The AI Talent Gap Is a Supply Problem Now, Not a Demand One

The story we keep telling is a year out of date

Almost every piece written about AI hiring since 2024 — including several of our own — has been a demand story. Roles are multiplying. Budgets are climbing. Salaries are re-pricing every quarter. All of that remains true, and all of it is now the less interesting half of the picture.

A demand story implies a particular kind of market: lots of buyers competing for a well-understood good, with price doing the rationing. That is not what the 2026 data describes. What it describes is a market where the good itself is in short supply and, more importantly, where buyers cannot reliably identify the good when they see it. Those are different problems with different consequences, and the second one is the one that now determines careers.

The clearest single figure is Indian. Demand for AI engineers in India is rising about 40% year on year, while the skilled talent pool is growing at only 15 to 20%. Hold those two numbers next to each other and the shape of the market becomes obvious: even at the optimistic end of the supply range, demand growth is running at twice the rate of supply growth. At the pessimistic end it is closer to 2.7 times. The gap is not closing; it is compounding, year on year, and every quarter it compounds the imbalance gets structurally harder to unwind, because you cannot manufacture five years of production experience in twelve months.

The UK tells a version of the same story from the other direction. AI engineering job adverts in the UK rose 1,133% between 2024 and 2026, per IT Jobs Watch data, reported by VIQU. That is not a demand signal that a training pipeline can absorb. Nothing about the British AI workforce grew by anything approaching elevenfold in two years.

We should be upfront about that number before we use it. It is single-origin: every subsequent appearance of the 1,133% figure traces back to the same IT Jobs Watch dataset, so it has not been independently reproduced. Directionally, though, it does not stand alone. PwC's 2026 AI Jobs Barometer points the same way, reporting that UK AI hiring and wages are outpacing the broader jobs market. We are deliberately not attaching a percentage to PwC, because the value of that source here is corroboration of direction rather than magnitude. One dramatic single-origin number plus one independent directional confirmation is a weaker evidence base than one might like — and, as we set out below, the argument is built so that it survives losing the dramatic number entirely.

What actually changes when supply is the binding constraint

Economists distinguish between markets that clear on price and markets that clear on search. AI hiring in 2026 is decisively the second kind, and the practical consequences for individual engineers are larger than the salary headlines suggest.

Consider a hiring manager in Bengaluru or Bristol with three open requisitions and no spare time. In a demand-constrained market — the market most of us learned to job-hunt in — that manager's problem is triage. Two hundred CVs arrive, most are unsuitable, and the entire hiring apparatus exists to filter them down. Every piece of conventional career advice you have absorbed is filtering advice: tailor the CV, hit the keywords, make it easy to say yes.

In a supply-constrained market, that manager's problem inverts. The applications that arrive are not enough, and the ones that do arrive skew towards people who are available rather than people who are good — which is not the same population. So the manager stops filtering and starts sourcing. They search. They ask their network. They open directories, portfolios and repositories and look for evidence of someone who has already built the thing they need built. Inbound applications become a minor input; outbound search becomes the primary one.

That single behavioural shift is the whole argument of this piece. If the buyer is doing the searching, then a candidate who cannot be found does not exist. Not "is at a disadvantage" — does not exist, in exactly the way that a book not in the catalogue does not exist to a librarian. This is why we think the practical advice for AI engineers in 2026 has to change: the discipline of making your projects discoverable rather than merely impressive is now doing more work than any amount of CV polish.

Two markets, two different constraints

It would be lazy to treat India and the UK as one market with different currencies. They are constrained by genuinely different things, and the difference matters for anyone deciding where to point their career.

The UK constraint is fundamentally one of headcount and mobility. An advert increase of 1,133% since 2024 lands against a smaller absolute talent base than India's, and the usual pressure valve — importing experienced engineers — carries visa and relocation friction that slows the response to a trickle. When the pool cannot be topped up quickly, the market clears through internal poaching and pay escalation, which is precisely what the senior band of roughly £90,000 to £150,000 base reflects.

India's constraint is not headcount. India has volume, and the volume is growing. The constraint is depth. Around 11.7% of all job postings in India now explicitly require AI skills, up from 8.2% a year earlier — and it is worth being precise about what that figure is. It is a demand signal, not a supply signal. It tells you how many employers are asking for AI skills; it says nothing about how many engineers can deliver them in production. Set it beside the 15 to 20% growth in the skilled pool and the gap becomes visible: the requirement is spreading through the job market considerably faster than the capability is spreading through the workforce.

We are deliberately not converting between these bands. A rupee figure and a sterling figure describe different cost bases, different tax regimes and different employer expectations, and pretending an exchange rate makes them comparable produces nonsense. What is comparable is the structure: both markets show demand signals accelerating faster than supply signals, and both show unusually wide dispersion within the same nominal job title. If you are benchmarking your own position, our guide to benchmarking and negotiating AI engineer pay in India and the UK works through each market on its own terms rather than through a conversion.

