Is There an AI Talent Shortage?
The sentence appears in nearly every article written about AI hiring in the last two years: there are not enough AI engineers. It is repeated by recruiters, by vendors, by consultancies, and by us. This article is a check on whether the evidence behind it says what everyone assumes it says.
It mostly does not. The claim is not fabricated, and the underlying market is genuinely difficult. But the specific proposition that there are too few AI engineers in the world is not supported by the figures used to support it, and the gap between the claim and the evidence changes what a company should do about it.
A Shortage Is a Ratio, Not a Trend
The word shortage describes a relationship between two quantities: how much of something is wanted and how much of it exists. Neither number means anything alone. A hundred thousand open roles is a shortage if there are fifty thousand engineers and a surplus if there are two hundred thousand. You cannot tell which from the first number.
This is worth stating plainly because the distinction disappears constantly in practice. Demand growth is emotionally persuasive. A chart going up looks like scarcity, and it reads like scarcity, and it is not scarcity. It is one half of a fraction with the denominator missing.
Every shortage claim therefore carries an implicit second number. When an article says the pipeline cannot keep pace, it is asserting something about supply. The question is whether it measured that, or assumed it.
What the Circulating Evidence Actually Measures
The most cited figure in AI hiring is job posting growth, drawn from platform data on how many roles employers opened year over year. It is a real measurement of a real thing, and the thing it measures is employer behaviour.
Posting growth is sensitive to factors that have nothing to do with the labour pool. Companies post more roles when they are funded, when a technology becomes board-level priority, and when they are unsure what they need and so advertise widely. A single role can generate several postings across job boards. Postings also do not have to be filled, or funded, or real; a proportion of any posting dataset consists of roles that were never going to be hired for. None of that makes posting data useless. It makes it a demand indicator, which is what the organisations publishing it say it is. LinkedIn's Economic Graph, the research arm behind the most quoted of these figures, describes its work as labour market insight drawn from platform activity, which is an accurate description of a demand-side instrument.
The same applies to the other evidence usually offered. Salary movement measures what employers paid, which reflects competition, urgency, and the funding environment. Time-to-fill measures how long a process took, which reflects the process. Both are outcomes of the market, not counts of the people in it.
We went looking for the other half. Across the published sources that assert an AI talent shortage, we could not find a single supply-side figure: no count of qualified engineers, no measure of pool size, no estimate of how many people can demonstrably ship production AI systems. Every number was a measure of demand. This article applies one standard to itself: it cites only figures we could trace to a named publication we were able to reach and read. Where a figure fails that test, it is left out.
Nobody Counts AI Engineers, Including the Government
A reasonable next question is why nobody simply looks the number up. The answer is that the occupation does not formally exist in the statistical system.
The US Bureau of Labor Statistics publishes no occupation code for AI engineer. There is no BLS median, no BLS employment level, and no BLS projection for the role, because the Standard Occupational Classification does not contain it. Writers who need a figure typically substitute Software Developers, SOC 15-1252. As a salary bracket that substitution is defensible and we use it ourselves. As a headcount it is close to meaningless, because that occupation contains millions of people, the overwhelming majority of whom do no AI work.
This is not a criticism of BLS. Classification systems update slowly by design, since the value of a time series is that its definitions hold still. It does mean that anyone quoting a number of AI engineers is quoting an estimate built on a definition they chose, and different definitions produce answers that differ by an order of magnitude. Is a backend engineer who integrates a model API an AI engineer? Different studies answer differently, and the answer moves the total more than any real change in the workforce.
The most careful public work in this area, including the Stanford AI Index economy chapter, reports AI labour market activity through the same demand-side instruments and is explicit about what it is measuring. The rigour is in the framing. It gets lost when the figures are quoted secondhand.
Where the Claim Does Hold
None of the above means AI hiring is comfortable. It means the difficulty has been misdiagnosed.
