Comparative LLM Usage Across Sectors

Narrative

The headline finding from Menlo Ventures' mid-2025 survey of 150 technical leaders is a market leadership reversal: enterprise LLM spend more than doubled from $3.5 billion in late 2024 to $8.4 billion by mid-2025, with Anthropic displacing OpenAI as the leading enterprise API provider. Menlo's year-end 2025 report refined the figures, estimating Anthropic holds 40% of enterprise LLM spend (up from 12% in 2023), while OpenAI fell from 50% to 27% over the same period. Google Gemini climbed to 21%, and the three firms together account for 88% of enterprise LLM API usage, leaving only 12% across Meta's Llama, Mistral, Cohere, and a long tail. Menlo's own portfolio investment in Anthropic is a declared conflict that tempers interpretation of these numbers, and the sample is self-selected among technically sophisticated teams.

McKinsey's State of AI 2025, published November 2025, provides a complementary and more sober read. Eighty-eight percent of respondents say their organisations use AI regularly across at least one function, up from 78% the year before, and 72% report using generative AI specifically, up from 33% in 2024. The scaling gap is the central finding: only about one-third of organisations report scaling AI enterprise-wide; two-thirds remain in what observers characterise as "pilot purgatory." Only 6% qualify as AI high performers attributing more than 5% of EBIT to AI, and just 39% report any enterprise-level financial impact. AI agents are the most prominent emerging deployment pattern, with roughly 62% of organisations at least experimenting and 23% scaling agentic systems. Adoption is highest in technology, media, telecommunications, and healthcare; financial services and legal are growing but face governance constraints. High performers are nearly three times more likely to be scaling agents and 2.8 times more likely to have fundamentally redesigned workflows.

A16z's survey of more than 100 CIOs, summarised in Kimberly Tan's "AI Adoption by the Numbers," shows enterprise LLM spend rising from $4.5 million to $7 million per organisation over two years, with CIOs projecting $11.6 million by end-2026. The report's standout structural finding is the death of the single-model regime: 81% of respondents now orchestrate three or more model families in production, up from 68% a year earlier, with frontier models reserved for high-stakes reasoning and smaller, cheaper alternatives handling routine workloads. AI budgets have migrated from innovation or R&D lines to core IT spend, growing at approximately 75% year-on-year, with innovation budget allocation dropping from 25% to just 7% of total AI spend. Separately, a16z payment data from over 200,000 startups (June to August 2025) confirmed that LLM assistants from OpenAI and Anthropic are the most-purchased AI applications among early-stage companies, with spend driven by multi-workflow ROI rather than point-solution capability.

Sequoia Capital's AI Ascent 2026, held in April 2026, declared the agent era the defining frame of the year. Sonya Huang argued that 2022 to 2024 was the chat interface phase, 2024 to 2025 the reasoning model phase, and 2026 the phase in which systems pursue goals rather than respond to prompts. Pat Grady identified a widening "diffusion gap" between what frontier labs ship and what Fortune 500 firms have deployed, framing the application layer as the primary investment opportunity. Gartner's 2025 Hype Cycle for Artificial Intelligence placed generative AI in the Trough of Disillusionment, noting that organisations struggle to demonstrate tangible ROI and face governance, hallucination, and integration challenges, while projecting that AI agents and AI-ready data are the two fastest-advancing technologies on the curve. CB Insights mapped more than 400 AI agent startups across 16 categories and identified coding AI agent revenue scaling fast in 2025, with big tech and leading LLM developers expected to own general-purpose agents while smaller players compete on vertical specificity. Open-weight models including Llama 4, Qwen 3, and DeepSeek R1 have reached performance parity with closed frontier models on most published benchmarks by 2026, but Menlo's production data shows closed-source models still controlling roughly 87% to 88% of observed enterprise API usage, with open-weight adoption strongest in self-hosted code generation and compliance-sensitive sectors where data residency matters.