Legora at $10B, Harvey at $15.5B: The Legal AI Valuation Sprint Is Outrunning Revenue. What Enterprise Legal Teams Must Know Before Signing Multi-Year Contracts.
Menlo Ventures' 2025 mid-year data shows enterprise LLM spend doubled to $8.4B in six months, and Anthropic owns the largest slice. OpenAI fell from 50% to 27%. Understanding why the flip happened tells you how the next enterprise AI procurement cycle will play out.
In mid-2025, Menlo Ventures surveyed 150 technical leaders across AI startups and large enterprises and found something the AI industry's consumer-facing narrative had obscured: in the market that actually pays frontier model providers for API access, Anthropic is winning — and winning decisively.
Anthropic holds 40% of enterprise LLM API spend. OpenAI, which held 50% in 2023 and effectively invented the enterprise LLM market through ChatGPT's rapid enterprise adoption, now holds 27%. Google, long considered the structural long-term threat to both, holds 21%.
Those three numbers — Anthropic 40%, OpenAI 27%, Google 21% — represent one of the most significant market share shifts in enterprise software in the past decade. Enterprise LLM API spend reached $8.4 billion in mid-2025, up from $3.5 billion in late 2024 — a market that doubled in approximately six months. At 40% of $8.4 billion, Anthropic's implied API revenue run rate from enterprise customers alone is approximately $3.4 billion annually.
For enterprise AI buyers, the market share data is a directional signal: the technical leaders who have been through the most rigorous API selection processes — the people responsible for LLM vendor decisions in organizations that are actually running production workloads at scale — are choosing Claude at a rate that has overtaken every competitor. Understanding why that happened, and what it means for the competitive landscape going forward, is the procurement intelligence that matters now.
The Market Share Timeline: Two Years of Dramatic Reversal
The speed of Anthropic's rise from 12% to 40% in roughly 24 months requires some context to appreciate fully.
| Vendor | Enterprise LLM API Share (2023) | Enterprise LLM API Share (Mid-2025) | Change |
|---|---|---|---|
| Anthropic | 12% | 40% | +28pp |
| OpenAI | 50% | 27% | -23pp |
| 7% | 21% | +14pp | |
| Other | 31% | 12% | -19pp |
The table tells three distinct stories. OpenAI built a 50% market share in 2023 primarily through brand recognition and first-mover advantage: ChatGPT's consumer launch drove enterprise curiosity, and many organizations signed GPT-4 API contracts before serious competitive alternatives existed. That structural advantage eroded as the market matured and buyers switched based on API performance benchmarks rather than brand familiarity.
Google's gain from 7% to 21% reflects its distribution advantage — enterprise teams running infrastructure on Google Cloud have lower-friction paths to Gemini API access through Vertex AI than to standalone Anthropic API accounts. Google's growth is real, but it is partially attributable to bundling and existing enterprise relationships rather than purely competitive API selection.
Anthropic's gain from 12% to 40% is the most structurally significant. It happened in a market where enterprise buyers are increasingly sophisticated — running side-by-side benchmark comparisons, evaluating on task-specific performance metrics, and making decisions based on cost-per-output-quality rather than default vendor relationships. Anthropic's share gain in that environment, against an OpenAI brand that started with a 4x advantage, tells you something real about where the capability and value proposition landed.
The Coding Market: Where Anthropic's Dominance Is Most Pronounced
The enterprise LLM market is not monolithic. Different use cases show different vendor distributions, and the divergence is instructive.
In the enterprise coding market — which covers API calls for code generation, code review, debugging, and developer workflow automation — Anthropic holds 54% of enterprise spend. OpenAI holds 21%. Google holds 11%. The remaining 14% is distributed across open-source models and other providers.
The coding market is significant because it is both the largest individual enterprise AI use case and the one with the most rigorous and measurable performance requirements. Developers can directly observe whether AI-generated code works, whether it handles edge cases correctly, and whether it produces maintainable output. The feedback loop in coding is faster and more objective than in most other enterprise AI use cases, which means the vendor selection data in coding represents the most technically informed procurement decisions in the enterprise AI market.
Anthropic's 54% coding market share implies that in the category most demanding of model capability, enterprises are choosing Claude by a margin wider than its overall API market share. This has two implications:
First, Claude's coding performance benchmark is real and decision-driving. The $2B+ ARR that Cursor achieved before hiring a significant enterprise sales team is downstream of the same underlying model quality that drives Anthropic's direct API coding share. Enterprise developers have switched coding vendors based on direct productivity measurement — not on marketing claims.
