Executive thesis.
The Q2 2026 compensation market has split into three tiers that no longer move together: AI frontier roles anchor at a $350,000 ceiling, service and trades roles post the fastest annual wage growth at 3.1 percent, and mid-tier tech compresses toward a $155,000 cost-adjusted floor shared across New York, San Francisco, Los Angeles, Chicago, and Seattle.
What the Orbyt data says.
By Q4 2029, Principal AI Research Scientist median crosses $500,000 while the cost-adjusted median across New York, San Francisco, Los Angeles, Chicago, and Seattle stays within 5 percent of $155,000, widening the frontier-to-floor ratio from 2.3x today to 3.2x.
Methodology: Frontier roles have compounded at roughly 9 to 11 percent annually in the Orbyt dataset since 2023, driven by a small number of labs competing on a thin supply curve. Cost-adjusted floors have moved less than 2 percent annually because COL multipliers in tier-1 metros absorb nominal gains. Straight-lining those two rates to 2029 yields the claim. Downside risk: a frontier-lab funding reset compresses the ceiling before 2029.
Stop benchmarking to San Francisco nominal. Benchmark to the $155,000 cost-adjusted floor for mid-tier engineering and product, and pay the $350,000 AI frontier number without apology for the one or two roles that actually move your roadmap. In my experience running comp at three companies, the most expensive mistake in 2026 is paying San Francisco nominal to a Data Analyst (1.3 percent growth, commoditizing fast) while underpaying the single Principal AI Research Scientist who decides whether your product ships. Split the band. Publish the split internally.
Top-paying roles.
Top-paying metros.
Best value metros (cost-adjusted).
Fastest-growing roles (YoY).
Downloads · 2026 · CC BY 4.0 · Free
The Orbyt Intelligence API.
The published CSV is the national-level summary. The API serves the full role-by-metro matrix, historical year-over-year time series, programmatic filtering, and embargo access to next quarter’s snapshot before it goes public. Build tier ($99/mo) included.
Read the API docs.Methodology.
The Orbyt Intelligence dataset anchors this report on 3,445 roles across 81 U.S. metros, representing 279,045 role-by-city data points. The Q2 2026 snapshot is the analytic baseline. Previous-quarter deltas reference the Q1 2026 snapshot.
Primary sources are the U.S. Bureau of Labor Statistics Occupational Employment and Wage Statistics program (BLS OES), the Department of Labor's H-1B Labor Condition Application disclosures, SEC 10-K and proxy-statement compensation data, and a proprietary aggregation across more than 50 job-posting platforms. Cost-of-living adjustments use the Bureau of Economic Analysis Regional Price Parities.
Figures are rounded to the nearest $1,000 unless otherwise noted. Experience-band estimates scale off the role median using a fixed multiplier set disclosed in the Appendix of the Enterprise Annual edition. Percentile bands (25th, 50th, 75th) derive from the underlying distribution rather than a model fit.
My read on the limitations: the Orbyt dataset reflects posted compensation and disclosed compensation, not realized compensation at the individual level. Equity values reflect grant-date fair value, not mark-to-market value, which understates realized comp in up markets and overstates it in down markets. Geographic coverage is strongest for the top 30 U.S. metros and weaker for secondary markets. The dataset does not capture private-company pre-IPO equity that is not disclosed through H-1B or SEC filings, which materially understates AI-native startup compensation at the senior-IC and staff level.
Projections through 2030 use an employer-posting panel, an occupational displacement model keyed off BLS category mappings, and an AI-capability premium regression run per role archetype. Confidence levels are disclosed in-line with every projection. High confidence means the projection would need a structural break to miss. Moderate confidence means the projection is sensitive to one or two known inputs. Low confidence means the data supports the direction but not the magnitude, and the claim is made to advance the conversation rather than to be defended as precise.
- BLS OES (Occupational Employment and Wage Statistics)
- H-1B LCA filings (U.S. DOL disclosure data)
- SEC filings (10-K, proxy statements)
- 50+ job-posting platforms (aggregated)
APA: Bartak, J. (2026). The AI Compensation Report 2026. Orbyt Intelligence. https://www.orbytjobs.ai/orbyt-intelligence/reports/2026-ai-compensation-summary
Chicago: Justin Bartak, "The AI Compensation Report 2026," Orbyt Intelligence, 2026, https://www.orbytjobs.ai/orbyt-intelligence/reports/2026-ai-compensation-summary.
Plain: Source: Orbyt Intelligence, 2026. The AI Compensation Report 2026, by Justin Bartak.
Dataset paths cited: macro.globalMedian, macro.globalDelta, macro.topPayingRoles, macro.topPayingCities, macro.topValueCities, macro.fastestGrowingRoles, sampledRoles, sampledCities
Published April 19, 2026. Generated June 4, 2026. Free to quote with attribution and a backlink. Up to 250 words per public use. Source: Orbyt Intelligence, 2026.