Research · Summary
Back to sweepResearch sweep · standard · 2025 – present
Enterprise LLM Vendor Selection and Consumption Models
Enterprise LLM vendor selection and consumption patterns (April 2025–present): how companies choose between OpenAI, Anthropic, Google, hyperscaler-hosted model access, and direct API relationships; what decision metrics they use across availability, quality, price, governance, and SLAs; and how adoption differs by company size, workload criticality, and realtime versus offline use cases
- Claude Opus 4.8
- financial
- frontier
- academic
- vc
- substack
Synthesised 2026-04-13
Overview
Enterprise LLM vendor selection became a distinct procurement discipline in 2025, and the headline event is an inversion at the top. Menlo Ventures' July 2025 survey of 150 technical leaders found Anthropic holding 32% of enterprise production workloads, ahead of OpenAI at 25% (down from 50% in 2023) and Google at 20%. Code generation drove the swing, with Claude taking 42% developer share against OpenAI's 21%.
Sources: Menlo Ventures (2025) (↗); Menlo Ventures (2025) (↗); Menlo Ventures (2025) (↗)
Spending grew faster than the share rankings shifted. Enterprise LLM spend nearly tripled from $3.5bn in late 2024 to $8.4bn by mid-2025, and 37% of enterprises now run five or more models in production. That last figure is the structural story: buyers are not standardising on one frontier vendor but building workload-based portfolios, routing Claude to code, GPT to retrieval and reasoning, and Gemini to multimodal tasks.
Sources: Menlo Ventures (2025) (↗); AI TechPark (amplifying Menlo Ventures data) (2025) (↗); Typedef.ai (2025) (↗)
The shift underneath the numbers is from "best model" to "best portfolio" selection, enabled by roughly 80% token-price compression between early 2025 and early 2026 and by cloud-mediated access lowering the cost of holding multiple relationships. As quality converged, differentiation moved to governance, compliance certification, SLA structure and price predictability. The unresolved question is whether value ends up with model providers or with the orchestration and governance layers that abstract them away.
Sources: Price Per Token (2026) (↗); Gartner (2025) (↗)
Timeline
- Claude Sonnet 3.5 sets coding benchmark
- Enterprise LLM spend at $3.5bn
- Claude Sonnet 3.7 release
- Gartner forecasts 3x task-specific model use by 2027
- Menlo survey shows Anthropic overtaking OpenAI
- Enterprise spend reaches $8.4bn
- Multi-model portfolios become default architecture
- Closed-source share rises to 87%
- Accenture-Anthropic partnership trains 30,000 staff
- Agentic enterprise licence agreements emerge
- Token prices down ~80% year-on-year
- Enterprise LLM gateways formalise as a category
Key Findings
Anthropic's lead concentrates in code, not across the board. Menlo's data shows Claude at 42% developer share versus OpenAI's 21%, and code generation is described as AI's first killer app. Other lanes put Anthropic's enterprise spend share as high as 40% by late 2025, citing the same Menlo research, so the precise figure varies by metric (workloads versus spend).
Sources: Menlo Ventures (2025) (↗); AI CERTs News (2025) (↗)
Switching is rare but upgrading is constant. Menlo found only 11% annual vendor switching, with 66% of enterprises upgrading within their existing vendor. Switching costs are material: migration runs 20-40 hours for shallow API work and 80-120 hours for deep integration with fine-tuning and embeddings.
Sources: Menlo Ventures (2025) (↗); SoftwareSeni (2025) (↗)
Pricing unpredictability is the dominant procurement complaint. CloudEagle reports 78% of IT leaders saw unexpected charges from consumption-based AI pricing, and average AI-native app spend hit $1.2m, up 108% year-on-year. This is driving the agentic enterprise licence, a fixed-fee structure CIOs and CFOs prefer over consumption metering.
Sources: CloudEagle.ai (2025) (↗)
Cloud channels win on operations, direct relationships win on access. Claude is the only frontier model on all three major clouds (Bedrock, Vertex, Azure), which buyers use for multi-cloud optionality. Azure OpenAI offers a 99.9% uptime SLA with ISO/SOC/HIPAA compliance across 27 regions, the kind of pre-certification that regulated industries cannot easily replicate via direct API.
Sources: Xenoss (2025)
System integrators are becoming a decision pathway. The Accenture-Anthropic partnership trains 30,000 professionals on Claude, and the AI system integration market reached $11bn in 2025, projected at $14bn in 2026.