Second, the coding market tends to foreshadow the broader enterprise API market. Developer tooling is where AI adoption typically enters enterprise organizations before expanding to legal, finance, marketing, and other functional areas. Anthropic's dominance in coding today positions it to expand into those adjacent markets as enterprise AI footprints grow. The pattern Signal documented in the IBM-OpenAI enterprise consulting partnership shows how coding tool adoption creates an organizational beachhead that expands through adjacent professional workflows — Anthropic's coding share advantage creates the same expansion vector.
Why Claude Won Enterprise: Three Structural Advantages
The market share shift from OpenAI to Anthropic in enterprise API procurement did not happen because of a single feature release or a single quarter of better performance. It happened because of three structural advantages that accumulated over 24 months.
1. Safety and compliance positioning as a procurement filter. Enterprise procurement processes include AI vendor risk assessments. For regulated industries — financial services, healthcare, legal, government — those risk assessments are intensive, multi-stakeholder reviews that evaluate vendor AI safety frameworks, data handling commitments, model behavior documentation, and incident response processes. Anthropic's Constitutional AI approach and its published Responsible Scaling Policy created documentation artifacts that enterprise risk and compliance teams could evaluate. When enterprise risk teams ask "what is your vendor's AI safety framework?" a detailed published methodology outperforms a general assurance. OpenAI has improved its enterprise documentation significantly since 2023, but Anthropic entered the enterprise market with safety documentation as a primary differentiator rather than as a compliance afterthought.
2. Model performance on enterprise-relevant benchmarks, not consumer benchmarks. The benchmarks that drove early narrative about OpenAI's GPT-4 — MMLU, HumanEval in its standard form, general-purpose reasoning — don't map directly to enterprise workflow performance. Enterprise buyers care about long-document summarization accuracy, instruction-following consistency across complex multi-step prompts, structured output reliability, and citation accuracy in research synthesis. On these task-specific benchmarks, Claude's architecture optimized for instruction-following and long-context performance showed advantages that compound in production workflows where the same prompt runs millions of times and where output quality variance creates real downstream costs.
3. Pricing and cost-per-quality economics at production scale. At the token volumes that enterprise production workloads generate, per-token pricing differences of even 10-20% translate into significant annual cost differences. Anthropic's pricing structure for Claude, particularly for the Claude Haiku tier (optimized for high-volume applications) and Claude Sonnet (the mid-tier production model), offered better cost-per-output-quality ratios for many enterprise use cases than equivalent OpenAI tiers. As all providers cut inference prices in the AI price war of 2025, the relative cost-per-quality ranking became more stable, and Anthropic's position in that ranking sustained its enterprise adoption momentum.
OpenAI's Position: What the 27% Share Tells You
OpenAI's decline from 50% to 27% of enterprise LLM API spend is less dramatic than it appears when you account for the absolute market growth. The total enterprise LLM market doubled from $3.5B to $8.4B during the same period. OpenAI's 50% share of $3.5B implied approximately $1.75B in enterprise API revenue. Its 27% share of $8.4B implies approximately $2.27B — a revenue increase despite a market share decline.
This is a critical distinction. OpenAI has not lost enterprise customers in aggregate; it has grown enterprise API revenue while failing to keep pace with the overall market growth and Anthropic's gains. The share decline reflects the pace of new enterprise customer acquisition going to Anthropic at a higher rate than to OpenAI, and existing OpenAI customers diversifying their API spend across providers rather than consolidating on GPT models.
OpenAI's confidential S-1 filing targeting a $1 trillion IPO valuation describes a company generating approximately $25 billion in annualized revenue across consumer, enterprise, and API segments — which means enterprise API is a significant but not dominant share of total OpenAI revenue. The consumer ChatGPT subscription base and the enterprise ChatGPT Workspace deployments are larger absolute revenue contributors than pure API. This creates a strategic split: OpenAI is simultaneously a consumer AI company, an enterprise API company, and an AI infrastructure company. Managing those three distinct business models — with different product, pricing, and compliance requirements — is organizationally complex in ways that Anthropic's more focused enterprise API strategy is not.
Google's 21%: The Distribution Threat That Could Accelerate
Google's rise from 7% to 21% of enterprise LLM API spend in two years is the market development that most concerns Anthropic's long-term position. The concern is structural: Google's growth is not primarily about model quality catching up to Claude (though Gemini Ultra has closed capability gaps significantly), but about distribution leverage that Anthropic cannot replicate.
