Enterprise AI Spending Hit $177 Billion in 2026 as Agents Take Over: What Menlo Ventures Found

Key takeaway: Menlo Ventures reports enterprise AI spending reached $176.8 billion by September 2026, up 5.4 times from $33 billion in 2025. AI agents, not new buyers, drove the surge: agents now generate 80% of tokens on OpenRouter, and companies spent roughly 2.5 times what they had budgeted.

Menlo Ventures released its 2026: The State of Enterprise AI report on November 3, 2026. Its headline number is striking: enterprise AI deployment spending surged from $33 billion to $176.8 billion in one year — more than $140 billion in new annual spending, in a market worth just $1.7 billion as recently as 2023.

The report combines a survey of nearly 400 U.S. enterprise decision-makers overseeing more than $4 billion in AI spending with a bottom-up market model. Its dollar figures are Menlo Ventures estimates, not a census — and Menlo discloses that it is an investor in Anthropic and OpenRouter, two companies whose data feature prominently in the findings. Keep that caveat in mind throughout.

What changed: agents became the biggest users of AI

Menlo’s central argument is that 2026 was the year AI stopped being mainly something people ask questions of, and became something that does work on its own. The firm points to Anthropic’s Opus 4.5, released in November 2025, as the inflection point: for the first time, models could carry out complex, multi-step tasks with limited human oversight.

On OpenRouter, a platform routing developers to hundreds of models, agents now account for 80% of all tokens — roughly seven times human-driven usage, or 130.9 trillion tokens per week. Agent traffic tripled in the week after Opus 4.5 launched, overtook human traffic in early February 2026, and has since grown 37 times, versus five times for human usage. OpenRouter is only one channel, Menlo notes, but it shows consumption shifting from humans asking questions to machines doing work.

That shift changes the economics. As Menlo principal Derek Xiao put it, spending is “no longer constrained by the number of employees using it, but by how much work it can get done.” Companies entered 2026 planning to nearly double AI spending; on average they spent 4.7 times their 2025 level against a planned 1.9 times — roughly 2.5 times budget. Median spending also beat plans, at 2.2 times last year’s level against 1.8 times expected.

2025 vs 2026: the numbers side by side

Measure	2025	2026
Enterprise AI deployment spending	$33 billion	$176.8 billion (by September)
Year-on-year growth	3.9x	5.4x
Share of enterprises paying for AI	55%	65%
Frontier labs’ share of deployment spending	50%	69% ($122.7 billion)
Anthropic / OpenAI / Google share of API spending	40% / 27% / 21%	55% / 31% / 9%
Open-source share of tokens / spending	11% / small	19% / about 4%
Spending on building models (training)	—	$102.5 billion

Where the money went

Of the $176.8 billion spent using AI, $99.5 billion went to infrastructure and $77.3 billion to applications. Coding ($34.0 billion) and AI assistants and productivity tools ($25.0 billion) make up 76% of application spending, because they offer environments where agents can act, workflows they can learn, and results that can be verified. Healthcare is the largest vertical market at $3.7 billion.

Anthropic, OpenAI, and Google captured 69% of deployment spending, up from 50% last year, and Anthropic plus OpenAI alone take 86% of enterprise API spending. Open-source models handle about one in five tokens but only about one in 25 dollars of model spending. The newcomer-incumbent split depends on category: AI-native firms capture 96% of coding spending, while incumbents lead IT (91%), data science (88%), and HR (68%).

Notably, Menlo found little evidence of companies replacing core systems such as Salesforce’s CRM or Workday’s HRIS with AI. New budgets funded 55% of incremental AI spending, while 22% came from existing software budgets — mostly staying with incumbents converting licences to AI offerings, such as Microsoft Office to Copilot.

What this means if you are buying AI

Cost: Budget for work, not seats. Agent workloads run in parallel and keep going after people step away, which is why average spending landed at 2.5 times plan. Set usage caps and track cost per completed task, not cost per token.

Prerequisites: Agents pay off where work can be checked — which is exactly why coding and assistant tools lead spending. Before deploying agents in a new workflow, confirm there is a fast, reliable way to verify what the agent did.

Risk: Concentration is real. With two vendors taking 86% of enterprise API spending, keep a second model in production, even for a small share of traffic, so a switch stays possible.

Knowns, unknowns, and what to watch

Known: The figures, methodology, and disclosures are public, and the direction — rapid growth driven by agents — matches other 2026 signals, including recent benchmarks measuring what agents can actually do at work.

Unknown: Whether the pace is sustainable. Growth came mostly from existing adopters spending more, not new buyers — adoption rose only from 55% to 65%. If agent results disappoint, the same mechanism could pull spending back just as fast.

What to watch: Whether Google regains API share with its next flagship model, whether open-source models convert token share into revenue, and how vendor pricing changes — such as OpenAI’s shared ChatGPT and Codex allowance from December — reshape what companies actually pay.

Analysis: spending follows work, not headcount

Analysis: The most important line in this report is not the $177 billion — it is the gap between 5.4-times spending growth and a 10-point rise in adoption. AI has crossed from a per-seat business into a per-task business, so the old budgeting habit of counting users and multiplying by a licence price no longer describes the bill. For sceptics, the report cuts both ways: it answers “where is the revenue?” with a very large number, but that number comes from a venture firm invested in the winners it names, so independent confirmation matters more than ever.

Frequently asked questions

How much did enterprises spend on AI in 2026?

Menlo Ventures estimates enterprise AI deployment spending reached $176.8 billion by September 2026, up from $33 billion in 2025 — a 5.4-times increase. Spending on building models added another $102.5 billion, so enterprises spent about $1.70 putting AI to work for every $1 spent training models.

Why did spending grow so fast if adoption barely rose?

The share of enterprises paying for AI rose only from 55% to 65%. Growth came from existing adopters spending much more, driven by AI agents: agents now generate 80% of tokens on OpenRouter, and agent workloads scale with the amount of work done rather than headcount, so costs rise faster than employee numbers.

Which companies capture most enterprise AI spending?

Anthropic, OpenAI, and Google together capture 69% of enterprise AI deployment spending, up from 50% last year. Within enterprise LLM API spending, Anthropic leads at 55%, followed by OpenAI at 31% and Google at 9%. Anthropic and OpenAI together account for 86% of API spending.

Where is enterprise AI spending concentrated?

Coding ($34.0 billion) and AI assistants and productivity tools ($25.0 billion) make up 76% of the $77.3 billion application market. Healthcare is the largest vertical market at $3.7 billion. AI-native companies dominate coding with 96% share, while incumbents lead established categories like IT, data science, and HR.

Should businesses trust the $177 billion figure?

Treat it as an estimate, not a census. Menlo built it from a survey of nearly 400 U.S. enterprise decision-makers plus a bottom-up model, and it discloses investments in Anthropic and OpenRouter, whose data underpin headline claims. The direction is consistent with other 2026 evidence, but the precise total awaits independent corroboration.

Sources

Menlo Ventures 2026 State of Enterprise AI Report announcement, GlobeNewswire, November 3, 2026

Menlo Ventures, 2026: The State of Enterprise AI (full report), November 3, 2026