the real 2026 numbers
- How much it costs to implement AI in a business in 2026
- 5 AI trends for businesses in 2026 and 2027
Frequently asked questions
How much do enterprises spend on AI in 2026?
Per Menlo Ventures, $176.8 billion by September 2026, 5.4 times the $33 billion of 2025.
How much AI usage already comes from agents?
On OpenRouter, agents generate 80% of all tokens, about 7 times human usage, and their consumption has grown 37x since February 2026.
Which vendor leads enterprise AI spending?
Anthropic, with 55% of enterprise model API spending per the report, followed by OpenAI at 31% and Google at 9%.
Does open source really compete?
In usage, yes: it handles about 1 in 5 tokens. In dollars, not yet: only about 1 in 25 dollars of model spending.