Google's enterprise AI distribution advantages include:
- Google Cloud's existing enterprise relationships: Enterprises already running workloads on GCP have procurement relationships, committed spend frameworks, and infrastructure integration that make adding Vertex AI model access low-friction. The incremental procurement step for a GCP customer evaluating Gemini is vastly simpler than a net-new Anthropic API account setup.
- Google Workspace AI integration: Gemini's embedding in Gmail, Docs, Sheets, and Meet puts AI capability in front of hundreds of millions of existing commercial users through application interfaces rather than API calls. This drives adoption that may not appear directly in enterprise LLM API spend data but creates organizational familiarity with Gemini that influences API procurement decisions.
- Vertex AI as the managed platform layer: Large enterprises that want managed AI deployment — fine-tuning, evaluation, deployment, monitoring — in a single cloud-native environment have Vertex AI as an integrated option that neither Anthropic nor OpenAI can match with cloud-agnostic alternatives.
The question for Anthropic is whether its model quality advantage can sustain a 19-percentage-point lead over Google as Google's distribution leverage and model quality both improve. The Menlo Ventures data is from mid-2025; the 2026 data, when published, will be the first real test of whether the Anthropic share peak has been reached or whether it is continuing to grow.
The $8.4 Billion Market: Where Enterprise LLM Spend Goes
Understanding the $8.4 billion enterprise LLM API market requires breaking it down by use case category.
| Use Case | Estimated Share of Enterprise LLM Spend | Why It Drives High API Volume |
|---|---|---|
| Coding and developer tools | ~35% ($2.9B) | High daily active usage, long context windows, structured output requirements |
| Document analysis and synthesis | ~25% ($2.1B) | Long-context processing, enterprise content volumes, compliance workflows |
| Customer-facing AI (chatbots, support) | ~18% ($1.5B) | High transaction volume, latency sensitivity, real-time inference |
| Internal knowledge and search | ~12% ($1.0B) | Enterprise knowledge base retrieval, RAG pipelines, agent orchestration |
| Code review and security | ~10% ($0.8B) | Continuous integration integration, automated review cycles |
The coding category's 35% share of enterprise LLM API spend — combined with Anthropic's 54% share of that category — explains a disproportionate amount of Anthropic's overall market position. If coding is $2.9B of the $8.4B market, and Anthropic holds 54%, that's approximately $1.57B from coding alone. Divided by the $3.4B implied by 40% overall market share, coding represents roughly 46% of Anthropic's total enterprise API revenue. This concentration creates both strength (deep coding market penetration creates durable enterprise relationships) and risk (any competitive degradation in Claude's coding benchmark performance has outsized revenue impact).
Menlo Ventures' full-year 2025 State of Generative AI report placed total enterprise generative AI investment at $37 billion in 2025 — a 3.2x increase from $11.5 billion in 2024. The $8.4 billion enterprise LLM API spend is a subset of that $37 billion, with the remainder flowing to AI-native applications, enterprise tooling, services, and consulting. The overall market growth trajectory suggests that enterprise LLM API spend will continue to grow faster than the broader enterprise software market as more workloads shift from traditional software to LLM-based processing.
What This Means for Enterprise Buyers Now
For enterprise teams making LLM vendor decisions in 2026, the market share data carries practical implications.
The default vendor assumption has shifted. In 2022 and 2023, the default assumption for enterprise LLM API procurement was OpenAI — the incumbent with the broadest name recognition, the most extensive enterprise documentation, and the largest developer community. That default has changed. Technical leaders surveyed by Menlo Ventures are choosing Claude at a 40% rate versus OpenAI's 27%. If your organization is still defaulting to GPT models without running a current side-by-side performance evaluation against Claude on your specific use cases, you are making a procurement decision on 2023 assumptions in a 2026 market.
The multi-vendor architecture is now standard, not advanced. The "other" category in enterprise LLM vendor distribution has shrunk from 31% to 12% — but 88% of enterprise LLM spend is now concentrated in three vendors rather than one. Enterprise teams should architect their LLM infrastructure to route different use cases to different providers based on benchmark performance: Claude for coding and long-document tasks, GPT-5 for use cases where OpenAI's specific fine-tuning capabilities or integrations have advantages, Gemini for GCP-native deployments. The multi-vendor architecture is not a hedge against vendor lock-in alone — it is the configuration that maximizes benchmark-matched performance across heterogeneous enterprise workloads.
