Every technical discipline has had its shortage moment. There was a time when mobile engineers were impossible to find. A time when cloud architects were scarce enough to command premiums that seemed disconnected from any reasonable market logic. Those shortages were real, they drove real salary inflation, and they eventually resolved partly through training pipelines catching up, partly through demand normalising, partly through adjacent skills transferring into the gap.
AI talent scarcity does not follow that pattern, and understanding why it doesn’t is more important than any individual data point in this index.
The difference is pipeline depth. Mobile engineering shortage resolved because the skills were learnable in one to two years by engineers with adjacent experience. Cloud architecture shortage resolved because the major platforms invested heavily in certification programs that created supply at scale. AI and machine learning at the level the market needs, not AI tool usage, not prompt engineering for simple workflows, but genuine model architecture, training infrastructure, and production ML systems, requires a foundation in mathematics, statistics, and systems thinking that takes years to build and cannot be meaningfully accelerated by a bootcamp or a certification course.
The pipeline feeding qualified AI talent into the market in 2025 is the one that was seeded by graduate programs in 2019 and 2020. It is not large enough for what the market needs now. And the pipeline being seeded today will not produce a meaningful supply until 2028 at the earliest. That is the structural reality behind every number in this index.
Applicant-to-Role Ratio: The Core Measure of AI Talent Scarcity
The applicant-to-role ratio is the cleanest single measure of market tightness and for AI talent in 2025, it is the number that most vividly illustrates how different this market is from every other technical discipline.
Applicant-to-Role Ratio by Technical Discipline : US + Europe, 2025
| Discipline | Qualified Applicants per Role | Change vs 2023 |
|---|---|---|
| General Software Engineering | 8.7 | Down from 11.2 |
| Frontend Engineering | 12.4 | Down from 14.8 |
| DevOps / Cloud Engineering | 3.2 | Down from 4.1 |
| Cybersecurity Engineering | 2.8 | Down from 3.4 |
| Data Engineering | 2.4 | Down from 3.1 |
| Machine Learning Engineer | 1.6 | Down from 2.2 |
| LLM / NLP Specialist | 1.1 | Down from 1.4 |
| Multimodal AI Engineer | 0.9 | Down from 1.3 |
| AI Research Scientist | 0.7 | Down from 0.9 |
| MLOps / AI Infrastructure | 1.4 | Down from 1.9 |
Source: LinkedIn Global Talent Trends 2025, Hired State of Software Engineers 2025, Robert Half Technology 2025
The multimodal AI engineer figure, 0.9 qualified applicants per open role, means the market is, in aggregate, operating with fewer qualified candidates than open positions. That is not a tight market. That is a market in which companies are structurally competing for candidates who do not yet exist in sufficient numbers for everyone who needs them to find one.
The direction of travel across every row in that table is the same: down. Every AI sub-discipline has seen its applicant-to-role ratio decline since 2023. That is not a demand spike resolving, it is a demand spike accelerating into a supply base that is not growing at the same pace.
What Counts as Qualified and Why That Matters
The qualified applicant figure deserves a specific note because it is doing significant definitional work. A qualified applicant for an ML engineer role, as defined across the sources used for this index, means someone with verifiable production ML experience, models shipped to real users, not personal projects or academic work. That filter removes the majority of candidates who apply with AI-adjacent backgrounds: data analysts who have run models, software engineers who have used ML APIs, and recent graduates with coursework but no production experience.
When the filter is loosened to include all applicants, verified experience or not, the ratios look dramatically different. General applicant-to-role ratios for AI roles run eight to twelve to one. The problem is that the conversion rate from general applicant to qualified candidate for senior AI roles runs at roughly 6–9%. The headline pipeline looks healthy. The functional pipeline is the crisis.
Where AI Talent Scarcity Is Most Acute
AI talent scarcity is not evenly distributed. It is concentrated geographically, institutionally, and to some extent generationally, in ways that shape the competitive dynamics for any company trying to hire.
