Where is AI talent concentrated in the US?

The Bay Area, Seattle, New York, and Boston hold the largest pools, and the Bay Area still dominates for research and frontier model work. For applied AI engineering, the talent is far more distributed, because the skill is built on top of ordinary software engineering, and strong applied AI engineers work remotely from most large US metros and from Canada. A startup that insists on one city is shrinking its market for the role it most likely needs.

How much does it cost to hire AI talent in the US?

Applied AI engineers at Series A and B companies are paid at a premium to equivalent software engineers, and research scientists with frontier-lab experience are paid at a level most startups cannot reach. The premium is smaller than headlines suggest for applied roles and larger for research roles, which is one more reason to be precise about which one you are hiring.

Should we hire AI talent or train our existing engineers?

For most applied AI work, train first. A strong software engineer with a few months of serious work on retrieval, evaluation, and model integration becomes an applied AI engineer, and they already know your product. Hire externally when the work needs experience you cannot build in time: production evaluation systems, cost and latency at scale, or a first hire who sets the pattern the rest of the team learns from.

Can a startup hire AI talent from outside the US?

Yes, and many do. Canada is the most common source for US startups because of time zone overlap and a deep applied AI market, and remote hiring from Europe and Latin America is routine for engineering roles. What changes is the closing process: candidates outside the US weigh equity differently, and the compensation calibration has to be done against their market, not against San Francisco.

How do we evaluate AI candidates when nobody on our team has done the work?

Have someone who currently does the work screen the shortlist before your interviews. AI titles are inconsistent enough that a keyword screen is close to useless, and a founder without the background will over-weight vocabulary and research credentials. The interview itself should be a realistic problem with your data: build the retrieval, set up the evaluation, explain the cost tradeoffs.

How to Hire AI Talent in the US

Q: How do I hire AI talent in the US?

Decide first whether you need research talent or applied AI engineering, because they are different markets with different prices and most startups need the second. Then source across the US and Canada rather than one city, calibrate pay against applied engineering rather than frontier-lab headlines, consider training a strong engineer before hiring externally, and have a practitioner currently doing the work screen candidates before your interview loop.

September 20, 2026 · Patrick Dyer

To hire AI talent in the US, decide first whether you need research talent or applied AI engineering, because they are different markets with different prices and most startups need the second. Then source across the US and Canada rather than one city, calibrate pay against applied engineering rather than frontier-lab headlines, consider training a strong engineer before hiring externally, and have a practitioner currently doing the work screen candidates before your interview loop.

"AI talent" is the vaguest label in technical hiring right now, and the vagueness is expensive. It covers people who train models, people who build products on models, people who run the data and infrastructure underneath both, and an increasing number of people who have added the words to a profile. The search only works once you know which of those you are looking for.

Which AI talent you actually need

Profile	What they do	US market	Most startups need
Research scientist	Trains and improves models, publishes, works at the frontier	Very small, concentrated in a few labs and cities, priced beyond most startups	Rarely, unless the model is the product
ML engineer	Trains, fine-tunes, and serves models in production; owns pipelines and infrastructure	Established market, competitive, mostly at companies with proprietary data at scale	Sometimes, once there is data worth training on
Applied AI engineer	Builds product on top of existing models: retrieval, evaluation, guardrails, cost and latency	Growing fast, distributed across the country, built on ordinary software engineering	Yes, almost always first
AI product manager, forward deployed engineer	Shapes what the AI feature should do; makes it work inside a customer's environment	New roles with unstable titles; the candidates are often not searching under those names	Once the product is in customers' hands

The most common mis-hire in this space is writing a job description for the second row and expecting the third, or the reverse. The distinction is laid out in what an applied AI engineer is. If the company's product is built on models it does not train, the applied profile is the hire, and it is a far larger and more affordable market than the headlines about AI talent wars suggest.

Where the talent is

Research and frontier work is concentrated: the Bay Area first, then Seattle, New York, and Boston, with the largest labs and the universities that feed them. If that is the profile you need, you are competing in a handful of cities against employers with compensation you cannot match, and the honest strategy is to find the people those employers overlook: strong practitioners without the famous logo, or people who want to see their work ship rather than be published.

Applied AI engineering is a different geography. Because the skill sits on top of software engineering, the people who have it are wherever strong software engineers are, which is most large US metros and a great deal of Canada. Toronto, Montreal, and Vancouver have deep applied AI markets, overlap US working hours, and are the most common source of AI hires for US startups that recruit across the border. A search restricted to one city for an applied role is throwing away most of the market.