Sources: Accenture Newsroom (2025) (↗)
Gartner expects fragmentation to deepen. By 2027, organisations will deploy small task-specific models at 3x the volume of general-purpose LLMs, with value accruing to orchestration platforms that route workloads across portfolios.
Sources: Gartner (2025) (↗)
Evidence & Data
The core numbers cluster around Menlo Ventures' July 2025 survey: Anthropic 32%, OpenAI 25%, Google 20% of production workloads; spend of $8.4bn up from $3.5bn; 11% switching and 66% within-vendor upgrades; 37% running five or more models.
Sources: Menlo Ventures (2025) (↗); Menlo Ventures (2025) (↗)
On consumption economics, output tokens run 3-10x input costs, batch APIs offer 50% discounts and prompt caching up to 90% savings on cached input. Closed-source models hold 87% of workloads, up from 81%. Spending intensity: 37% of enterprises spend over $250,000 annually, 73% over $50,000.
Sources: IntuitionLabs (2025) (↗); Menlo Ventures (2025) (↗); Typedef.ai (2025) (↗)
On vendor sustainability, OpenAI reportedly burned $8bn on compute in 2025 and projects $14bn cumulative losses by end-2026, a factor in long-term contract negotiation.
Sources: Menlo Ventures (2025) (↗)
Tensions & Open Questions
The share figures disagree. Workload share (32%) and spend share (40%) for Anthropic come from the same Menlo research but measure different things, and OpenAI's December 2025 State of Enterprise AI report tells a more favourable story. Buyers should treat any single ranking with caution.
Sources: AI CERTs News (citing Menlo Ventures research) (2025) (↗)
SLAs remain weak where they matter. Reports note SLAs frequently lack clarity on uptime, latency or accuracy, with little financial recourse. Hard SLAs concentrate in cloud-mediated channels; direct best-effort APIs dominate elsewhere, and the gap is unresolved for mission-critical realtime workloads.
The academic record is nearly empty. Only the NAACL 2025 Industry Track paper by Zhang et al. and a few preprints address enterprise evaluation, none covering procurement or lock-in empirically. The evidence base is consultant surveys, not peer review.
Sources: NAACL 2025 Industry Track (Association for Computational Linguistics) (2025) (↗)
Whether the agentic enterprise licence holds is untested. CFOs want predictability, but vendors burning billions may resist fixed-fee terms that cap revenue. If quality re-diverges, the portfolio logic that weakened lock-in could reverse just as fast.
![[sources-enterprise-llm-vendor-selection-and-consumption-pa]]
Sources
Summary: ↑ Back to summary
Financial Press
| ID | Title | Outlet | Date | Significance |
|---|---|---|---|---|
| f1 | 2025 Mid-Year LLM Market Update: Foundation Model Landscape + Economics | Menlo Ventures | 2025-07 | Survey of 150+ technical leaders showing Anthropic now holds 32% enterprise LLM market share (vs OpenAI 25%, Google 20%); documents decision metrics: code generation capability, inference economics, and sticky vendor behavior (66% upgrade within provider, only 11% switch) |
| f2 | Generative AI - 2025 | Bloomberg Professional Services | 2024-11 | Bloomberg Intelligence CIO survey showing 75% plan IT-infrastructure budget increases; addresses hyperscaler capex competition and inference cost trends; context for vendor positioning on infrastructure/SLA reliability |
| f3 | Evolving LLM Market: Anthropic Leads 2025 Enterprise Share | AI CERTs News (citing Menlo Ventures research) | 2025-12 | Multi-model adoption data (37% of enterprises deploy 5+ models); documents governance pressures and vendor lock-in concerns as drivers of portfolio strategies; 42% Claude adoption for coding |
| f4 | [Enterprise LLM Market | Global Market Analysis Report - 2035](https://www.futuremarketinsights.com/reports/enterprise-llm-market) | Future Market Insights | 2025-09 |
| f5 | Large Language Model Market Forecast 2032 | Persistence Market Research | 2025-01 | Market consolidation narrative: shift from fragmented startups to major tech consolidation (acquisitions, partnerships); OpenAI paying users grew 3M–5M (June–Aug); closed-source models dominate (87% of usage) |
| f6 | AI Pricing in 2025: A Detailed Guide | CloudEagle.ai | 2025-11 | Enterprise procurement focus: per-token pricing vs. credit systems; vendor lock-in through data portability restrictions; contract intelligence and benchmarking; SLA clarity gaps on uptime, latency, accuracy |
