The $-Per-Week CIO: How 100 Enterprise Leaders Are Building and Buying Gen AI (2025)

See also (wiki): wiki/ai-platform-selection.md, wiki/ai-maturity-models.md, wiki/firm-size-ai-outcomes.md, wiki/workflow-redesign.md

Frontmatter

• Source: Andreessen Horowitz (a16z), “How 100 Enterprise CIOs Are Building and Buying Gen AI in 2025”
• URL: https://a16z.com/ai-enterprise-2025/
• Date: May 8, 2025
• Methodology: n=100 CIOs, 15 industries; 25+ enterprise buyer interviews
• Source tier: TIER 2 — CIO survey + qualitative interviews; a16z has portfolio financial interest in AI-native vendors; apply VC bias caveat. Consistent with Menlo Ventures 2025 and other independent build/buy data.
• Last updated: 2026-04-23

Key Findings

• Enterprise AI spend growing ~75% year-over-year with innovation budget share collapsing from 25% (2024) to 7% (2025) of total LLM spending — AI has moved from experiment to operating infrastructure line item (a16z, n=100 CIOs, May 2025).
• The shift from build to buy is complete in customer support: 90%+ of respondents are testing third-party customer support solutions rather than building internally — the build/buy question has been answered for this use case (a16z, 2025).
• Software development is the “killer use case”: One CTO reported 90% of code is now AI-generated, up from 10–15% a year prior. Consistent with Anthropic Agentic Coding Trends 2026 data and a16z’s April 2026 Fortune 500 penetration finding.
• Multi-model adoption is accelerating: 37% of enterprises now use 5+ AI models (vs. 29% in 2024). Model differentiation by use case is primary driver — enterprises are not converging on one vendor (a16z, 2025).
• Cost and security now outrank accuracy in model selection — this is a procurement maturity signal. When accuracy was the main criterion, enterprises were still evaluating. Prioritizing cost and security means they are operating (a16z, 2025).
• Outcome-based pricing remains blocked: CIOs are “still uncomfortable” with outcome metric definition and attribution. Usage-based pricing dominates. This limits vendor ability to capture value from AI-driven productivity gains (a16z, 2025).
• AI-native vendors outpacing incumbents on product velocity: Cursor’s NPS versus GitHub Copilot is the specific data point cited. Consumer brand strength (ChatGPT halo) is driving enterprise procurement — the consumerization-to-enterprise pipeline is confirmed (a16z, 2025).
• Agentic workflows creating model lock-in: Extensive prompt engineering and custom guardrails in agent deployments are raising switching costs — early platform commitments will have multi-year stickiness (a16z, 2025).

The Build/Buy Signal Worth Tracking

The shift from 25% to 7% innovation-budget share is the critical leading indicator. When AI spend moves out of the innovation budget and into core IT and business-unit budgets, it means:

1. The CFO has line-item visibility — this is no longer discretionary
2. Procurement cycles apply — vendor selection is more formal
3. ROI is now expected, not just hoped for
4. The deployment risk profile changes — budget owners are accountable for outcomes

This maps directly to why outcome-based pricing is stalled: budget owners who are newly accountable don’t want to commit to outcome metrics they don’t yet know how to measure.

Cross-References to Existing Research

• Extends: a16z “Where Enterprises Are Actually Adopting AI” (April 2026, 29% Fortune 500 penetration) — the CIO survey is the “how” to the April 2026 report’s “how many.”
• Corroborates: Menlo Ventures State of GenAI 2025 (n=495, 76% buy vs. build) — a16z CIO data is consistent with the broader build/buy shift.
• Corroborates: Anthropic Agentic Coding Trends 2026 — software dev as dominant use case confirmed across two vendor sources.
• Tensions with: METR RCT (experienced developers 19% slower on open-ended tasks) — the “90% of code is AI-generated” CTO claim is uncontrolled; vendor-selected positive cases.

Data Table

Source Credibility Assessment

Tier: TIER 2 — 100 CIOs is a meaningful sample for enterprise procurement decisions, but small for statistical inference. a16z financial interest in AI-native startup adoption creates selection bias risk. Directional findings are consistent with independent corroboration (Menlo Ventures, Anthropic) — use for trend direction, not precise percentages.

Ingested: 2026-04-23