Enterprise AI spending reached $176.8 billion by September 2026, up more than fivefold from $33 billion a year earlier, according to Menlo Ventures’ fourth annual State of Enterprise AI report, published November 3. Agents are the engine of the surge: they now generate 80 percent of model tokens on OpenRouter, while Anthropic, OpenAI and Google together capture 69 percent of deployment dollars. The figures are Menlo’s modelled estimates, not audited accounts, but they put hard numbers on a year in which AI stopped being an experiment and became a budget line.
The headline numbers
Menlo Ventures has published its fourth annual State of Enterprise AI report, and its headline figure is hard to ignore. Enterprise spending on deploying AI reached $176.8 billion by September 2026, up from $33 billion a year earlier. That is a 5.4-fold increase in twelve months, faster than the 3.9-fold growth recorded a year ago.
For every dollar the industry spends building models, enterprises now spend about $1.70 putting them to work. Menlo puts spending on using AI at $176.8 billion, against $102.5 billion on building models. The report’s central claim is that agents explain the jump, because agents tie spending to work performed rather than users served.
Agents take over the token supply
The report’s most striking chart comes from OpenRouter, a platform that gives developers access to hundreds of models. Agents now account for 80 percent of all tokens there, roughly seven times human-driven usage. That is about 130.9 trillion tokens a week, a scale that shows how quickly machines doing work have overtaken people asking questions.
Menlo dates the inflection to November 2025, when Anthropic’s Opus 4.5 first let AI move beyond conversational turns into complex, multi-step tasks. Agent traffic tripled the week after that launch, then passed human traffic in early February. OpenRouter is only one channel, the report concedes, but the direction of travel is unmistakable.
That shift shows up in what enterprises buy. The AI application market reached $77.3 billion, up from $19 billion last year, and two markets account for 76 percent of it. Coding stands at $34.0 billion and enterprise chatbots at $25.0 billion, while healthcare, one of the world’s largest industries, ranks a distant third at $3.7 billion.
Three labs take more than two-thirds of the money
Concentration is the report’s second big theme. Anthropic, OpenAI and Google together command $122.7 billion of enterprise deployment spending, or 69 percent of the total. That share is up from 50 percent last year, and their spending grew 7.5-fold year over year, against 3.3-fold for everyone else.
Within that trio, Anthropic widened its lead. Its share of enterprise API spending rose from 40 percent to 55 percent, while OpenAI’s rose from 27 percent to 31 percent. Google fell from 21 percent to 9 percent, and Menlo says it has not matched its rivals’ flagship models since Gemini 3 Pro.
The scale of the commitments behind those shares is visible in the industry’s own disclosures. Anthropic’s amended filing, for example, lays out roughly $895 billion of purchase commitments and leases, nearly all for computing power. Menlo’s figures and that filing tell the same story: buyers pay a premium for capability, and suppliers lock that demand into long contracts.
Budgets blown, but mostly new money
Enterprises did not plan this surge. The average enterprise planned to nearly double its AI spending in 2026 but ended up spending 4.7 times last year’s level. That is roughly 2.5 times what it budgeted, while median spending reached 2.2 times last year’s level against a 1.8-fold plan.
Where the money came from matters for the software industry’s anxieties. New budgets funded 55 percent of the incremental AI spend, against 22 percent from existing software budgets. Only 9 percent, or $12.6 billion, moved to AI-native vendors, so Menlo concludes AI is expanding the market rather than simply displacing software.
Open-source models tell a subtler story. Their share of tokens nearly doubled, from 11 percent to 19 percent, and their volume rose 22.5-fold to about 2.9 quadrillion tokens a month. Yet their share of spending fell from 6 percent to 4 percent, because average open-source token prices fell 3.3-fold to $0.13 per million.
Post-training, the practice of specialising an open model, remains a minority pursuit. Adoption rose from 6 percent of AI builders in 2025 to 10 percent this year, and 91 percent of those who post-train rely on outside partners. Among the specialists Menlo names are firms such as Applied Compute, whose $350 million raise shows how that niche is commercialising.
Our read
The number to trust most is not $177 billion but the shape of it. A market that quintuples in a year on one use pattern, agents doing verifiable work, depends on agents keeping their reliability gains. If agent quality stalls, the token volumes justifying today’s infrastructure spending stall with it.
Read the concentration figures with Menlo’s own interests in mind. The firm discloses that it invests in Anthropic and OpenRouter, two of the report’s star performers, and its totals are modelled estimates, not audited accounts. Menlo’s own Anthropic stake, worth roughly $14 billion at the company’s private valuation, is part of the story it is telling.
The most durable insight may be the budget miss. Enterprises overshooting their AI budgets by 2.5 times is not careful planning; it is a sign that usage-based pricing for autonomous work breaks the budgeting model entirely. Finance teams that forecast seats and licences now face workloads that spin up in parallel and keep going unattended.
Why It Matters
This report matters because it replaces anecdote with arithmetic for the AI economy’s biggest question: who pays, and for what. A $177 billion market, growing 5.4-fold while adoption rose from 55 percent to 65 percent of enterprises, says the boom is driven by depth of use. That leaves headroom if the long tail follows, and exposure if power users retrench.
It also reframes the competition. For now, money follows capability, and the frontier labs take more than two-thirds of every enterprise AI dollar. But open-source usage is compounding faster than closed-source usage, and agents are pulling blended token prices down. The next report will show whether capability keeps its premium, or whether the price war climbs the stack.
Background
Menlo Ventures has published its State of Enterprise AI report annually since 2023, and the 2026 edition is its fourth. This year’s edition combines a survey of nearly 400 United States enterprise decision-makers responsible for more than $4 billion in AI spending, fielded by Morning Consult from September 8 to 26, 2026, with a bottom-up market model.
Source: Menlo Ventures report and GlobeNewswire release
0 Comments