| f7 | AI Pricing: What's the True AI Cost for Businesses in 2026? | Zylo | 2026-02 | Enterprise cost governance: AI-native spending doubled YoY to $1.2M avg.; 78% of IT leaders report unexpected charges due to consumption-based pricing; contrasts Microsoft Copilot ($30/user/month fixed) vs. Salesforce Agentforce (per-resolution) pricing models |
| f8 | How to Price AI Products: The Complete Guide for PMs (2026) | Aakash G. (practitioner analysis) | 2026-02 | Documents economics crisis: OpenAI burned $8B on compute in 2025 (projects $14B cumulative losses by end 2026); GitHub Copilot lost money per user at launch; illustrates why vendor selection on pricing/SLA reliability matters; Cursor pricing collapse case study |
| f9 | 13 LLM Adoption Statistics: Critical Data Points for Enterprise AI Implementation in 2025 | Typedef.ai | 2025-10 | Enterprise financial commitment: 37% spend >$250K annually, 73% spend >$50K; 3.7x average ROI; 72% plan to increase spending in 2025; shift from innovation budgets (25%) to mainstream infrastructure (7%) |
| f10 | Agentic AI Providers Comparison 2025: Features, Pricing Models, and Best-Fit Use Cases | Monetizely (SaaS pricing research) | 2025-11 | Decision framework for enterprise agentic AI: stack alignment, latency/throughput requirements, reliability/SLAs, security posture; hybrid pricing structures (platform fee + usage, committed volume deals); governance as differentiator for board-approved decisions |
Frontier Lab & Model News
| ID | Title | Outlet | Date | Significance |
|---|---|---|---|---|
| t1 | 2025 Mid-Year LLM Market Update: Foundation Model Landscape + Economics | Menlo Ventures | 2025-07 | Primary market research tracking enterprise LLM adoption by vendor (Anthropic 32%, OpenAI 25%, Google 20%), key drivers of vendor selection including code generation dominance, and shift toward inference-driven workloads. |
| t2 | Evolving LLM Market: Anthropic Leads 2025 Enterprise Share | AI CERTs News | 2025-12 | Quantifies enterprise LLM spending surge ($3.5B to $8.4B in six months), Anthropic market leadership in coding (42% adoption vs OpenAI's 21%), and evidence that multi-model strategies are gaining traction to mitigate vendor lock-in. |
| t3 | Comparing OpenAI Anthropic and Google for Startup AI Development in 2025 | SoftwareSeni | 2025-12 | Analysis of vendor lock-in risk, migration costs (20–120 hours depending on integration depth), and strategic contracting recommendations centered on source code access and data portability. |
| t4 | Top 11 LLM API Providers in 2026 | Future AGI (Substack) | 2026-02 | Comprehensive coverage of enterprise SLA requirements (99.9% uptime, ISO/SOC/HIPAA compliance), cloud platform options (Azure OpenAI, Bedrock), and deployment architectural trade-offs across regions and dedicated infrastructure. |
| t5 | LLM API Pricing Comparison (2025): OpenAI, Gemini, Claude | IntuitionLabs | 2025-10 | Pricing evolution and forecasts showing shift toward premium-controlled markets (Western providers focus on SLAs and compliance) versus commodity use moving to open-source; evidence that pricing has become a chief competitive factor by 2026. |
| t6 | LLM API Pricing 2026 - Compare 300+ AI Model Costs | Price Per Token | 2026-03 | Real-time pricing comparison tool tracking cost dynamics across 300+ models, reflecting aggressive pricing compression (~80% reductions 2025–2026) and token-based cost as enterprise selection criterion. |
| t7 | Accenture and Anthropic Launch Multi-Year Partnership to Drive Enterprise AI Innovation | Accenture Newsroom | 2025-12 | Signals enterprise contracting and integrator partnerships; Accenture training 30,000 professionals on Claude for regulated industries (finance, healthcare); demonstrates move from pilots to production deployment with governance frameworks. |
| t8 | Claude in the enterprise: case studies of AI deployments and real-world results | DataStudios | 2025-09 | Real-world enterprise case studies (TELUS 57K users, Tines cybersecurity, NNSA 94.8% detection rate) showing multi-cloud deployment patterns (Anthropic API, AWS Bedrock, Google Vertex AI, private endpoints), model diversity strategies, and operational SLA requirements. |
| t9 | Anthropic Economic Index report: Uneven geographic and task-level patterns | Anthropic Research | 2025-09 | Official research on enterprise Claude deployment patterns showing 77% automation rate (task delegation vs collaboration), task concentration analysis, and infrastructure requirements (lengthy inputs for complex tasks creating data centralization barriers). |
