Skip to content
Report · United States

The US AI talent market, mapped

Most AI hiring difficulty is a specification problem, not a supply problem. This is the five-tier structure of the US market, where each tier actually sits, how its pay behaves, and which adjacent populations are being overlooked.

Report By the Audentia Research desk Published 12 min read US market focus
Share on LinkedIn

Key facts

Direct answer
The US AI talent market is not one market. A very small frontier-research tier sits above a much larger applied tier, and most hiring difficulty comes from writing a research specification for an applied job.
Five tiers
Frontier research, applied research, ML/AI engineering, ML platform and infrastructure, and AI product and safety. Supply, pay and mobility differ completely between them.
Deepest pool
ML and AI engineering — large, growing quickly, and behaving much like senior software engineering, because to a significant degree that is what it is.
Geographic concentration
Extreme by US standards. The Bay Area dominates the research tiers; Seattle, New York, Boston, Austin, Pittsburgh and the Research Triangle hold distinct and specialized populations.
Density over cost
A metro 25% cheaper that holds one fifth of the qualified population is not cheaper once search duration and failure rate are priced in.
Compensation structure
Equity frequently exceeds base plus bonus in the research tiers, the distribution is bimodal rather than normal, and bands age materially within two quarters.
Adjacency
Distributed systems, quantitative finance, computational science, data engineering and non-CS post-docs are the reachable populations most employers overlook.
Top retention risk
Work that never reaches production, not compensation. A year of prototypes nothing depends on is a credential problem for strong applied people.
The shape of it

What the US AI talent market actually looks like

Direct answer

The US AI talent market is not one market. It is a narrow, severely constrained research and frontier-model tier sitting on top of a much larger, far less constrained applied tier — and most organizations are trying to hire from the first when they need the second.

That mismatch, rather than absolute scarcity, is what makes AI hiring feel impossible. The population of people who can train a frontier model from scratch is genuinely tiny and effectively unhirable for most employers. The population who can take a foundation model, build a reliable production system around it, and keep it running is far larger, far more reachable, and growing quickly. Confusing the two produces a twelve-month search for a role that did not need to be specified that way.

The specification test. Before concluding that AI talent is unobtainable, write down what the person will actually do in their first year. If the honest answer is “integrate models into our product and make it reliable at scale,” you are hiring from the applied tier and the market is much better than the headlines suggest. If the answer is genuinely “advance the state of the art,” the market is as difficult as you have been told, and you should plan accordingly rather than hoping.

Segmentation

Five tiers, five completely different markets

These tiers are routinely collapsed into one requisition. They should not be, because their supply, pay and mobility characteristics have almost nothing in common.

Tiers of the US AI and machine learning talent market, with supply and mobility characteristics.
TierWhat they doUS supplyMobility
Frontier researchNovel architecture and training-methodology research at the edge of published work.Very small — low thousands nationally, concentrated in a handful of labs and universities.Very low. Compensation is rarely the binding constraint; access to compute, data and collaborators usually is.
Applied researchAdapting published methods to a specific domain, fine-tuning, evaluation design.Small but real — tens of thousands, spread far more widely than tier one.Moderate. Responsive to problem interest and to genuine data access.
ML / AI engineeringBuilding, serving and maintaining models and model-backed systems in production.Large and growing fast. The deepest pool in the market by a wide margin.High. This tier behaves like senior software engineering, because that is largely what it is.
ML platform & infrastructureTraining and inference infrastructure, GPU orchestration, evaluation and deployment pipelines.Moderate, and the most persistently under-hired tier relative to need.Moderate to high. Frequently reachable from adjacent distributed-systems and SRE populations.
AI product & safetyProduct management for model-backed features, evaluation, red-teaming, governance and risk.Small but expanding quickly, and drawn from unusually varied backgrounds.High. The least title-standardized tier in the market, which makes it the hardest to search for and the easiest to map.

