By Hamza Faisal

Corporate spending on artificial intelligence has surged to $176.8 billion in 2026, more than five times last year's total, as autonomous agents quietly become the biggest consumers of AI models. The figures come from Menlo Ventures' fourth annual State of Enterprise AI report, released on November 3, which is based on a survey of nearly 400 US enterprise decision-makers overseeing more than $4 billion in AI spending.

Spending climbed from $33 billion in 2025 to $176.8 billion by September 2026, a 5.4-fold jump that actually accelerated from the 3.9-fold growth recorded the year before. The striking part is who is doing the spending: the share of enterprises paying for AI rose only from 55% to 65%, meaning the boom is being driven by existing adopters spending far more, not by a wave of newcomers.

Agents now use more AI than people do

The report's most eye-catching finding concerns who, or what, is actually consuming all this AI. On OpenRouter, a platform that routes developers to hundreds of models, autonomous agents now generate 80% of all tokens, roughly seven times human-driven usage. That works out to about 130.9 trillion tokens per week.

Agent traffic overtook human traffic in February 2026, following the launches of Anthropic's Opus 4.6 and OpenAI's GPT-5.3-Codex, and has since grown 37-fold, compared with five-fold growth for human usage.

That shift is also blowing up corporate budgets. Companies entered 2026 planning to nearly double their AI spending, but on average they spent 4.7 times their 2025 levels against a planned 1.9 times, roughly 2.5 times what they had budgeted. "Companies entered 2026 expecting to spend more on AI, so they budgeted nearly twice what they'd spent in 2025. But even that wasn't enough," said Derek Xiao, a principal at Menlo Ventures.

A handful of labs take most of the money

The money is concentrating fast. Anthropic, Google and OpenAI together capture 69% of enterprise AI spending, about $122.7 billion, up from 50% last year. Spending on deploying AI, $176.8 billion, now far exceeds the $102.5 billion spent on building the models themselves, and Menlo estimates every dollar spent developing models generates $1.70 in deployment revenue.

Within applications, spending reached roughly $77.3 billion, up from $19 billion a year earlier. Coding tools account for $34 billion and enterprise chatbots and assistants $25 billion, together making up 76% of the application market. Healthcare, at 5%, and customer success, at 3%, remain small by comparison.

Open-source models are gaining ground in usage but not in revenue: they now handle about one in five tokens, up from one in nine last year, yet account for only about one in 25 dollars of model spending. Meanwhile, 55% of the incremental spending came from entirely new budgets, while 22% came out of existing software budgets, mostly as incumbents converted licenses into AI tiers, such as Microsoft Office to Copilot.

What the numbers do not prove

Menlo is candid about the limits of its own figures. The dollar totals are modeled estimates built from survey responses, public filings and third-party reports, not audited accounts, and the firm discloses that it invests in both Anthropic and OpenRouter, two of the companies whose data underpin the report.

The report also measures spending, not returns. It does not show whether the $176.8 billion is paying off in productivity or profit. For now, though, the direction is unmistakable: AI spending has grown roughly 100-fold in three years, from $1.7 billion in 2023, and the agents doing the work are the ones running up the bill.