| t10 | LLM API Pricing Comparison 2025: Complete Cost Analysis Guide | Binadox | 2025-08 | Documents shift toward enterprise SLA-based pricing tiers (mission-critical, standard, budget), commitment-based discounts, and integration of pricing with compliance features justifying premium tiers for regulated workloads. |
Academic & arXiv
| ID | Title | Outlet | Date | Significance |
|---|---|---|---|---|
| a1 | Evaluating Large Language Models with Enterprise Benchmarks | NAACL 2025 Industry Track (Association for Computational Linguistics) | 2025-04 | Directly addresses enterprise LLM evaluation across 25 domain-specific benchmarks spanning financial services, legal, climate, and cybersecurity - core evidence for vendor selection criteria in regulated and critical workloads. |
| a2 | Large Language Model Evaluation in 2025: Smarter Metrics That Separate Hype from Trust | TechRxiv Preprint | 2025 | Proposes multidimensional evaluation framework for enterprise-grade LLMs covering latency, privacy, energy efficiency, and hallucination - directly mapped to procurement decision criteria beyond raw benchmark accuracy. |
| a3 | Enterprise Large Language Model Evaluation Benchmark | arXiv | 2025-06 | Benchmark-focused paper evaluating six leading models on enterprise performance gaps, offering actionable optimization insights relevant to cost-performance tradeoffs in vendor selection. |
VC & Analyst Reports
| ID | Title | Outlet | Date | Significance |
|---|---|---|---|---|
| v1 | 2025 Mid-Year LLM Market Update: Foundation Model Landscape + Economics | Menlo Ventures | 2025-07 | Landmark VC-backed survey of 150 technical leaders quantifying enterprise LLM consumption: Anthropic 32% (up from niche), OpenAI 25% (down from 50%), Google 20%. Documents $3.5B→$8.4B spending surge, closed-source dominance (87%), and code generation as killer app (Claude 42% vs OpenAI 21%). |
| v2 | Gartner Predicts That by 2030, Performing Inference on an LLM With 1 Trillion Parameters Will Cost GenAI Providers Over 90% Less Than in 2025 | Gartner | 2026-03 | Cost trajectory forecast shaping vendor selection logic: 90% inference cost reduction by 2030, but overall costs rising due to token consumption surge. Emphasizes portfolio orchestration across small domain-specific models vs. commodity frontier models as strategic imperative. |
| v3 | Gartner Predicts by 2027, Organizations Will Use Small, Task-Specific AI Models Three Times More Than General-Purpose Large Language Models | Gartner | 2025-04 | Predicts 3:1 volume shift toward task-specific over general-purpose LLMs by 2027, driven by accuracy and cost. Recommends multi-model portfolio strategies with RAG/fine-tuning, implying vendors must compete on specialization and integration, not monolithic capability. |
| v4 | Enterprise LLM Spend Hits $8.4B as Anthropic Tops OpenAI | AI TechPark (amplifying Menlo Ventures data) | 2025-08 | Replicates Menlo findings with added vendor-switching insight: only 11% of teams switch providers annually; 66% upgrade within vendor; 23% make no changes. Documents market consolidation and sticky dynamics despite rapid share shifts. |
| v5 | Responsible Innovation: A Strategic Framework for Financial LLM Integration | Academic/Industry (multi-institutional) | 2025 | Six-step governance decision framework for regulated sectors (finance, healthcare). Maps selection criteria beyond performance: compliance frameworks, ROI justification, data governance, risk management - critical for high-stakes workload segments. |
| v6 | Large Language Model Evaluation in 2025: Smarter Metrics That Separate Hype from Trust | TechRxiv (peer-reviewed preprint) | 2025 | Documents evolution of enterprise LLM evaluation metrics (2020–2025): semantic accuracy, latency, explainability, adversarial robustness, fairness. Emphasizes production trade-offs (latency vs. benchmark score) shaping procurement decisions beyond leaderboard rankings. |
| v7 | Buy versus Build an LLM: A Decision Framework for Governments | Academic/Policy (arXiv) | 2026-02 | Cites Menlo Ventures data (88% market share held by Anthropic, OpenAI, Google) as evidence of concentration. Frames buy-vs-build decision tree relevant to enterprise SOI: diversification, talent, ecosystem maturity - mirrors commercial vendor selection trade-offs. |
