Nooks Secured $500 Million Without Giving Up a Single Share. The Revenue Agent Platform Covering First Contact to Renewal Is Rewriting Sales AI.

Menlo Ventures' fourth annual State of Enterprise AI report, published November 3, 2026, opens with a number that reframes every conversation about AI disrupting the enterprise software market: $176.8 billion. That is what enterprises spent on AI deployment in the twelve months through September 2026 — not on building AI models, not on research, but on running AI in production. Up 5.4x from approximately $33 billion the prior year.

The number will be reported as a headline. The strategic story is in the second number: $12.6 billion. That is the total amount that moved from traditional SaaS budgets to AI-native vendors. Of $144 billion in incremental AI spending, only 9% came at the direct expense of incumbent software. The rest came from new budget lines, hardware refreshes, and infrastructure reallocation. AI is not primarily replacing enterprise software in 2026. It is expanding the market while incumbents convert existing customers to higher-priced AI tiers.

This piece works through every significant finding in the Menlo data, examines the incumbent survival thesis, analyzes the agent compute story, dissects the coding market concentration, and translates the report's implications into a 2027 product and procurement strategy.

The $177 Billion Number: Deployment, Not Development

The $176.8 billion figure Menlo reports is a deployment number. It represents the money enterprises are paying to use AI in production — inference costs, platform licensing, enterprise contracts with AI-native companies, and the AI-related portion of cloud infrastructure bills. It explicitly excludes spending on foundation model training, research, and academic AI.

This distinction matters because most of the AI spending that attracts public attention — Anthropic's $895 billion in compute commitments disclosed in its revised S-1, OpenAI's $20 billion fundraise — is on the development and infrastructure side. The $176.8 billion is what happens after all that investment, at the enterprise deployment layer where AI actually generates business value.

The infrastructure/application split within that $176.8 billion is itself revealing. Infrastructure accounts for $99.5 billion — 56% of the total — while applications account for $77.3 billion. That infrastructure-heavy split reflects two things: the compute cost of running AI agents at scale is substantial (inference for a complex agent workflow that handles thousands of transactions per day consumes more compute than the same workflow run as a traditional software application), and enterprises are still building the AI infrastructure layer rather than primarily buying AI applications.

The growth rate of 5.4x in a single year is the fastest one-year growth in enterprise software spending in modern history. For comparison, cloud computing — the last major enterprise infrastructure transition — grew approximately 2.5x in its fastest single year (2012-2013). AI deployment spending is growing more than twice as fast, compressing the technology transition timeline dramatically. The enterprises that are behind in AI deployment today are behind in a market that is moving faster than any prior enterprise technology wave.

Menlo's report draws on a survey of nearly 400 U.S. enterprise decision-makers responsible for more than $4 billion in AI spending, fielded by Morning Consult from September 8 through September 26, 2026. That is a credible methodology for a market sizing study of this kind, though the figures are Menlo's own estimates rather than audited financial data. Readers should also note that Menlo Ventures is an investor in Anthropic, one of the major beneficiaries of the market share data in this report.

Where the Money Actually Came From

The budget source breakdown is the most strategically important finding in the report for CFOs, CIOs, and product leaders evaluating their AI investment strategy.

The 55% new-budget figure tells you that most AI spending is not substituting for prior technology investment — it is incremental enterprise technology spending. CFOs are approving new AI budget lines rather than canceling existing SaaS contracts to fund them. This has two implications.

First, the ROI case for AI is being made on expansion terms, not displacement terms. Enterprises are not telling their AI vendors "we cut Salesforce to fund you." They are saying "we created a new budget for AI because the productivity case was compelling enough to justify incremental spend." That is a weaker version of the ROI case than displacement — it means AI has to justify its spend as a productivity multiplier, not as a cost replacement.

Second, the 22% from software budgets represents real SaaS displacement, but at a small scale relative to the total software market. Twenty-two percent of the incremental $144 billion is approximately $32 billion in SaaS spending that moved to AI — but that $32 billion was spread across all AI vendors, not concentrated in AI-native startups. Most of it went to incumbents' AI products: Microsoft's Copilot tiers replacing standard Office 365 licenses, Salesforce's Agentforce tiers replacing standard Sales Cloud licenses, Workday's AI products replacing standard HRIS seats.