The most common scoping error. Writing a tier-one specification for a tier-three job. A requisition asking for publications at major conferences, distributed training experience at scale, and production ownership of a customer-facing service describes perhaps a few dozen reachable people nationally. Dropping the publication requirement — which the actual work does not need — can expand the reachable population by an order of magnitude without lowering the bar on anything that matters.

Geography

Where US AI talent actually sits

Concentration in this market is extreme by US standards. The Bay Area holds a disproportionate share of the frontier and applied research tiers — not because the people were born there, but because the labs, the compute and the funding are. Seattle follows, largely on the strength of cloud and large-scale infrastructure. New York has built a substantial applied and AI-product population on the back of financial services, media and healthcare. Boston is strongly weighted toward life-sciences and academic-adjacent research. Austin, Denver, San Diego, Pittsburgh and the Research Triangle each hold meaningful, specific pockets.

Three things follow from that, and they matter more than the ranking itself.

  • Density beats cost in this market. A metro 25% cheaper that holds one fifth of the qualified population is not cheaper, because the search takes three times as long and fails more often. In markets this tight, talent density should dominate the location decision.
  • Remote widens the applied tier substantially and the research tier barely at all. Tier-three engineers are widely distributed and comfortable remote. Tier-one researchers cluster around compute, collaborators and institutions, and remote work does not relocate those.
  • The secondary metros are specialized, not smaller versions of the primary ones. Pittsburgh is not a small Bay Area; it is a distinct population with distinct strengths. Mapping a secondary metro against a Bay Area comparator set produces a misleading answer in both directions.

Why published rankings are the wrong tool here. Metro rankings count job postings or self-reported profile keywords. Neither measures how many people could actually do your job, and in AI the keyword noise is worse than in any other function — profile self-description has run well ahead of demonstrated capability. Counting the population against a defined capability standard is a mapping exercise, and in this market it is the only method that produces a number you can plan against.

Compensation

How AI compensation behaves differently

AI compensation is not simply higher than comparable software engineering pay. It is structurally different in four ways, and each one breaks a standard benchmarking assumption.

Equity dominates at the top

In the research tiers, equity frequently exceeds base and bonus combined. A benchmark built on base salary does not merely understate these packages — it ranks them in the wrong order.

The distribution is bimodal

A small set of employers pays far above everyone else for the same nominal title, producing two clusters rather than a bell curve. A median across both describes nobody, and a peer set that mixes them produces an unusable band.

Bands age in months

In tiers one and two, a benchmark more than two quarters old is materially unreliable. Annual survey cycles cannot track this, which is why primary research earns its cost here more clearly than almost anywhere else.

Non-cash terms are decisive

Compute budget, publication freedom, data access and problem selection genuinely move decisions in the research tiers. Employers who can only compete on cash tend to lose candidates they had assumed were priced in.

The practical consequence: benchmark AI roles against a peer set drawn from the tier you are actually hiring from, on total compensation, refreshed at least every six months. The general method is set out in our US compensation benchmarking guide; the difference here is cadence and peer-set discipline, not technique.

Supply expansion

The adjacent populations most employers ignore

In a market this tight, the reachable population is usually several times larger than the exact-match population — if you are willing to define the role by capability rather than by prior title.

01

Distributed systems engineers

The scarcest competence in ML platform work is running large distributed systems reliably, not knowing model internals. That population is large, and the model-specific knowledge is a two-to-three-month ramp.

02

Quantitative finance

Deep applied statistics, production modeling under real consequences, and a strong evaluation culture. Frequently overlooked because the domain language differs, not the capability.

03

Computational science and bioinformatics

Large-scale numerical computing, GPU familiarity and rigorous experimental design, often with more methodological discipline than the tier-three average.

04

Data engineering

Model quality is a data-pipeline problem far more often than a model-architecture problem. Strong data engineers move into ML engineering more successfully than their interview performance usually predicts.

05

Academic post-docs outside CS

Physics, statistics, computational neuroscience and operations research produce people with the mathematical foundation and the research habits. The gap is engineering practice, which is teachable.

06

Senior software engineers with evaluation instincts

For AI product and safety work, the binding competence is rigorous evaluation design and judgment about failure modes. That is found in senior engineering and in testing-heavy disciplines, not only in ML backgrounds.