What it costs

Pay for AI roles has a wide spread, and the spread follows the table above. Research scientists with frontier-lab experience are paid at levels that only a few companies can sustain. Applied AI engineers at Series A and B startups are paid at a premium to an equivalent software engineer, but the premium is a fraction of what the research end commands, and it narrows as more engineers add the skill. The mistake is to calibrate an applied search against research-lab figures, which produces an offer that is either uncompetitive for the wrong role or wildly overpriced for the right one. Where AI-era engineering roles actually land is covered in what AI-era engineering roles pay.

Equity matters more here than in most searches, because the strongest applied AI candidates are choosing between a startup and a large company offering more cash. Meaningful equity, real scope, and the ability to ship to production without a review committee are what a startup can offer that a large company cannot, and they should be in the first conversation.

Train or hire

For applied work, the strongest first move for most startups is not a hire. A senior software engineer already on the team, given a real AI feature to own and a few months to do it properly, becomes an applied AI engineer who also knows your product and your customers. That is faster and cheaper than an external search, and it establishes the standard the eventual hire will be measured against.

Hire externally when the work needs experience that cannot be built in time: a production evaluation system, cost and latency engineering at scale, or a first hire whose job is to set the pattern for a team that will grow. The same logic applies at leadership level, where the answer to whether to hire a Head of AI is usually not. AI fluency at a startup is a bar applied to every technical role, not a department.

Assessing candidates when nobody internally has done the work

This is the part of the search that goes wrong most often. AI titles are inconsistent, the vocabulary is easy to acquire, and a founder without the background will over-weight research credentials, famous employers, and fluency in the jargon, none of which predicts whether someone can ship a reliable AI feature.

Practitioner screening first. Have someone currently doing applied AI work read the shortlist before your team interviews anyone. They can distinguish shipped systems from adjacent ones in ways a CV review cannot.
A realistic problem with your data. Not a whiteboard, not a take-home essay. Set up retrieval over a sample of your content, build a small evaluation, explain what would make it cheaper. Watch for the judgment calls, not the completion.
Ask about failure. Every production AI system has failed in a way that surprised its builders. Candidates who have actually shipped one can describe theirs in detail. Candidates who have not will describe a hypothetical.
Test the tradeoff instinct. Cost, latency, and quality pull against each other in every AI feature. Ask which they would give up first for your product, and why. The answer reveals whether they think like an engineer or a demo builder.

Common questions

Where is AI talent concentrated in the US?: The Bay Area, Seattle, New York, and Boston hold the largest pools, and the Bay Area still dominates for research and frontier model work. For applied AI engineering, the talent is far more distributed, because the skill is built on top of ordinary software engineering, and strong applied AI engineers work remotely from most large US metros and from Canada. A startup that insists on one city is shrinking its market for the role it most likely needs.
How much does it cost to hire AI talent in the US?: Applied AI engineers at Series A and B companies are paid at a premium to equivalent software engineers, and research scientists with frontier-lab experience are paid at a level most startups cannot reach. The premium is smaller than headlines suggest for applied roles and larger for research roles, which is one more reason to be precise about which one you are hiring.
Should we hire AI talent or train our existing engineers?: For most applied AI work, train first. A strong software engineer with a few months of serious work on retrieval, evaluation, and model integration becomes an applied AI engineer, and they already know your product. Hire externally when the work needs experience you cannot build in time: production evaluation systems, cost and latency at scale, or a first hire who sets the pattern the rest of the team learns from.
Can a startup hire AI talent from outside the US?: Yes, and many do. Canada is the most common source for US startups because of time zone overlap and a deep applied AI market, and remote hiring from Europe and Latin America is routine for engineering roles. What changes is the closing process: candidates outside the US weigh equity differently, and the compensation calibration has to be done against their market, not against San Francisco.
How do we evaluate AI candidates when nobody on our team has done the work?: Have someone who currently does the work screen the shortlist before your interviews. AI titles are inconsistent enough that a keyword screen is close to useless, and a founder without the background will over-weight vocabulary and research credentials. The interview itself should be a realistic problem with your data: build the retrieval, set up the evaluation, explain the cost tradeoffs.

Hiring AI talent for a US or Canadian startup?

We settle which profile the work needs before the search opens, source across both countries, and have working AI practitioners screen every shortlist.

AI recruiting for AI-native companies