How 100 Enterprise CIOs Are Building and Buying Gen AI in 2025

• Enterprise LLM budgets grew past forecasts; avg +75% YoY.
  • “What I spent in 2023 I now spend in a week.”
  • Innovation budget share collapsed: 25% → 7%; AI is core IT now.
• Multi-model world: 37% use 5+ models (vs. 29% prior year).
  • Claude leads coding; Gemini leads system design; OpenAI leads production deployments.
  • xAI showing strong early enterprise interest.
• Model costs dropping order-of-magnitude every 12 months.
  • Gemini 2.5 Flash: $0.26/M tokens vs. GPT-4.1 mini at $0.70/M.
  • Fine-tuning declining; long context + prompt engineering cheaper.
• Reasoning models (o3) in 23% of enterprises; DeepSeek at 3%.
  • Still early but optimism is high; few in production yet.
• Marked shift: enterprises buying off-the-shelf AI apps over building.

  • 90% testing third-party customer support apps.

  • Software dev is killer use case: one CTO reports ~90% AI-generated code via Cursor + Claude Code.
• AI-native vendors outpacing incumbents on quality and velocity.
  • Prosumer pull (ChatGPT, Cursor) driving enterprise adoption bottom-up.
  • Outcome-based pricing remains contentious; usage-based preferred.

Sarah Wang, Shangda Xu, Justin Kahl, Tugce Erten — a16z · 2025-06-10 · Read on a16z.com