AI Talent Concentration, Top Metro Areas, US + Europe (2025)
| Metro Area | Share of Regional Qualified AI Talent | Competition Intensity Score* |
|---|---|---|
| San Francisco Bay Area | 28% of US total | 9.8 / 10 |
| New York Metro | 14% of US total | 8.4 / 10 |
| Seattle / Redmond | 11% of US total | 8.1 / 10 |
| Boston / Cambridge | 8% of US total | 7.9 / 10 |
| London | 31% of Europe total | 9.2 / 10 |
| Berlin | 12% of Europe total | 7.8 / 10 |
| Paris | 10% of Europe total | 7.6 / 10 |
| Amsterdam | 7% of Europe total | 7.1 / 10 |
| Warsaw | 6% of Europe total | 6.4 / 10 |
| Stockholm | 5% of Europe total | 6.8 / 10 |
Competition Intensity Score = composite of applicant-to-role ratio, salary inflation rate, and offer rejection rate normalised to 10-point scale.
Source: LinkedIn Global Talent Trends 2025, PwC Global AI Jobs Barometer 2025, Glassdoor
Sixty-two percent of all globally qualified AI talent sits in six metro areas: San Francisco, New York, Seattle, Boston, London, and Berlin. That concentration is the product of where the research institutions are, where the major AI labs are headquartered, and where the first generation of production ML engineers built their careers. It does not reflect where companies hiring AI talent are located, which is everywhere.
The practical consequence of that geographic mismatch is that companies outside those six metros are not just competing for AI talent in a tight market. They are competing for AI talent that has every structural incentive to stay in the cluster, peer networks, career infrastructure, lab adjacency, and in the US, the highest compensation packages available anywhere.
London as Europe’s AI Talent Epicentre
London’s position as the primary concentration point for European AI talent reflects several reinforcing factors. DeepMind’s presence, the density of fintech and enterprise AI adoption, proximity to Oxford and Cambridge research output, and a long-established culture of international technical talent recruitment have made London the default landing point for AI-specialised engineers moving into European markets.
The competition intensity score of 9.2 out of 10 for London means that hiring AI talent there in 2025 is nearly as competitive as hiring in San Francisco. Companies entering the London AI talent market without a compelling employer brand, a strong remote flexibility policy, and compensation benchmarked to current rather than 2023 data are not running a difficult search, they are running an unsuccessful one.
The Eastern European AI Talent Emergence
Warsaw’s appearance in the top ten, with 6% of European qualified AI talent, is the data point that most surprises hiring managers who associate Eastern Europe primarily with backend and QA engineering. Poland has been quietly building a credible AI talent base, partly through domestic university programs, partly through the return of Polish engineers who trained or worked in Western European AI labs, and partly through the relocation of Ukrainian AI talent following 2022.
Warsaw’s competition intensity score of 6.4, meaningfully below London and Berlin, reflects a market where qualified AI talent exists in genuine depth but has not yet been fully discovered by international buyers. For companies willing to hire remotely into Poland specifically for AI roles, the combination of lower competition intensity and lower salary expectations relative to Western European markets represents the most significant underutilised opportunity in the European AI hiring landscape.
Salary Inflation Rate : What AI Talent Scarcity Has Done to Compensation
The salary inflation rate for AI specialists between 2023 and 2025 is 47%. That figure, drawn from composite data across Glassdoor, Levels.fyi, and Hired, sits alongside a general software engineering salary inflation rate of 8% over the same period. The AI premium has not just held — it has expanded dramatically.
AI Specialist Salary Inflation : US + Europe, 2023–2025
| Role | 2023 Median (US) | 2025 Median (US) | Inflation Rate | 2023 Median (W. Europe) | 2025 Median (W. Europe) | Inflation Rate |
|---|---|---|---|---|---|---|
| ML Engineer (Senior) | $148,000 | $192,000 | +30% | €95,000 | €128,000 | +35% |
| LLM / NLP Specialist | $165,000 | $245,000 | +48% | €105,000 | €158,000 | +50% |
| Multimodal AI Engineer | $175,000 | $268,000 | +53% | €112,000 | €172,000 | +54% |
| MLOps Engineer | $142,000 | $198,000 | +39% | €88,000 | €122,000 | +39% |
| AI Research Scientist | $195,000 | $310,000 | +59% | €130,000 | €205,000 | +58% |
| AI Product Manager | $138,000 | $185,000 | +34% | €85,000 | €115,000 | +35% |
Source: Glassdoor, Levels.fyi, Hired State of Software Engineers 2025, PwC Global AI Jobs Barometer 2025
The AI research scientist figure of 59% salary inflation in the US over two years, is the one that registers most viscerally with hiring managers who approved budgets in 2023 and are now trying to hire against them. A role budgeted at $195,000 two years ago requires $310,000 to be competitive today. That is not a negotiation problem or a candidate being unreasonable. It is a market that has moved significantly faster than planning cycles could track.