83% of CFOs Are Increasing AI Budgets While Just 31% Are Satisfied With the Outcomes — that satisfaction gap in CFO surveys is consistent with the new-budget-creation story. CFOs are creating new AI budgets under pressure from business unit leaders and board expectations, but the ROI measurement infrastructure for most AI tools is not mature enough to show clear returns at the portfolio level. The budget growth is real. The satisfaction is lagging.

Why Incumbents Are Still Winning: The Systems-of-Record Defense

The most counterintuitive finding in the Menlo report is the degree to which enterprise software incumbents are holding their positions in core categories. Incumbents hold 91% of IT spending, 88% of data science spending, and 68% of HR software spending. Only 9% of enterprise AI spending — the $12.6 billion — moved to AI-native vendors.

This is not what the AI disruption narrative predicted. The standard prediction, articulated by many AI-native startups and their investors since 2023, was that AI-native applications would displace incumbents by offering fundamentally better experiences: more intelligent, more automated, more integrated with LLM capabilities. Three years into the enterprise AI wave, that displacement is happening more slowly than predicted.

The structural reason is switching cost asymmetry. Salesforce is not just a CRM — it is a system of record that holds ten or fifteen years of customer interaction data, is integrated with every adjacent sales tool and marketing automation platform in the enterprise stack, and is embedded in the workflows of every person who touches revenue. Replacing Salesforce requires migrating that data, rebuilding those integrations, retraining hundreds or thousands of users, and accepting months of reduced sales team effectiveness during the transition. An AI-native CRM would need to be not just better at AI features — it would need to be better by enough to justify the transition cost and risk.

The incumbent response has also been more effective than predicted. Salesforce's Agentforce, launched in October 2025, is an AI agent platform built directly on the CRM data layer — it offers AI-native workflows to customers who already have their data in Salesforce, without requiring migration. Workday's AI platform does the same for HRIS. ServiceNow's AI features extend its ITSM platform. These incumbents are capturing the AI spend from their existing customer base rather than ceding it to startups.

The 68% incumbent share in HR — the lowest category the report covers — suggests that HR software is the most contested category and the one where AI-native startups are making the most headway. HR workflows are more modular and less data-intensive than CRM or ITSM, which lowers switching costs and makes the AI-native value proposition more compelling relative to transition risk.

Agents Are Now the Majority Compute Consumer

The agent compute story in the Menlo report is a tipping point disclosure. Agents now generate approximately 80% of all tokens on OpenRouter, the aggregated AI usage tracking platform, amounting to roughly 130.9 trillion tokens per week — approximately seven times the volume of human-driven usage. For the first time in the history of enterprise AI, the majority of AI compute is consumed not by humans interacting with AI but by AI interacting with AI.

This is not a semantic distinction. Human-driven AI usage is primarily conversational or document-based: a person asks a question, the model responds, the exchange ends. The latency tolerance is seconds to minutes, the context windows are moderate, and the usage is intermittent — a person uses the model when they have a task, then stops. Agent-driven AI usage is fundamentally different: an agent might execute hundreds of tool calls in the course of completing a complex workflow, consuming large amounts of compute per task, running continuously rather than intermittently, and potentially spawning sub-agents that run in parallel.

The 7x ratio of agent to human compute consumption is the signal that enterprise AI has crossed a threshold. When agents consume seven times the tokens that humans do, the infrastructure architecture for enterprise AI changes: you are no longer primarily optimizing for human query latency and cost, you are optimizing for agent workflow throughput, reliability, and cost at scale. The economics of inference — cost per token, latency per call, context window pricing — matter more for agent workflows than they did for conversational AI.

Gartner Forecasts That 40% of Agentic AI Projects Will Be Canceled by 2027 — the cancellation risk is partly driven by compute cost surprises. Enterprises that designed their AI agent deployments for human-usage economics and are now seeing agent-usage economics are facing budget overruns that are driving cancellation decisions. The 80% agent token share in the Menlo data is a warning for every enterprise that approved an AI agent budget based on conversational AI cost assumptions: the actual compute bill will be higher, potentially much higher, than the original estimate.

