AI cut 18,400 jobs in Q2 2026 — and created 31,300. Here's the full picture.

Q2 2026 recorded 18,400 job cuts where AI was cited with explicit or strong attribution, against roughly 31,300 new AI-related roles from labor-market data providers — a net of +12,900. That makes three consecutive net-positive quarters, and the margin is widening: Q4 2025 was +4,100, Q1 2026 was +8,600.

The usual caveat applies with more force than ever: percentage-based and ranged announcements sit outside these totals, so the true cut figure is higher. We publish them in the live ledger with their own precision labels rather than guessing them into a sum.

Two categories dominated. Customer support and network operations — led by Telstra's 2,800-role simplification in April — and content moderation, where the execution of Meta's February restructuring made trust-and-safety the single most-cut role family of the quarter. Together they account for 61% of Q2's attributed cuts.

“For every role cut with AI cited this quarter, 1.7 new AI-related roles were posted — the widest ratio since tracking began.”

Geographically, Q2 broke the U.S. concentration pattern: Australia and Japan together contributed 29% of attributed cuts, up from 8% a year earlier, as agentic operations tooling reached telecom and IT-services firms outside the American tech sector.

AI-engineering and ML-operations postings grew 22% quarter-over-quarter, and the wage premium for AI-skilled roles reached +34% — up from +25% in 2024. The fastest-growing single category was AI integration consulting, concentrated in Japan and Germany, where enterprise adoption programs are hiring faster than the local layoff rate.

The asymmetry is the story: the median cut role paid below the national median wage; the median created role paid 1.6× above it. A net-positive quarter is not a symmetric one — displaced support workers are not, by and large, the people filling ML-ops seats.

Across the ledger, AI capex growth leads attributed cuts by roughly two quarters (r = 0.71 on quarterly buckets) — companies spend first, restructure second. Revenue per employee in the AI-first cohort rose to $2.1M, nearly five times the S&P 500 median, and headline sentiment improved 6 points even as cut announcements continued.

Treat the correlation as descriptive, not causal — a small ledger is not a panel study. The signal we trust most is the disclosure effect: since New York's WARN AI question took force, attributed cuts in filings run ~40% higher than press-release attribution alone would suggest.

Three things. California's displacement-disclosure rule begins reporting in September, which should surface cuts that never reached the press. The zero-person cohort's growth rate (+51 YTD) implies it crosses 200 companies by year-end. And if the capex-to-cuts lag holds, the record Q1–Q2 2026 spending wave points to an acceleration of attributed cuts in Q4 2026–Q1 2027 — the first real test of whether the 1.7× creation ratio survives a heavier cutting cycle.