The AI talent market has passed through three phases: the ZIRP-fueled hiring spree of 2021-2022, the correction of 2023, and the AI-driven rebound of 2024-2026. Each phase produced its own distortions and myths. This post is the honest numbers on where compensation sits, where the real supply-demand imbalances are, and what is bubble vs substance in AI hiring.
The numbers
These are the US-fully-loaded numbers we see across client searches and our own hiring. Figures are approximate and vary by location, company stage, and negotiation.
Junior AI engineers (0-2 years production experience): $140K-$200K at mid-size tech, $250K-$400K at frontier labs. The gap narrowed in 2025-2026 as non-lab companies got more serious about AI compensation.
Senior AI engineers (5-8 years): $260K-$400K at market, $500K-$900K at frontier. This is where the gap widens dramatically — frontier labs are paying senior talent in a different tier, and non-lab companies need strong equity or mission to compete.
Staff/principal engineers: $400K-$700K at market, $900K-$2M+ at frontier. For truly top senior engineers, frontier labs effectively set the market and everyone else either matches or loses.
Research scientists with publications: $300K-$600K at market, $1M-$5M+ at frontier. The top 50-100 researchers globally command compensation that is genuinely unusual for technical roles.
AI engineering managers: $350K-$600K at market, $700K-$1.5M at frontier.
International calibration: India compensation is roughly 25-40% of US equivalents (reflecting PPP and local norms). Europe is 50-70%. Singapore and Hong Kong are closer to 70-80% of US.
What is real (substance, not bubble)
Senior AI engineer compensation has genuinely reset higher. It is not coming back to 2022 levels. The supply of people with real production AI engineering experience — not just prompt tutorials — is limited, and that scarcity is durable enough to justify premium.
Frontier-lab compensation for top talent is real and growing. This is a small market (hundreds of people globally), and it distorts the broader market in ways that matter. Companies competing for this talent are paying extreme premiums that make sense only if AGI-scale outcomes materialize.
Applied AI skills (shipping to production, evals, retrieval, ops) command consistent premium over prompt-engineering-only skills. Market has sorted. Hirers who know what they're looking for reward production depth.
What is bubble
Prompt engineering as a standalone role. Salaries for pure 'prompt engineers' peaked in 2023 and have collapsed. The role was real briefly; now it's a skill every engineer has.
AI ethicists and AI policy specialists at unreasonable levels. There's a real demand for governance expertise; there's not demand for a new profession at $500K/year. Most companies need governance expertise as part of a legal or compliance function, not as a new senior role.
'Head of AI' titles without clear accountability. Someone paying $400K for a VP of AI without a concrete mandate is mostly signaling. Real AI leadership roles have specific product or platform outcomes attached.
Chief AI Officer roles with no reporting line to product or engineering. Political more than operational. Expect the title to settle as companies figure out where AI actually lives in the org.
The skills that actually pay
From our hiring observations across 2024-2026, the combinations that command the biggest premiums:
Senior software engineer + production AI experience: rare because strong systems engineers who also have shipped AI-specific production systems (evals, guardrails, observability) are not plentiful. These candidates often name their price.
Research background + ability to ship: valuable because pure researchers often can't ship and pure engineers often don't get the theory. The bridge people are scarce and highly sought.
Domain expertise + AI skills: legal + AI, medical + AI, financial + AI. When depth in a vertical combines with AI fluency, the candidate can lead applied-AI work in that vertical with unusual effectiveness.
What hiring managers should ignore
Resume claims about 'working with GPT-5' or similar — meaningless, everyone has done this. Certifications from vendors — valuable for junior signal, irrelevant for senior. LinkedIn title inflation — senior AI engineers with 2 years of experience typically means 1.5 years of prompt tutorials. Interview on actual production work and judgment (see hiring post).