The Coding Market: When Two Providers Take 70% of a $34 Billion Category

The enterprise AI coding market data is the most concentrated market share story in the Menlo report. Claude Code and OpenAI's Codex together account for $24 billion — 70% — of a $34 billion enterprise AI coding market. No other enterprise software category of comparable size has two vendors controlling 70% of market spend.

GitHub Copilot, which dominated the AI coding tool category in 2023 and 2024, has ceded significant market share. The shift reflects a capability gap: Claude Code and Codex have demonstrated significantly superior performance on complex coding tasks — multi-file refactoring, context-window-intensive codebases, test generation, debugging — compared to Copilot's original inline completion model. Enterprise developers doing real work on real codebases care about complex task performance, and the performance gap has driven adoption decisions.

The concentration matters for three reasons. First, pricing power: when two providers control 70% of a $34 billion market, they have substantial ability to raise prices as competition fails to close the capability gap. Enterprise developers who have customized their entire coding workflow around Claude Code's capabilities face high switching costs — switching requires retraining, reconfiguring integrations with IDEs and CI/CD pipelines, and accepting a capability regression during the transition period.

Second, data moats: the coding tools with the highest market share are accumulating the most training data from enterprise codebases. This creates a compounding advantage — better models attract more users, more users generate more training signal, better training signal produces better models. AI-native coding tools that do not achieve scale before the data moat solidifies face an increasingly difficult catch-up curve.

Third, security risk concentration: 70% of enterprise coding activity flowing through two providers creates a supply chain security risk that few enterprise CISOs have fully modeled. A vulnerability in Claude Code or Codex's code generation layer could affect a majority of enterprise code output simultaneously.

The Model Market Share Shift: Anthropic's 15-Point Surge

The model market share data in the Menlo report shows Anthropic's enterprise API share rising from 40% to 55% — a 15-point gain in a single year — while Google's fell from 21% to 9%. OpenAI grew from 27% to 31%. The combined frontier lab share rose to approximately 69% of enterprise deployment dollars.

Anthropic's 15-point gain is the most dramatic single-year market share movement the report documents, and it deserves contextual reading. Menlo Ventures is an investor in Anthropic, which the report discloses. That relationship does not invalidate the data — Menlo surveys real enterprise decision-makers, not Anthropic employees — but it means the Anthropic figures in this report should be verified against additional data sources before being treated as definitive. Developer surveys from other sources have consistently shown Anthropic's coding tool dominance, which is consistent with a rising API share, but the magnitude of 55% market share has not been independently verified as of publication.

Google's decline from 21% to 9% is the most unexpected finding in the report. Google has significant structural advantages in enterprise AI: deep relationships through Google Cloud, Workspace integration that embeds Gemini in tools employees already use, and TPU infrastructure that could support lower inference pricing. The 12-point API share decline despite those advantages suggests that raw model capability — the benchmark on which Gemini has trailed Claude in complex reasoning tasks — matters more than distribution and integration at the enterprise API layer.

The combined 69% frontier lab share of deployment dollars confirms what the Anthropic S-1 data disclosed in November 2026: the AI value chain is concentrating at the foundation model layer. Infrastructure spending, the larger half of the $176.8 billion total, flows primarily to compute providers and hyperscalers. Application spending is split between frontier labs' own products (like Claude Code and ChatGPT Enterprise) and AI-native application vendors who build on top of frontier models. The enterprise AI market is not a winner-take-all market, but it is a market where the foundation model providers are capturing a disproportionate share of the value.

The SaaS Slowdown Signal and What It Actually Means

Menlo estimates that AI has slowed enterprise application SaaS growth from approximately 13% annually to approximately 11%, on a market exceeding $600 billion. That 200-basis-point deceleration represents approximately $12 billion in annual growth that the traditional SaaS market is not capturing.

For public market investors and SaaS company leaders, this is the number to watch. The SaaS market has grown at approximately 15-20% annually for a decade. The deceleration to 11% — if it persists and accelerates — is the first observable impact of AI on the structural growth rate of enterprise software. It is not yet a dramatic reversal, but it is directional evidence that AI is beginning to constrain SaaS expansion.