What the evidence supports is a shortage at a price, in a place. In the US market, at US salary levels, competing against well-funded labs, a company with an ordinary budget will struggle to hire an experienced AI engineer quickly. That is real, it is well documented by the salary and time-to-fill data, and nothing here disputes it.
That is a materially different claim from global scarcity, and the difference is not academic. Consider what each diagnosis implies:
| Diagnosis | What it implies | Rational response |
|---|---|---|
| Global scarcity of skill | The people do not exist anywhere. Waiting will not help, and neither will searching harder. | Pay more, wait longer, or abandon the project. |
| Scarcity at US prices in the US market | The people exist, but not at that price, in that place, on that timeline. | Widen where you look, or change what you are willing to pay for. |
| A search and screening problem | Candidates exist and are hard to identify because titles are inconsistent and references are thin. | Improve the brief and the assessment before widening the budget. |
Companies that accept the first diagnosis stop looking earlier than the evidence justifies. That is the practical cost of the imprecision, and it is why this is worth arguing about rather than letting it pass.
The Supply Side Nobody Cites
Here is the asymmetry that makes the one-sided evidence conspicuous. Supply-side measurement is not impossible. It is simply not done in the articles making the shortage claim, even though the instruments exist.
India publishes technical education statistics through the All India Council for Technical Education and higher education participation through the government's All India Survey on Higher Education. The industry body NASSCOM reports on the technology workforce. These are the same kind of instruments that would be used to establish a shortage if anyone wanted to establish one rather than assert it.
We are deliberately not pulling headline numbers out of those sources and putting them in this article, because doing so would repeat the error the article is about. A graduate count is not a count of production AI engineers any more than a posting count is. What those sources establish is narrower and still decisive: large technical pipelines exist outside the United States, they are measured by public bodies, and no serious attempt has been made to set them against global AI demand. Until someone does, the honest description of the situation is that one side of the ratio is well instrumented and the other is not.
What Would Settle It
A defensible shortage claim needs one thing that does not currently exist: a count of people with demonstrable production AI experience, set against open roles, using a single definition applied to both sides.
That study would need to state its definition explicitly, since the definition determines the answer. It would need to count globally, because labour for remote roles is not confined to one country. And it would need to distinguish people who can build AI systems from people who list AI skills, which is the hard part and the reason nobody has done it.
Until that exists, the position this article takes is the narrow one: demand is well measured, supply is not measured at all in the public discussion, and claims about the ratio are therefore unsupported in either direction. We are not asserting that there is no shortage. We are asserting that the evidence offered does not establish one, which is a different and more defensible statement.
What This Means for Your Hiring
If you are hiring and have absorbed the shortage narrative, three things follow.
Stop treating the market as closed. The evidence for global scarcity is absent, not negative. Companies that widen their search geographically routinely find candidates, which is difficult to reconcile with a genuine worldwide shortage of the skill.
Separate the price problem from the skill problem. If you cannot hire at your budget in your city, that is a pricing and location constraint, and it has direct remedies. If you cannot identify a good candidate when you see one, that is a screening problem, and no budget fixes it.
Be suspicious of any number offered without a denominator. Apply it to every source, this one included. If a statistic tells you how much demand grew, it has told you nothing about scarcity, however large it is.
The Bottom Line
The AI talent shortage is real as a description of the US market at US prices and unproven as a description of the world. The distinction matters because the two have different remedies, and the stronger claim discourages the search that the weaker claim rewards.
F5 Hiring Solutions places AI engineers from India as a managed remote workforce, starting at $600 per week, all-inclusive, with a shortlist in 7-14 business days. That model rests on the narrow claim, not the broad one: the engineers exist, and the constraint most US companies hit is where and at what price they are looking. If you want to check the shape of your own problem first, our note on how long it actually takes to hire an AI engineer sets out where the time genuinely goes, and why AI projects fail on the talent gap covers how much of the failure rate talent explains and how much it does not.
To scope a role, hire remote AI and ML engineers from India or book a 15-minute call with Joel Deutsch at https://calendly.com/joel-f5hiringsolutions/f5.