Anthropic's enterprise dominance is now a procurement negotiation fact. The credit war between Anthropic and OpenAI competing for early-stage startup adoption reflects both companies' awareness of how important enterprise API lock-in is at the early adoption stage. Enterprise teams renewing or expanding Claude API contracts in 2026 are negotiating with a vendor that holds 40% market share and has pricing power to match. Volume commitments, SLA terms, and data handling agreements should be negotiated explicitly — with the awareness that Anthropic's revenue dependence on enterprise retention is the leverage the buyer holds.
The Anthropic IPO: Market Share as the Valuation Foundation
Anthropic's $2 trillion IPO target is premised on $100 billion in annualized revenue — a target that requires substantial growth from mid-2025 levels. The Menlo Ventures market share data provides the first concrete external validation of where Anthropic's revenue is actually anchored.
At 40% of an $8.4B enterprise LLM API market in mid-2025, Anthropic's enterprise API revenue was approximately $3.4 billion annually — a small fraction of the $100 billion IPO revenue target. The path from $3.4B to $100B requires either dramatic market expansion (the enterprise LLM API market growing more than 10x from mid-2025 levels) or Anthropic expanding significantly beyond pure API revenue into AI products, services, and consumer subscriptions. Anthropic is pursuing all of the above — and the legal AI valuations covered in the Legora/Harvey analysis illustrate how vertical AI applications built on top of frontier APIs can command premium valuations that dwarf the API infrastructure layer.
For investors and enterprise buyers alike, the Menlo Ventures data establishes Anthropic's market position credibility in a way that Anthropic's own claims cannot. An independent survey of technical leaders showing 40% enterprise API market share — versus a market share data point that could be cherry-picked or defined favorably — is the foundation on which the IPO narrative rests.
Takeaway: Anthropic's 40% enterprise LLM API market share is not a temporary survey artifact. It reflects 24 months of accumulated procurement decisions by the most technically informed buyers in enterprise AI — the engineering leaders who run production workloads at scale and measure vendor performance on task-specific benchmarks rather than general capability claims. The reversal from OpenAI's 50% to Anthropic's 40% happened because the enterprise AI market matured from brand-familiarity procurement to benchmark-driven selection. For enterprise buyers, the implication is twofold: first, revisit any LLM vendor decisions made before 2025 that have not been benchmarked against Claude on current production task profiles; second, negotiate your Anthropic contracts with awareness that the vendor's 40% market share reflects revenue dependence on enterprise retention that you can use as procurement leverage. The market share data is not a reason to default to Claude — it is a reason to benchmark seriously and negotiate from knowledge of where the market has actually moved.
Frequently Asked Questions
How did Anthropic capture 40% of enterprise LLM API spend?
Anthropic's rise from 12% of enterprise LLM API spend in 2023 to 40% by mid-2025 was driven by three converging factors. First, Claude's coding capabilities outperformed competitors: Anthropic now holds 54% of the enterprise coding market specifically, which means the category with the highest daily active usage and clearest productivity ROI flows disproportionately through Claude. Second, Anthropic's safety and compliance positioning resonated with enterprise procurement teams who needed AI vendors with demonstrable risk management frameworks — a structural advantage in a market where enterprise buyers increasingly require vendor risk assessments. Third, Claude's context window, instruction-following accuracy, and low hallucination rate in structured document workflows outperformed GPT-4 on the specific benchmarks enterprise teams actually care about — contract review, long-document synthesis, code generation — not on the consumer chatbot metrics that dominated early adoption stories. The market share shift happened fastest among technical leaders who switched API vendors based on benchmark performance comparisons rather than brand familiarity, which is why the Menlo Ventures survey of 150 technical leaders showed results so skewed relative to consumer-facing data.
Why has OpenAI's enterprise API market share fallen from 50% to 27%?
OpenAI's decline from 50% of enterprise LLM API spend in 2023 to 27% by mid-2025 reflects several structural challenges. First, the company's initial enterprise advantage was brand recognition and first-mover position, both of which depreciate as the market matures and buyers switch based on API performance rather than name recognition. Second, OpenAI's pricing strategy — historically higher per-token costs relative to Claude and some open-source alternatives — created cost headwinds for teams scaling to production volume. Third, OpenAI's product strategy split attention between consumer ChatGPT, which is growing to hundreds of millions of users, and enterprise API customers, whose needs around compliance, customization, and reliability sometimes conflicted with consumer-facing development priorities. Fourth, Anthropic's sustained focus on enterprise use cases — particularly coding, document analysis, and agentic workflows — produced measurable benchmark improvements in the specific task categories enterprise buyers care about. OpenAI still holds 21% of the enterprise coding market specifically, but the overall enterprise API share reflects a market where technical leaders are choosing based on capability-per-dollar benchmarks rather than default vendor relationships.
