Enterprise AI spending hit $176.8 billion in 2026, up 5.4x — agents now generate 80% of model tokens

Menlo Ventures released its 2026 State of Enterprise AI report on November 3, 2026, its fourth annual edition of the survey-based study. The report, authored by Tim Tully, Joff Redfern, Deedy Das and Derek Xiao, combines a survey of nearly 400 U.S. enterprise decision-makers overseeing more than $4 billion in AI spending with a bottom-up analysis of the market. Its headline finding: enterprise AI spending reached $176.8 billion by September 2026, up from $33 billion in 2025.

That is growth of 5.4 times in a year, accelerating from 3.9 times the year before. “Wall Street is questioning whether the massive investments in AI will pay off, but our latest data shows just how quickly enterprise demand is growing,” Tim Tully, a partner at Menlo Ventures, said in the announcement. “Enterprise AI spending has jumped more than fivefold, from $33 billion last year to $177 billion this year.”

Agents are doing the consuming now

Menlo’s explanation for the jump is agents. On OpenRouter, agents generate 80% of all tokens, roughly seven times human-driven usage, the report finds. Since agent traffic overtook human traffic in February, agent token consumption has grown 37 times, compared with five times for human-driven usage. Derek Xiao, a principal at Menlo Ventures, put the shift in economic terms: “When AI can do the work itself, enterprise spending is no longer constrained by the number of employees using it, but by how much work it can get done.”

The spending followed the work. Companies entered 2026 expecting to spend more on AI and budgeted nearly twice what they had spent in 2025 — and still ended up spending 2.5 times what they had budgeted, according to the survey. For every dollar spent on AI model training this year, enterprises spent $1.70 putting AI to work, Xiao said.

Deeper pockets, not many more buyers

The growth did not come mainly from new buyers. The share of enterprises paying for AI rose only modestly, from 55% to 65%, Menlo says, suggesting the increase was driven primarily by greater spending among adopters rather than a wave of new buyers.

New budgets funded 55% of the incremental enterprise AI spending, while 22% came from existing software budgets. Most of that existing-budget spending stayed with incumbents as they converted licenses to AI offerings, such as Microsoft Office to Copilot, the report says — and Menlo found little evidence of enterprises replacing core systems such as Salesforce’s CRM or Workday’s HRIS.

Where the money concentrates

Applications are now a major engine of growth. Enterprise AI application spending reached approximately $77 billion, up from $19 billion last year. Coding, at $34 billion, and AI assistants and productivity tools, at $25 billion, together account for 76% of application spending. Healthcare remains the largest vertical industry market, at $3.7 billion.

The split between newcomers and incumbents depends on the category. AI-native providers capture 96% of coding spending, while incumbents lead IT (91%), data science (88%) and HR (68%). “When AI changes how work gets done, new companies have an opening to reshape a market,” Joff Redfern, a partner at Menlo Ventures, said. “When AI improves the way people already work, established players are harder to displace.”

At the model layer, concentration is sharper still. Anthropic, OpenAI and Google together account for 69% of enterprise AI deployment spending, up from 50% in 2025. Within enterprise LLM API spending, Anthropic extended its lead from 40% to 55%, OpenAI rose from 27% to 31%, and Google fell from 21% to 9%. Open-source models now handle approximately one in five tokens but account for only about one in 25 dollars of model spending.

Read the numbers with care

Menlo’s totals are estimates, not audited revenue. The firm says its dollar figures are built from a bottom-up model informed by survey responses, public filings and third-party reports, and its token-share figures draw on OpenRouter data and Menlo’s own estimates of model mix. The survey itself covers U.S. enterprise decision-makers, so the findings describe that market, not global spending.

Menlo also discloses an interest in the story it is telling: the firm is an investor in Anthropic and OpenRouter. That does not invalidate the numbers, but it is a reason to treat them as one firm’s estimate — and to watch whether other spending trackers land in the same place. What the report establishes, on its own terms, is the shape of demand in 2026: agent-led, concentrated among a few labs, and larger than even the buyers themselves budgeted for. The full report is published on Menlo Ventures’ site.