The mechanism is budget competition, not displacement. SaaS companies are not primarily losing existing customers to AI-native competitors. They are losing incremental budget growth — the new seats, new modules, and expansion deals that would have been approved in prior years are now being partially deferred as enterprises allocate incremental budget to AI tools instead. A sales team that would have added 20 new Salesforce seats is adding 10 seats and allocating the other 10 seats' budget equivalent to Nooks or another AI sales tool.

This budget competition will intensify before it resolves. The enterprise software market is not yet making hard trade-off decisions between incumbent SaaS and AI-native tools — it is adding AI tools on top of existing stacks while maintaining incumbent licenses. As AI tools mature and demonstrate ROI clearly enough to justify displacement rather than addition, the budget competition will shift to displacement decisions. That transition is likely to begin in 2027 for the highest-ROI categories, particularly those where AI-native tools have demonstrated the clearest performance advantage over incumbent features.

The 2027 Product Strategy Playbook

The Menlo 2026 State of Enterprise AI report implies a specific set of strategic decisions for enterprise product leaders, procurement teams, and startups building for the enterprise market. The findings — incumbent survival, budget expansion rather than displacement, agent compute dominance, and coding market concentration — point toward a coherent 2027 framework.

1. Measure AI ROI at the category level, not the tool level. The 83% of CFOs increasing AI budgets while only 31% expressing satisfaction with outcomes reflects a measurement problem at the portfolio level. Build a category-level ROI dashboard that shows the return on the full AI investment in each functional category — not just per-tool metrics that cannot be aggregated. The enterprises that resolve the measurement gap will have higher confidence in their AI investment decisions and will accelerate toward platforms that demonstrate clear attribution.

2. Audit the agent compute cost assumptions in every existing AI deployment. The 80% agent token share means that enterprises running agentic workflows built on conversational AI cost assumptions are likely underestimating their actual compute cost. Audit every production AI agent deployment for actual token consumption versus projected, and model the cost trajectory as agent usage scales. Identify deployments where the agent compute cost will exceed the economic value generated at current scale.

3. Treat systems-of-record AI features as the upgrade path of least resistance. For categories where incumbents hold 88-91% share — IT and data science — the fastest path to AI value is not evaluating AI-native alternatives. It is working with existing incumbent vendors to activate their AI features. Salesforce Agentforce, Workday AI, and ServiceNow AI are available to existing customers today. The activation barrier is configuration and change management, not procurement or migration.

4. Rationalize point tools in categories where platform consolidation is happening. The sales AI category consolidation story illustrated by Nooks is not unique to sales. Customer success AI, marketing AI, and HR AI are all experiencing the same fragmentation → consolidation pattern. In each category, identify whether a platform that covers the full workflow is now viable — if it is, the ROI case for consolidating from a five-tool stack to a single platform is likely compelling on integration cost savings alone.

5. Build AI strategy around data moats, not feature parity. The coding market concentration demonstrates that feature differentiation at the foundation model layer translates directly into enterprise market share, and that market share compounds through data accumulation. For enterprises selecting AI tools, prefer tools that are accumulating training signal from your domain — your codebase, your customer data, your document workflows — over tools that compete primarily on feature parity with alternatives. The tools building the best domain-specific models from your data today will have the best domain-specific models from your data in 2029.

Takeaway: Menlo Ventures' 2026 State of Enterprise AI report delivers a counterintuitive conclusion for a year in which every AI headline was about disruption: incumbents are winning the enterprise AI transition in 2026. AI spent $177 billion in enterprise deployment, but only $12.6 billion of that displaced existing SaaS vendors. The rest created new budget lines, refreshed hardware, and paid the compute bills of AI agents consuming tokens at seven times the rate of humans. The strategic read for 2027 is not "which incumbent is about to be disrupted?" — it is "which AI-native vendors have solved the ROI attribution problem clearly enough to justify a displacement decision rather than an addition decision?" That is a much smaller set than the AI-native startup market implies. The enterprises that identify that set early, test them with rigorous attribution experiments, and build data moats around the tools that perform will capture the compounding advantage of early deployment. The enterprises that add AI tools without clear attribution experiments will continue to increase budgets while expressing dissatisfaction with outcomes — a pattern that the Menlo data suggests is the dominant enterprise AI behavior in 2026.