What is the total size of the enterprise LLM API market in 2026?
Enterprise LLM API spend reached $8.4 billion by mid-2025, according to Menlo Ventures' 2025 Mid-Year LLM Market Update — more than double the $3.5 billion in late 2024, meaning the market roughly doubled in approximately six months. Menlo Ventures' full-year 2025 State of Generative AI in the Enterprise report placed total enterprise generative AI investment at $37 billion, which includes LLM API spend alongside AI-native application development, enterprise tooling, and consulting. The $8.4 billion figure represents direct API infrastructure spend — the money flowing to AI model providers for inference — and is the segment most directly relevant to the market share analysis. For context, total global AI software spending is projected at $184 billion in 2026, and worldwide SaaS revenue is forecasted to reach $488.5 billion. Enterprise LLM API spend represents a significant but still-early fraction of total enterprise software spending, suggesting the market has substantial remaining growth potential as AI penetrates categories currently running on traditional software.
How does Google's enterprise AI position compare to Anthropic and OpenAI?
Google has grown from 7% of enterprise LLM API spend in 2023 to 21% by mid-2025 — a tripling of market share over two years. Google's growth has been faster than OpenAI's retention on a percentage basis, but Google's absolute position still trails Anthropic by a substantial margin. Google's enterprise AI strategy differs structurally from Anthropic's and OpenAI's: Google operates Gemini both as a standalone API product and as an embedded component of Google Workspace, which means a portion of enterprise Gemini usage is embedded in productivity applications rather than direct API calls. In the coding-specific market, where the Menlo Ventures data shows Anthropic at 54%, Google holds 11%. In the broader enterprise API market, Google's 21% share is growing from a base of deep enterprise relationships through Google Cloud, which gives it a distribution advantage — enterprise teams already running infrastructure on GCP have lower friction paths to Vertex AI and Gemini API access than to standalone Anthropic or OpenAI API accounts. The long-term question is whether Google's distribution advantage converts to API share faster than Anthropic's benchmark advantage sustains its current lead.
What are the implications for enterprise AI buyers of a more concentrated LLM vendor market?
The concentration of enterprise LLM API spend across three vendors — Anthropic at 40%, OpenAI at 27%, Google at 21% — creates both stability and lock-in risk for enterprise buyers. The stability comes from vendor concentration: three vendors with significant market share are each incentivized to maintain enterprise-grade reliability, support, and compliance infrastructure at scale. The lock-in risk comes from API dependencies: enterprise teams that have built production workflows on Claude, GPT-5, or Gemini face real migration costs if they need to switch vendors — not because the underlying models are interchangeable (they're not, on specific benchmark tasks), but because production codebases, fine-tunes, and prompts are rarely portable without rework. Enterprise teams should build their LLM architecture with explicit vendor switching costs in mind: abstract the model API call behind an internal interface that can route to different providers, maintain prompt libraries that have been tested across at least two frontier providers, and resist building deeply on any single vendor's proprietary features — function calling conventions, tool definitions, system prompt structures — that don't have cross-provider equivalents. The market share data also suggests that enterprise teams defaulting to OpenAI from 2022-2023 familiarity should revisit their vendor assumptions; the performance landscape has changed materially since the ChatGPT launch era.
How does Anthropic's enterprise API dominance affect its IPO valuation case?
Anthropic's 40% enterprise LLM API market share is the central revenue story behind its $2 trillion IPO target — and the Menlo Ventures data provides the most concrete third-party validation of why investors have accepted that valuation. At 40% of an $8.4 billion enterprise LLM API market, Anthropic's implied share is approximately $3.4 billion in annual API revenue from enterprise customers alone. Combined with consumer Claude.ai subscriptions, team plans, and operator API usage, the $100 billion annualized revenue target disclosed in Anthropic's IPO reporting requires approximately 12x growth from the mid-2025 enterprise LLM market size — which is plausible if the overall enterprise AI market grows at the rate Gartner and Menlo Ventures are projecting. The risk to the IPO story is whether Anthropic's enterprise API share is sustainable as OpenAI improves its enterprise capabilities, Google leverages GCP relationships more aggressively, and open-source models continue improving to the point where some workloads shift away from frontier API providers. Enterprise buyers who have read Anthropic's IPO coverage and understand the market share data are now in a position to negotiate from awareness of the vendor's revenue dependence on enterprise retention — a leverage point few buyers use explicitly.