Primary source: Menlo Ventures, 2026: The State of Enterprise AI
Enterprise AI spending just rose more than fivefold in a year, according to Menlo Ventures. For revenue cycle operators, the more consequential finding is that agents are moving from answering questions to executing work—but most AI spending still comes from new budgets rather than displaced labor.
Menlo's estimated annualized enterprise AI deployment spending as of September 2026, up 5.4x year over year. This is not a measure of RCM spending.
Menlo Ventures estimates that enterprise AI deployment spend reached a $176.8 billion annualized run rate by September 2026, compared with $33 billion the year before. The 5.4x increase was modeled using a Morning Consult survey of nearly 400 U.S. enterprise buyers, public disclosures and other research. It is not an audited revenue total, and it is not specific to healthcare revenue cycle.
Healthcare is the largest vertical AI application category at $3.7 billion, up 2.5x. That includes clinical decision support and documentation products, not merely RCM tools. The scale is impressive, but the spending mix matters more than the headline number.
| Menlo finding | Reported figure | Operating implication |
|---|---|---|
| Enterprise AI deployment | $176.8B, up 5.4x | AI has become a meaningful enterprise spending category. |
| Healthcare vertical AI | $3.7B, up 2.5x | Healthcare is the largest vertical AI segment; this is not an RCM market estimate. |
| New budgets | 55% of incremental AI spend | Much of the expansion is additive, rather than already replacing costs. |
| OpenRouter token volume | 80% agentic | Agent workloads now dominate traffic on that platform; not a cross-enterprise share. |
| AI-native applications | 77% of spend | Products built around agent workflows are taking meaningful market share. |
| Builders using agents | 74% | Multi-step execution is becoming an application design norm. |
Menlo says 55% of incremental AI spending came from new budgets. Another 22% came from software budgets. That is a different story from the idea that companies have already replaced large portions of their workforce with AI. A billing organization can spend more on models, connectivity, governance and experimentation while keeping most of its existing staff.
For RCM leaders, the test is not how many employees can access an assistant. It is whether claims move from encounter to payment with fewer human touches, fewer errors and lower fully loaded cost. Productivity gains can precede headcount savings, and a model that handles easy cases may leave the expensive exceptions untouched.
Menlo identifies three conditions that favor successful agents: connected systems in which to act, complete operational context, and outcomes that can be independently checked. Claims, eligibility, denials, prior authorization and payment posting fit that framework in principle. In practice, each is constrained by payer rules, fragmented systems, compliance requirements and delayed feedback.
A denial agent needs the original claim, documentation, payer contract, explanation of benefits and historical appeal outcomes. It must know when it can act, when to wait and when to escalate. A useful result is a verified claim resolution or a correctly documented next step, not a convincing summary of why a claim denied.
These requirements explain why data, payer-specific intelligence, system access and exception routing may become more durable advantages than a generic AI user interface.
The conventional outsourced RCM business built scale by moving repeatable work to specialized teams, often in lower-cost labor markets. Agentic automation introduces another form of arbitrage: completing routine work with materially less human effort. An early operator that lowers production costs while retaining existing contract prices could expand margins.
But that benefit carries its own competitive risk. If software makes routine claim work cheaper for everyone, purchasers can seek lower fees, outcome-based contracts or in-house automation. What looks like a powerful margin tailwind today could become pricing pressure at renewal. Menlo does not measure that RCM-specific trade directly; it is RevCycleAI's interpretation of the operating implications.
Incumbents retain important assets: billing integrations, client trust, payer connections and established contracts. Specialists and new AI-native entrants gain an opening when they can redesign a full workflow around execution, not just assist the person doing it. Both advantages can coexist.
Menlo estimates $34 billion in AI coding application spending. That is software engineering—products such as Claude Code, Codex and Cursor. It is not autonomous diagnosis or procedure coding and should not be cited as the size of the medical coding automation market.
For providers and RCM vendors, the next scorecard should include clean claim rate, straight-through resolution, days in accounts receivable, denials reopened, compliance exceptions and cost per paid claim. For investors, track whether labor leverage improves without a matching fall in revenue per claim—and whether customers reprice contracts when automation savings become visible.
Enterprise AI spending is up 5.4x, yet 55% of incremental spend comes from new budgets. RCM economics change when agents resolve complete workflows at lower cost—not when a company simply adds another AI subscription.
RCAI take: Automating claims work without immediate repricing can create near-term margin expansion. As the cost of the work falls, labor-based pricing faces a longer-run challenge. The most defensible asset is likely to be payer-specific intelligence, connected systems and a trusted client relationship.
Methodology: Menlo's estimates cover enterprise AI broadly; the RCM analysis is RevCycleAI's interpretation, not a claim or forecast by Menlo Ventures. Source: 2026: The State of Enterprise AI, Menlo Ventures.