Three angles this week

Observation

Enterprises are blowing through AI budgets in months, not years. Uber's engineers burned their entire 2026 token allocation in four months. 71% of companies exceeded AI budgets in 2025.

Angle

Most carriers are still pricing AI like SaaS — fixed seats, predictable renewals. Agentic workloads don't work that way. The cost unit is consumption, not license, and it scales with usage in ways finance teams have no model for yet.

Implication for P&C carriers

P&C carriers running claims triage, subrogation, or underwriting agents need consumption-based cost models before scaling. A single complex claims workflow can amplify token usage 55x versus a simple query. Budget overruns aren't a risk — they're the default outcome without architecture guardrails.

Observation

The Starbucks AI inventory failure — scrapped after nine months across 11,000 stores — wasn't AI losing to humans. It was a pre-LLM computer vision vendor getting lapped by foundation models that can simply read a label.

Angle

The first AI casualties in enterprise aren't human jobs — they're the 2021-vintage AI point solutions that carriers bought during the last wave. The real obsolescence risk sits inside existing vendor contracts, not in future AI replacing staff.

Implication for P&C carriers

P&C carriers running narrow-trained ML models for fraud detection, first notice of loss triage, or document extraction should audit those vendor vintages now. Foundation models handle the same tasks with broader generalization and no per-category retraining. Contracts signed in 2022-2023 may be locking carriers into pre-LLM architecture at post-LLM prices.

Observation

Public opposition to data centers is intensifying — contaminated water, booed commencement speakers, community vetoes. Stratechery frames this as ordinary people gaining veto power over AI infrastructure they couldn't stop before.

Angle

P&C carriers aren't building data centers, but they operate in the same communities bearing the costs. The backlash is a leading indicator of regulatory and reputational risk that the industry hasn't priced into its AI governance posture. Being seen as responsible AI users — not just deployers — is now a competitive differentiator with regulators and policyholders.

Implication for P&C carriers

State insurance regulators are watching public sentiment on AI closely. Carriers that can demonstrate explainable, auditable AI decisions in claims and underwriting — and communicate that clearly — are better positioned as regulatory scrutiny tightens. The trust deficit in AI is now a material risk in a regulated industry.