Anthropic Captures 40% of Enterprise LLM Spend as GenAI Hits $37B in 2025

Anthropic Takes Market Lead as Enterprise Spend Reaches $37B

Anthropic now captures an estimated 40% of enterprise large language model spending, according to Menlo Ventures' 2025 State of Generative AI in the Enterprise report. The figure represents a sharp climb from 24% in 2024 and 12% in 2023, reshaping vendor selection strategy for organizations that assumed OpenAI would remain the default choice.

The report pegs total enterprise GenAI spend at $37B in 2025, consuming roughly 6% of a $300B software budget pool. Menlo calls it "the fastest scaling software category in history." For CIOs, the implication is clear: GenAI is no longer an experimental line item. It requires explicit budget allocation, multi-year planning, and vendor risk assessment.

Coding Tools Drive $4B in Spend, Deliver Measurable Time Savings

AI coding assistants account for $4B of the $37B total—approximately 11% of GenAI spend but 55% of departmental AI budgets. Products in this category include GitHub Copilot, Amazon Q Developer, Google Code Assist, and Anthropic's coding agents.

OpenAI survey data cited in the report shows ChatGPT Enterprise users in engineering and data science roles save 60 to 80 minutes per day. Overall, 73% of engineers report faster code delivery, and 87% of IT workers report faster issue resolution. These are not marginal gains. They translate to measurable headcount leverage in functions where labor costs are high and hiring timelines are long.

The productivity data provides the ROI justification procurement teams require. When a coding copilot costs $20 to $40 per seat per month and delivers an hour of time savings daily, the unit economics support expansion even in cost-constrained environments.

Multi-Model Strategy Becomes Requirement, Not Option

Anthropic's rise to 40% market share signals that enterprises are diversifying away from single-vendor LLM dependencies. The shift from 12% to 40% in two years indicates buyers are running competitive evaluations and selecting models based on task-specific performance, cost, and compliance posture.

RFPs now increasingly demand multi-model support and model routing capabilities. Vendors that lock buyers into a single LLM provider face disadvantage. Platform buyers must plan for architectures that route prompts to different models based on workload—whether that means Claude for long-context tasks, GPT-4 for reasoning, or open-source models for high-volume, low-cost inference.

Menlo's data also shows enterprises building more GenAI applications in-house. The ratio shifted from 80% vendor-purchased and 20% internally built in 2023 to roughly 53% vendor and 47% in-house in 2024. This means platform and infrastructure buyers need API orchestration layers, developer tooling, and governance frameworks that support internal development, not just off-the-shelf SaaS tools.

Usage Growth Creates Cost Management Risk

ChatGPT Enterprise seats grew 900% year-over-year, and weekly enterprise messages increased roughly 800%. Custom GPT usage climbed 19x, now accounting for 20% of all enterprise messages. Most concerning for finance teams: reasoning token consumption per organization jumped 320x in 12 months.

These usage patterns create unpredictable, usage-based cost exposure. Enterprises running GenAI at scale report monthly bills that swing by 30% to 50% based on user behavior they cannot forecast. Procurement must demand hard usage caps, real-time dashboards, and predictable pricing tiers. Contracts that rely solely on per-seat pricing without usage controls expose buyers to budget overruns.

The report notes enterprise reluctance around Chinese open-source LLMs, which account for only 10% of enterprise open-source model usage. This reflects compliance, data sovereignty, and geopolitical risk considerations that now factor into model selection alongside technical performance.

Application Spend Now Exceeds Infrastructure Spend

Menlo's breakdown shows $19B in spend on AI applications versus $18B on infrastructure. This means application vendors—Salesforce, ServiceNow, Workday, Adobe, SAP, and vertical SaaS providers—collectively capture more GenAI dollars than raw model and infrastructure providers.

For buyers, this complicates vendor consolidation strategy. GenAI spend is distributed across dozens of application vendors, each embedding models into workflows. IT leaders need visibility into which applications justify incremental AI pricing, which duplicate functionality, and where internal development can replace vendor markup.

What to Watch

Track whether Anthropic's market share continues to climb or stabilizes as OpenAI releases new models. Monitor vendor pricing responses to multi-model competition—expect pressure on per-token costs and introduction of consumption commitments.

Watch for enterprises publishing internal ROI benchmarks beyond productivity surveys. The next phase of GenAI buying will demand unit-economics justification at the workflow level, not just aggregate time savings. If your organization lacks usage telemetry that ties GenAI costs to specific outputs—lines of code shipped, tickets resolved, deals closed—that becomes the next procurement priority.