How to use this properly. Adjacency is not a lowering of the bar; it is a correction of the search. The disciplined version is to write down which capabilities are genuinely non-negotiable on day one, which can be acquired within ninety days, and then to count both populations. If the adjacent pool is four times the exact-match pool — which in AI engineering it very often is — the constraint was the specification rather than the market. The same diagnostic is set out in skills that are getting scarce.

Retention

Why AI teams leave, and what actually holds them

Attrition in AI teams is high, and compensation is a less reliable explanation than employers assume. Four patterns recur across US organizations.

  • The model never shipped. The most common reason strong applied people leave is that their work does not reach production. A year of prototypes that nothing depends on is a credential problem for them, and they know it.
  • Compute and data access. In research tiers, the constraint that drives departures is usually infrastructure rather than salary. An employer that cannot supply compute is not competing, however well it pays.
  • The vesting cliff. Equity-heavy packages concentrate mobility at predictable moments. Teams where several people joined in the same quarter have a correlated retention risk that nobody has diarized.
  • Organizational placement. AI teams buried three layers below the decision they are meant to inform lose senior people fastest. This is a reporting-line problem, and no compensation adjustment fixes it.

All four are observable from outside, which makes them useful in both directions. Competitor teams showing these patterns are reachable, and your own team showing them is a retention problem you can act on before it becomes a search.

Method

How to build an AI hiring plan that survives contact with the market

Assign each role to a tier

Before writing a specification, decide which of the five tiers the role sits in. Most organizations discover that two thirds of what they had called AI roles are tier three, where the market is far more tractable than they believed.

Separate day-one requirements from ninety-day requirements

Write both lists explicitly. The second list is where the adjacency argument lives, and it is the difference between a 40-person and a 400-person addressable market.

Count both populations

Map the exact-match and adjacent pools in your target metros. This is the step that converts an argument about difficulty into a number you can plan against.

Benchmark against the right tier

Use a peer set from the tier you are hiring from, on total compensation, refreshed at least twice a year. Mixing tiers produces a median that loses offers at the top and overpays at the bottom.

Sequence approaches against vesting

In equity-heavy populations, timing is a large part of reachability. Knowing where individuals sit in their vesting schedule turns a cold approach into a well-timed one — the basis of effective pipelining in this market.

Fix what makes the role hold

Reporting line, compute access and a credible path to production do more for both attraction and retention than the last 10% of cash. Establish them before the search, not after the second decline.

Answers

AI talent market: frequently asked

Is there really an AI talent shortage in the US?

It depends entirely on which tier you are hiring from. The frontier research population, capable of advancing model architecture and training methodology, is genuinely tiny and effectively unhirable for most employers. The applied tier, which builds and operates production systems around existing models, is large and growing quickly. Most reported AI shortages turn out to be specification problems: a research-level requirement written into a job that does not need it.

What are the tiers of the US AI talent market?

Five. Frontier research, which advances the state of the art. Applied research, which adapts published methods to a domain. ML and AI engineering, which builds and serves models in production. ML platform and infrastructure, covering training and inference systems, GPU orchestration and deployment pipelines. And AI product and safety, covering product management for model-backed features, evaluation, red-teaming and governance. Their supply and mobility characteristics have almost nothing in common.

Where is AI talent concentrated in the United States?

Concentration is extreme by US standards. The Bay Area holds a disproportionate share of the frontier and applied research tiers because the labs, compute and funding are there. Seattle follows on cloud and large-scale infrastructure strength. New York has a substantial applied and AI-product population from financial services, media and healthcare. Boston is weighted toward life sciences and academic-adjacent research. Austin, Denver, San Diego, Pittsburgh and the Research Triangle hold meaningful specialized pockets rather than smaller versions of the primary metros.

Can we hire AI engineers remotely?

For the applied tiers, largely yes. ML and AI engineers are widely distributed across US metros and generally comfortable working remotely. For the research tiers the effect is much weaker, because those populations cluster around compute, collaborators and institutions, and remote work does not relocate any of those. Remote materially widens the applied pool and barely widens the research pool.

