Enterprise AI Spending Hits $177 Billion: Why Agents Are Now the Dominant Consumer of AI

Menlo Ventures has published its 2026 State of Enterprise AI report, and the headline figure is difficult to ignore: enterprise AI deployment spending reached $177 billion by September 2026. That is a 5.4x increase from the $33 billion recorded a year earlier, and it represents more than $140 billion in new annual spending flowing into enterprise AI in a single year.

The report, which draws on a survey of nearly 400 US enterprise decision-makers responsible for more than $4 billion in AI spending, also contains a second figure that redefines how the industry should think about AI consumption. On OpenRouter, a platform that gives developers access to hundreds of AI models, agents now generate 80% of all tokens. That is roughly 130.9 trillion tokens per week, seven times human-driven usage. Agents have overtaken humans as the dominant consumers of AI.

The scale of the shift

To put the $177 billion figure in context, consider that just three years ago, in 2023, enterprise generative AI deployment spending was $1.7 billion. It grew to $8.5 billion in 2024, then $33 billion in 2025. The acceleration is not linear. It is compounding. Every dollar that went into developing models this year generated $1.70 in deployment revenue, suggesting that the enterprise AI market is not merely expanding but becoming self-sustaining.

Critically, 55% of that incremental spend came from net new budgets rather than existing software or labour budgets. AI is not just eating the software market. It is growing it. Menlo Ventures notes that the "SaaSpocalypse" fears earlier this year, which predicted AI would cannibalise traditional software revenue, have not materialised. The labs' coding agents grew from $1 billion to $24 billion, yet the rest of the coding market still grew 2.9x to $10 billion. Even excluding the labs' own applications, the largest AI app markets are the ones the labs target most aggressively: coding, AI assistants ($9.5 billion) and healthcare ($3.7 billion).

Agents are the main event

The most important structural change in the report is the prominence of agents. The launch of Anthropic's Opus 4.5 in November 2025 marked an inflection point, enabling AI to execute complex, multi-step tasks with limited human oversight rather than simply answering questions in conversational turns. Agent traffic on OpenRouter tripled the week after Opus 4.5's launch, overtook human traffic in early February following subsequent model releases, and has since grown 37x. Human-driven usage, by comparison, grew 5x.

This shift from humans asking questions to machines doing work has direct consequences for enterprise buyers. AI spend is concentrating where agents work best: coding ($34.0 billion) and chatbots ($25.0 billion) make up 76% of the $77.3 billion app market because these are environments in which agents can act, workflows they can learn, and results they can verify. For buyers, this means the highest-ROI applications of agentic AI are not hypothetical. They are already measured, funded and scaling.

The lab landscape is concentrating

Frontier labs now command 69% of deployment dollars, up from 50% last year. Anthropic's share of enterprise API spend increased from 40% to 55%, while OpenAI's grew from 27% to 31%. Google's share fell from 21% to 9%. The duopoly is real, and it is deepening. Open-source models, meanwhile, represent 19% of usage but only 4% of spend, suggesting significant latent demand that has not yet translated into commercial revenue.

For enterprise buyers, this concentration carries procurement implications. The platform decisions made today are likely to persist for years. Companies committing to a single enterprise-wide AI platform with a common architecture are showing better outcomes than those stitching together multiple vendor stacks. Vendor selection is becoming more consequential, not less.

What the numbers mean for enterprise buyers

The report offers four practical takeaways for companies buying agentic AI.

First, the proof-of-concept phase is ending. With $177 billion in deployment spend already in market, the question is no longer whether agentic AI works. It is which platforms deliver value fastest and integrate cleanly with existing systems. Buyers should prioritise vendors that can demonstrate named customers, clear metrics and time-to-value in months rather than years.

Second, coding and customer-facing automation remain the strongest entry points. These categories account for the majority of agent-driven spend because they offer measurable outcomes and environments where agent actions can be verified. Buyers should evaluate agent platforms on where they have demonstrated production traction, not where they claim readiness.

Third, budget is available, but it is new budget. Because 55% of spend is coming from net new allocations, procurement teams may need to build new business cases rather than repurposing existing software budgets. The organisations winning funding are those that frame agentic AI as a growth enabler, not a cost reduction tool.

Fourth, infrastructure readiness remains the bottleneck. OpenAI's experience is instructive: it had to build an entirely new Git backend, Pando, after Codex drove commits up 1,378% and reliability dropped below 90%. If the world's leading AI lab could not scale its infrastructure for agentic coding without a ground-up rebuild, enterprises should assume their own infrastructure will face similar strain and plan accordingly.

What the numbers mean for suppliers

For companies building and selling agentic AI, the report confirms that the market is large, growing and increasingly competitive.

The first signal is that buyer expectations have shifted. Enterprises are no longer testing whether agents work. They are measuring how quickly platforms deliver ROI and how well they fit into hybrid human-agent workflows. Suppliers that can articulate a clear value timeline, with named customers and specific metrics, will win procurement cycles against those still pitching possibility.

The second is that data readiness is the upstream battle. Menlo Ventures notes that AI-native companies are winning categories where agents can operate in clean, structured environments, while incumbents are gaining ground in IT (91%), data science (88%) and HR (68%) where existing systems and data integrations matter more. Suppliers that help buyers unify, clean and govern data alongside deploying agents are solving the right problem in the right order.

The Agentic Expo angle

Agentic Expo, at Olympia London on 23 to 24 March 2027, exists because the shift Menlo Ventures documents is already underway. The buyers attending the event are not exploring whether agents are real. They are deciding which platforms, governance frameworks and infrastructure partners will carry them from pilot to production at enterprise scale.

The $177 billion figure is the headline. But the underlying story is that agents have become the primary consumers of AI, that spend is compounding faster than most organisations planned for, and that the infrastructure to support agent scale is still catching up. The gap between ambition and readiness is where Agentic Expo operates.