The Inflation Differential Between Sub-Disciplines
The variation within the AI salary inflation data is as important as the overall rate. LLM and multimodal specialists have inflated at rates 20–25 percentage points above the already-high ML engineer baseline. The reason is the same across both: the commercial deployment of large language models and multimodal AI systems in 2023 and 2024 created immediate demand for engineers who understood those systems at a production level, and the number of engineers with that experience at the moment demand appeared was extremely small.
The population of engineers with genuine production LLM deployment experience not experimentation, not API integration, but full-scale deployment with monitoring, fine-tuning, and infrastructure management, was measured in the low thousands globally when demand for that capability emerged. The resulting auction dynamic drove compensation to levels that are still surprising even to people who track this market closely.
Contractor Rate Inflation Alongside Salary Inflation
The salary inflation in full-time roles has been matched and in some cases exceeded by day rate inflation in the AI contractor market. Senior ML engineers who were billing at $120–$140/hr in early 2023 are billing at $165–$200/hr in 2025. LLM specialists who entered the contractor market in 2023 at $150/hr are now quoting $220–$280/hr for engagements, with the strongest profiles commanding more.
For companies using contractor AI talent as a cost-management strategy relative to FTE salaries, the 2025 data suggests that the advantage has partially eroded. The contractor premium over FTE which historically ran at 20–30% to compensate for the absence of benefits and stability, has compressed as FTE salaries have risen faster than contractor rates in some sub-disciplines, and expanded in others where contractor scarcity has driven day rates above what the FTE market will bear.
Also read : Tech Layoff to Rehire: What the 2024–2025 Data Really Shows Across the US and Europe
Offer Rejection Rates : The Clearest Signal of AI Talent Scarcity
If salary inflation shows what AI talent scarcity costs, offer rejection rates show who holds the power in the negotiation. In 2025 that answer, for senior AI roles, is unambiguously the candidate.
Offer Rejection Rates — AI vs General Engineering, US + Europe (2025)
| Role | US Rejection Rate | Europe Rejection Rate | Change vs 2023 |
|---|---|---|---|
| General Software Engineer (Senior) | 28% | 24% | Stable |
| DevOps / Cloud Engineer (Senior) | 36% | 31% | Up 5pp |
| ML Engineer (Senior) | 54% | 48% | Up 14pp |
| LLM / NLP Specialist | 61% | 54% | Up 18pp |
| Multimodal AI Engineer | 67% | 59% | Up 22pp |
| AI Research Scientist | 72% | 65% | Up 24pp |
| MLOps Engineer | 49% | 43% | Up 12pp |
Source: Hired State of Software Engineers 2025, LinkedIn Global Talent Trends, Glassdoor
A 72% offer rejection rate for AI research scientists in the US means that nearly three in four extended offers are declined. The company that fills an AI research scientist role on its first offer attempt in 2025 is the exception. The company that extends three or four offers before closing one, burning recruiter time, engineering interview capacity, and hiring manager attention at every failed attempt , is the norm.
The 24 percentage point increase in AI research scientist rejection rates since 2023 is the steepest two-year movement in offer rejection data that the sources used for this index have on record for any technical sub-discipline. It is a direct reflection of the applicant-to-role ratio: when there are 0.7 qualified candidates per open role, the candidates who exist are simultaneously evaluating multiple offers. The one that gets accepted is almost never the first one extended.
Why Candidates Are Rejecting Offers
Offer rejection data broken out by stated reason, drawn from exit surveys and candidate feedback collected by Hired and LinkedIn shows a consistent pattern across both US and European markets.
Reasons for Offer Rejection — AI Roles, 2025
| Reason | US % | Europe % |
|---|---|---|
| Competing offer with higher compensation | 38% | 32% |
| Compensation below expectations | 24% | 28% |
| Remote / flexibility policy concerns | 18% | 21% |
| Company mission or direction misalignment | 12% | 14% |
| Process too slow — accepted elsewhere first | 8% | 5% |
Source: Hired State of Software Engineers 2025, LinkedIn Global Talent Trends
The most actionable finding in that table is the 8% of US offer rejections that happened because the process moved too slowly and the candidate accepted another offer first. That is not a compensation problem. It is a process problem and it is entirely preventable. Companies with streamlined AI hiring processes that move from final interview to written offer within 48 hours capture candidates that slower-moving competitors lose without ever understanding why.