Why is AI compensation so hard to benchmark?

Four reasons. Equity frequently exceeds base plus bonus in the research tiers, so base-salary benchmarks rank packages in the wrong order. The distribution is bimodal rather than normal, because a small group of employers pays far above the rest for the same nominal title, so a median across both clusters describes nobody. Bands move materially within two quarters, which annual survey cycles cannot track. And non-cash terms such as compute budget and publication freedom genuinely move decisions.

What adjacent backgrounds can move into AI roles?

Distributed systems engineers are the strongest source for ML platform work, because the scarce competence there is running large systems reliably rather than knowing model internals. Quantitative finance, computational science and bioinformatics bring applied statistics and rigorous evaluation habits. Data engineers move into ML engineering more successfully than interviews usually predict, because model quality is a data pipeline problem more often than an architecture problem. Non-CS post-docs in physics, statistics and operations research bring the mathematical foundation, with engineering practice as the teachable gap.

How do we know whether our AI role is genuinely hard to fill?

Count both populations. Establish how many people meet the non-negotiable day-one requirements in your target metros, then how many meet all but one requirement that could be closed within ninety days. If the adjacent pool is several times larger, the constraint is your specification rather than the market. In AI engineering the adjacent pool is very often four or more times the exact-match pool.

Why do AI teams have high attrition?

The most common reason strong applied people leave is that their work never reaches production, which is a credential problem for them rather than a pay problem. In research tiers, compute and data access drive departures more reliably than salary. Equity-heavy packages concentrate mobility at vesting cliffs, creating correlated risk when several people joined in the same quarter. And AI teams placed several layers below the decision they are meant to inform lose senior people fastest, which no compensation adjustment fixes.

Are published AI talent rankings reliable?

They are directionally interesting and operationally weak. Metro rankings generally count job postings or self-reported profile keywords, and neither measures how many people could actually do a specific job. Keyword noise is worse in AI than in any other function, because profile self-description has run well ahead of demonstrated capability. Counting a population against a defined capability standard requires primary research, not aggregation.

Should we build AI capability internally or hire it?

Usually both, but the decision should follow the count rather than precede it. If the adjacent internal population who could move into the role within ninety days is larger than the external exact-match population in your metro, building is faster and cheaper than hiring, and it retains people you already have. If neither pool exists at the level you need, the honest answer is that the plan needs re-phasing rather than a better recruiter.

How often should AI compensation benchmarks be refreshed?

At least every six months for the applied tiers and every quarter for the research tiers. A benchmark more than two quarters old is materially unreliable in the upper tiers, and using one is how employers end up losing candidates while believing their band is competitive.

What is the first thing to fix in an AI hiring plan?

Assign every role to a tier before writing a single specification. Most organizations find that around two thirds of what they had labelled AI roles are applied engineering roles, where the market is far more tractable than the headlines suggest. That reclassification, on its own, usually changes the hiring plan more than any sourcing change would.

Audentia Research desk

Talent research & intelligence

This report sets out the structural view Audentia applies when mapping AI and machine learning populations for US clients. It describes market structure and research method rather than presenting a numerical market survey.

Audentia has been conducting talent research since 2012, works on a fixed project fee with no placement commission, and hands every dataset to the client to keep. Questions about the method behind this page can go to sales@audentiaresearch.com.

This page describes market structure observed across Audentia Research’s US AI and machine learning mapping projects as at September 2026. It deliberately contains no headcount, salary or growth-rate statistics, because any such figure in this market is stale within a quarter and none should be attributed to Audentia unless it appears in a commissioned project report. Metro characterisations describe relative concentration, not measured population counts.

Count the population before you accept that it does not exist

Tell us the capability, not the title, and the metros that are genuinely in scope. We will count the exact-match and adjacent populations and tell you which of the two your plan actually depends on.

sales@audentiaresearch.com  ·  USA: +1 929 235 1786  ·  UK: +44 2038 077392