| v8 | A Cost-Benefit Analysis of On-Premise Large Language Model Deployment: Breaking Even with Commercial LLM Services | Academic (arXiv) | 2025-08 | Quantifies on-prem vs. cloud API economics: breakeven analysis for open-source (Llama, Qwen) vs. commercial (OpenAI, Anthropic, Google). Evaluates data privacy, switching costs, and long-term TCO drivers shaping consumption model selection. |
| v9 | LLM in Enterprise: A Complete Guide | TrueFoundry (practitioner/infrastructure vendor) | 2026-01 | Contrasts on-premise (governance, control, compliance) vs. cloud-managed (OpenAI, AWS Bedrock, Google, Azure) consumption models. Documents operational shift from experimentation to production: governance, observability, billing controls as decision criteria. |
| v10 | Emerging Patterns for Building LLM-Based AI Agents | Gartner | 2025 | Gartner research on agentic AI architecture patterns and vendor capabilities. Relevant to workload-specific selection (agents vs. chat vs. retrieval), multi-step orchestration, and vendor-specific tool-use maturity. |
Substack Thesis Validation
| ID | Title | Outlet | Date | Significance |
|---|---|---|---|---|
| undefined1 | Enterprise LLM Platforms: OpenAI vs Anthropic vs Google | Xenoss | 2025-09 | Directly compares enterprise LLM vendor selection across TCO, integration, and security benchmarks; documents Anthropic valuation tripling in 6 months (March–September 2025) and enterprise spending rising to $8.4 billion by mid-2025. |
| undefined2 | Top 5 Enterprise LLM Gateways in 2026 | Maxim AI | 2026-04 | Describes multi-provider routing architectures (OpenAI, Anthropic, Google Gemini, AWS Bedrock, Azure) as de facto enterprise practice; documents failover, load balancing, and cost control mechanisms. |
| undefined3 | Evolving LLM Market: Anthropic Leads 2025 Enterprise Share | AI CERTs News | 2025-12 | Cites Menlo Ventures research: Anthropic enterprise share 40% (up from 12% in 2023), OpenAI 27% (down from 50%), Google Gemini 20%. Documents inference cost dominance and multi-model avoidance-of-lock-in strategies. |
| undefined4 | OpenAI, Anthropic, and Google: Who's Winning the AI Race in 2026? | Clear AI News | 2026-04 | Reports Anthropic's enterprise-first strategy generating $5 billion revenue by 2025, with Claude 3.5 Sonnet capturing 32% of enterprise LLM API market vs OpenAI GPT-4o's 25%. |
| undefined5 | LLM API Pricing Calculator for Enterprise Deployment in 2026 | Iternal | 2026-04 | Live pricing tracker documenting 80% price reductions across industry 2025–2026; quantifies asymmetric token economics (output tokens 3–10x input cost) and batch API savings (50% discount for non-interactive workloads). |
| undefined6 | A $100k Blueprint for AI-Native Procurement in 2026 | Supernegotiate Substack | 2026-04 | Substack author detailing cost displacement of legacy enterprise software (supplier management, spend analysis, contract management modules) via Claude-based agents; demonstrates direct vendor selection for AI agents. |
| undefined7 | Enterprise Agentic AI Landscape 2026: Trust, Flexibility, and Vendor Lock-in | Kai Waehner Blog | 2026-04 | Documents structural shift: all major vendors (Anthropic, OpenAI, Google, Cohere) formalizing partnerships with system integrators (Accenture, Deloitte, McKinsey); AI system integration market $11 billion 2025, projected $14 billion 2026; 95% of enterprise AI pilots fail to scale. |
| undefined8 | Enterprise technology 2026: 15 AI, SaaS, data, business trends to watch | Constellation Research | 2026-01 | Documents vendor-driven shift from consumption models to Agentic Enterprise License Agreements (AELAs) due to CIO/CFO demand for budget predictability; describes data access tolls and vendor leverage in agent ecosystems. |
| undefined9 | Buy versus Build an LLM: A Decision Framework for Governments | arXiv | 2026-02 | Academic framework for buy-vs-build spanning sovereignty, cost, safety, resource capability; extends vendor evaluation criteria beyond pure commercial actors to public-sector risk posture relevant for regulated enterprise segments. |
| undefined10 | The State of Enterprise AI 2025 Report | OpenAI | 2025-12 | Official OpenAI enterprise usage data: 8x volume growth, 320x API reasoning token consumption growth year-over-year; documents shift from prompt layer to structured workflows (Projects, Custom GPTs, 19x growth); 40–60 minutes daily time savings. |