The 62% combined share of rejections driven by compensation, either a competing offer or a gap between the offer and the candidate’s expectation, is where most hiring managers focus their attention. The data suggests they are right to. But the framing matters: the compensation gap is almost never a negotiation gap that a counter-offer can close. It is a benchmarking gap, the offer was constructed using data that was 12–18 months old, and the market has moved significantly in that window.
The Equity Dimension in AI Offer Rejections
For AI research scientists and senior ML engineers specifically, equity is a disproportionate driver of offer decisions, more so than in any other technical discipline. These candidates have watched peers at AI labs and well-funded AI startups generate significant equity outcomes, and they are evaluating offers with that context in mind. A total cash compensation package that is competitive on base and bonus but thin on equity is rejected at significantly higher rates than packages that lead with total compensation transparency.
Companies that present offers as salary-first and treat equity as a footnote are consistently losing to companies, often AI-native startups with less brand recognition but more aggressive equity packages that present total compensation as the primary number. The framing of the offer, not just its value, drives materially different outcomes in the AI talent market.
2026 Predictions : Where AI Talent Scarcity Goes From Here
The 2025 data establishes the current state of AI talent scarcity across the US and Europe. The forward-looking question is whether the supply gap closes, widens, or stabilises heading into 2026. The evidence points in one direction.
AI Talent Supply-Demand Forecast — 2026
| Metric | 2025 Actual | 2026 Projection | Direction |
|---|---|---|---|
| Qualified applicants per AI role | 1.3 | 1.1 | Worsening |
| AI specialist salary inflation (annual) | +47% cumulative | +18–22% additional | Continuing |
| Senior AI offer rejection rate (US) | 54% | 58–62% | Worsening |
| Geographic concentration (top 6 metros) | 62% | 59% | Modest dispersal |
| New qualified AI graduates entering market | — | +12% YoY | Growing but insufficient |
| AI role posting growth (YoY) | +163% | +80–100% | Slower but still growing |
Source: PwC Global AI Jobs Barometer 2025, LinkedIn Global Talent Trends, Stack Overflow Developer Survey 2025
The picture for 2026 is one of continued tightening at the top of the AI talent market with modest dispersal of geographic concentration. The applicant-to-role ratio is projected to fall further, from 1.3 to approximately 1.1 , as role posting growth continues to outpace supply additions from graduate pipelines.
The one moderating signal is geographic: the concentration of qualified AI talent in six metro areas is projected to decline modestly to around 59% as Eastern European markets, particularly Warsaw and Kyiv, and secondary US markets like Austin and Toronto develop deeper AI talent bases. That dispersal will not resolve the overall scarcity, but it will create pockets of relatively lower competition intensity that represent genuine opportunities for companies willing to hire across those geographies.
What Structural Resolution Actually Requires
The AI talent scarcity index will not correct until the graduate pipeline produces supply at a scale proportional to demand and that is a 2028–2030 story at the earliest under optimistic assumptions. Three things would accelerate the correction: a significant scaling of AI-specialised graduate programs at universities in both the US and Europe; a meaningful transfer of adjacent skills like strong mathematicians, physicists, and statisticians, into production ML roles through structured upskilling pathways; and a normalisation of AI tooling that reduces the specialisation depth required for some categories of AI engineering work.
All three are happening. None are happening fast enough to materially change the supply picture before 2027.
What This Means for Hiring Strategy in 2026
Companies that accept the supply constraint as a fixed variable and design their hiring strategy around it will outperform those still waiting for the market to normalise. Practically, that means four things.
Building internal AI capability through upskilling existing engineers, identifying strong ML-adjacent talent inside the organisation and investing in structured development paths, rather than relying exclusively on external hiring. Expanding geographic reach to markets with lower competition intensity, specifically Eastern Europe and secondary US cities, where qualified talent exists in smaller absolute quantities but with significantly better odds of a successful close.
Restructuring compensation reviews to run quarterly rather than annually for AI roles, so that offers are benchmarked against current rather than historical data. And investing in employer brand specifically within the AI research and ML engineering communities, academic partnerships, open-source contributions, and published technical work, because in a market where candidates hold the power, reputation within the community is a more durable competitive advantage than any individual compensation package.
