Menlo's $19B 2025 GenAI Spend: Apps Win

TL;DR: The Menlo Ventures 19 billion gen AI spend 2025 estimate says enterprises put more than $19 billion into generative AI applications last year — over half of a $37 billion total. The money went to the application layer: copilots, coding assistants, and role-specific tools that live inside knowledge work. The pattern is clear: buyers are paying for software that does the work, not just the models underneath. That makes connected, findable institutional knowledge the thing that decides whether all that spending pays off.

Two years ago most enterprise AI was a pilot. In 2025 it became a line item. Menlo Ventures, in its third annual survey of enterprise AI buyers, reports that companies spent more than $19 billion on generative AI applications during the year (No Jitter, 2025). That is the slice that sits closest to actual work, and it is where the budget is concentrating.

The headline number is big. The shape of it is more useful.

What does the Menlo Ventures 19 billion gen AI spend 2025 estimate actually measure?

The $19 billion is the application layer: the user-facing products built on top of foundation models, as opposed to the model APIs and infrastructure beneath them. Menlo Ventures puts total generative AI spend at $37 billion in 2025, up from $11.5 billion in 2024, a 3.2x jump in one year (Menlo Ventures, 2025). Applications took more than half of that.

To put the application number in context, Menlo Ventures notes it already represents more than 6% of the entire software market, reached within three years of ChatGPT’s launch (Menlo Ventures, 2025).

The estimate comes from a survey of roughly 495 U.S. enterprise AI decision-makers conducted in November 2025 — C-suite executives, VPs of engineering and product, and technical leaders responsible for AI purchasing (No Jitter, 2025). These are buyers reporting what they bought, not analysts forecasting what they might.

Where the $19 billion went

Menlo Ventures splits application spending into three buckets:

• Horizontal AI — $8.4 billion. Tools that raise productivity across every function. This is the largest slice.
• Departmental AI — $7.3 billion. Tools built for specific roles like software development, sales, or customer success.
• Vertical AI — $3.5 billion. Industry-specific tools, led by healthcare (No Jitter, 2025).

Inside the departmental category, coding tools dominated at $4.2 billion. By comparison, AI for customer success — ticket routing, sentiment analysis, proactive outreach — drew about $630 million, with IT operations and marketing at similar levels (No Jitter, 2025). As the report put it, each of these targets “repetitive workflows where productivity gains are immediate and measurable.”

The common thread across all three buckets is knowledge work. Coding, support, sales, healthcare documentation — these are jobs where the bottleneck is finding the right information and acting on it.

Copilots took most of it. Agents are still small.

The horizontal category tells its own story. Of that $8.4 billion, generative AI assistants and copilots took $7.2 billion — about 86% of horizontal spending (No Jitter, 2025). These are the assistants bundled into communication and collaboration platforms (Microsoft Copilot, Zoom AI Companion, Webex AI Assistant) plus standalone tools like ChatGPT Enterprise and Claude for Work.

Agentic AI — platforms where AI agents act autonomously, maintain memory of interactions, reason over what they know, and orchestrate actions across systems — drew only about $750 million, roughly 10% of horizontal spend (No Jitter, 2025). Named examples include Salesforce Agentforce, Writer, and Glean.

That 86%-to-10% gap is the interesting part. Most of the money is still going to assistants that respond when asked. The smaller, faster-growing edge is agents that act on their own. Both have the same dependency, and most teams discover it the hard way.

The dependency every layer shares: context

A copilot is only as good as the context it can reach. An agent that orchestrates across systems is only as good as the systems it can actually see into. When the underlying knowledge lives in fragmented tools — one fact in a ticket, another in a wiki, a decision buried in chat — the AI answers from a fraction of what the organization knows.

This is the quiet limit on the $19 billion. Buyers paid for tools that do knowledge work. Whether those tools deliver depends on whether the knowledge is connected. A model upgrade does not fix a context gap. A connected knowledge layer does.

A semantic layer — a knowledge graph plus AI search that links people, documents, tickets, and projects across tools — gives both copilots and agents a single place to find and reason over institutional knowledge. It is the substrate the application layer assumes but rarely has.

A concrete example

Consider Vantage Health, a mid-size health system rolling out AI across departments. They buy a coding assistant for their engineering team, a copilot bundled into their collaboration suite, and a vertical clinical-documentation tool — spending that maps neatly onto Menlo Ventures’ three categories.

Three months in, the picture is uneven. The coding assistant works well; it has the codebase in front of it. The general copilot underperforms because it cannot see across the EHR notes, the policy wiki, and the support queue where the real answers live. A care coordinator still asks the copilot “what was last year’s exception for this payer?” and gets a confident, generic non-answer, because the source documents sit in three disconnected systems.

The fix is not a fourth tool or a bigger model. Vantage Health connects its sources through a semantic layer so that one question can traverse the EHR, the wiki, and the queue at once. The same copilot they already paid for now answers from the organization’s actual knowledge. The agentic workflows they pilot next inherit that connected context instead of starting blind. The spend was never the problem. The disconnection was.

Key takeaways

• Menlo Ventures estimates enterprises spent more than $19 billion on generative AI applications in 2025 — over half of a $37 billion total, and more than 6% of the whole software market.
• The money concentrated in the application layer: horizontal tools ($8.4B), departmental tools ($7.3B), and vertical tools ($3.5B), with coding alone at $4.2B.
• Copilots and assistants took 86% of horizontal spend ($7.2B); agentic AI is still early at about $750M (10%).
• Every layer of that spend depends on context. AI applied to knowledge work is only as good as the knowledge it can reach.
• A semantic layer — knowledge graph plus AI search — connects fragmented tools so the copilots and agents enterprises already bought can find and reason over institutional knowledge.