AI/ML Job Postings Are Up 163% Year Over Year and the Talent to Fill Them Doesn't Exist Yet

Every hiring cycle has its category of role that seems to defy the usual rules of supply and demand, the one where the reqs multiply, the salaries climb, and the offers keep going out even as the pipeline visibly runs dry. This cycle, that category is AI and machine learning. What almost nobody is planning for is that the qualified candidates to fill those roles don't exist in the numbers the demand implies.

Now, the market is trying to hire faster than the pipeline can produce.

More roles, same small pool

Recent data puts AI and machine learning job postings up 163% year over year, with 78% of tech leaders reporting plans to increase headcount in the second half of 2026. Those are the demand-side numbers, and they're stunning on their own. The supply-side numbers are the ones getting ignored.

There isn't a corresponding growth on the candidate side. Ph.D. programs in machine learning graduate a fixed number of students per year, and that number was set by admissions committees three to five years ago. Applied ML boot camps and certification programs are growing, but not at rates that close the gap. The self-taught pipeline of engineers cross-training into ML from adjacent roles is real, but slow.

Every company posting those roles is fishing in the same small pond. Unfortunately, the pond didn't get any bigger between 2025 and 2026.

The requirements problem

Looking at the job descriptions, the mismatch gets worse. Companies are posting for senior ML engineers with five years of experience deploying LLM-based systems in production. LLMs weren't in production at most companies five years ago. The requirements the market is asking for describe a candidate profile that didn't exist when the requirements were written.

The same pattern shows up across the AI/ML posting boom. Job specs written by pattern-matching to elite candidates at Anthropic, OpenAI, and Google are getting cloned across companies that won't ever compete for those candidates in the first place. The result is a market where every posting demands the same top-1% profile and every company wonders why nobody's applying.

The role itself keeps moving, too. What an AI/ML engineer needs to know in Q3 2026 is materially different from what the same title required in Q3 2025. New model releases, new frameworks, new deployment patterns. The hiring team writing a spec today is describing a job that'll have changed by the time the offer letter's signed, and the candidate pool is chasing the same moving target.

Some of the shortage is genuine skill scarcity. Much of it is manufactured scarcity, generated by hiring teams writing requirements that would rule out most of the people currently building AI systems.

The salary spiral

The predictable second-order effect is already happening: salaries for AI/ML roles have detached from the rest of the technical labor market. Meta reportedly offered signing bonuses in the tens of millions to poach senior researchers from competitors. Numbers at that tier reset the market for everyone underneath them, and every ML engineer now benchmarks compensation against those headlines, whether or not the offer is realistic for their role.

The compounding problem is retention. The same candidates being aggressively bid up are the ones easiest to poach in the next cycle. Companies paying premium salaries to lock down AI talent in Q1 are watching those same hires leave for higher premiums by Q4. The cost per successful hire keeps climbing, and the average tenure of an AI/ML engineer at a single company keeps dropping.

The equity math adds on another layer. Companies are handing out four-year vesting packages to engineers whose median tenure at a single AI role is now closer to eighteen months. The stock grants that were supposed to buy loyalty are getting cashed out at the second cliff and the engineer is moving to the next auction. What used to function as a retention tool is now a signing bonus with extra steps.

This raises a question the CFOs signing off on those req expansions probably haven't asked yet: what's the cost per shipped ML product when the average engineer working on it leaves before the product hits GA?

For a lot of companies posting those 163% more jobs, the answer's uncomfortable.

Where the best are hiring

The companies pulling ahead here have stopped competing for the same profile as everyone else. Their hiring models look different, and the results are showing up in what they ship.

Some are hiring adjacent talent and training. Data engineers, backend engineers, and analytics professionals with strong quantitative backgrounds can be brought into ML work in six to twelve months with the right investment. The process is slower than a lateral hire, and the resumé won't say "ML engineer" on day one, but the retention curve is flatter and the cost per productive engineer is a fraction of the auction price.

Others are hiring on capability first. What the person has shipped, whether or not the last three job titles included "ML," turns out to be a more reliable indicator of what they can build than a checklist of framework experience. Companies running that approach are willing to interview differently and trust their own technical assessments over resumé keyword filters.

The rest of the market is still trying to hire the same profile everyone else is trying to hire. The approach works if the company can outbid Anthropic, OpenAI, and Meta. For everyone else, it's the most expensive way to lose.

The 163% number is real, and so is the talent gap. What's optional is the assumption that the only way to close it is to write bigger checks.