One further Indian data point deserves attention, because it explains where the demand pressure is coming from. Indian AI startups raised $676 million across the first six months of 2026, up more than fourfold from $162 million across 30 deals in H1 2025, according to Inc42. Nearly 40 AI startups reached unicorn status in the first half of 2026, at valuations between $1 billion and $41 billion. Capital at that velocity does not sit still; it converts into headcount within a quarter or two. The 40% demand growth is not a forecast. It is already funded.

Wide pay bands are not noise. They are failed price discovery.

Look again at those salary ranges and notice how strange they are. In India, the same job title spans roughly INR 6 LPA to INR 80 LPA and above — better than a thirteenfold spread. In the UK, senior AI engineers sit at roughly £90,000 to £150,000, while prompt engineers occupy a band of roughly £42,000 to £77,000 that averages about £57,000. Even within a single seniority label, the top of the range is often close to double the bottom.

The conventional reading is that this is seniority mix: bands look wide because "AI engineer" covers everyone from a fresh graduate to a staff-level specialist. That is partly right, and we return to it in the counter-argument below. But it does not explain the dispersion within tiers, and it does not explain why the spread in AI roles is so much wider than we would expect in a comparably senior engineering role.

Our reading is that the width is the market failing to price talent it cannot evaluate. Pricing requires assessment, and assessment in this field is genuinely hard. A hiring manager cannot tell from a CV whether "built RAG pipelines" means a weekend tutorial or a system serving production traffic with an eval harness behind it. Faced with that uncertainty, employers do one of two things: they anchor low and hedge, or they anchor high and buy a signal — a brand-name employer, a well-known repository, a public track record. Both behaviours widen the band.

Which leads to the point that matters. In a market where the buyer cannot assess capability cheaply, proof of work is the pricing mechanism. It is not a nice-to-have appended to a CV; it is the instrument that moves you from the hedged bottom of the band to the confident top of it. The engineer with a public, verifiable, quantified project history is not merely more visible than an equivalent engineer without one. They are cheaper to assess, and in a supply-constrained market cheapness of assessment converts directly into offers and into pay. Our walkthrough of building an AI engineer portfolio around proof of work rather than a résumé covers what that evidence needs to contain to actually do the job.

The honest case against this argument

An opinion piece that only presents its own evidence is advocacy, not analysis. Here is the strongest version of the case against the supply-constraint thesis, and it is a good deal stronger than we would like.

The second objection is title inflation, and it is the one we find hardest to dismiss. A substantial share of the growth in "AI engineering" adverts is almost certainly reclassification: roles that were advertised as data engineering, backend or analytics in 2024 now carry an AI label because that is what attracts applications and justifies budget. If a third of the 1,133% is relabelling, the underlying demand growth is dramatically less alarming than the headline. The same critique applies to the Indian figure of 11.7% of postings requiring AI skills. "Requires AI skills" is doing enormous unexamined work in that sentence, and it plausibly includes roles where the requirement means competence with an off-the-shelf assistant rather than the ability to build a system.

Third, the salary bands really do reflect seniority mix. An INR 6 LPA to INR 80 LPA range is not principally evidence of pricing failure; it is principally evidence that the same two words describe a graduate and a specialist with eight years of production experience. Attributing that spread to evaluation difficulty overstates our case, and we should say so plainly.

Where does that leave the thesis? Weakened at the edges, intact at the core — and deliberately so. Delete the 1,133% figure entirely and the argument still stands: the Indian comparison does not rest on advert counts at all, since it sets demand growth against pool growth, and a roughly two-to-one ratio survives a great deal of scepticism about its inputs. The UK half of the case then rests on PwC's directional finding that AI hiring and wages are outpacing the wider jobs market, which is less dramatic but not single-origin. If you distrust the headline number — and you are entitled to — you should still finish this piece agreeing about the direction of travel. That was the test we set the argument.

The salary-dispersion section is the softest plank, and we present it as interpretation rather than as fact. But the behavioural consequence — that hiring teams facing scarcity shift from filtering to searching — does not require any of these numbers to be precise. It only requires them to point the same way, and they do, in both markets, from separate sources with separate methodologies.

What follows for builders in India and the UK

If you accept the argument, three things change about how you manage your career this year.

• Optimise for retrieval, not for persuasion. A CV is a persuasion document sent to someone who already knows you exist. In a search-driven market, the prior step is being retrievable at all — indexed, tagged with the vocabulary hiring managers actually type, and present in the places they look. Our LinkedIn playbook for ranking in recruiter search deals with the mechanics of exactly that.
• Publish evidence that survives scepticism. Named systems, measured outcomes, honest scope. "Built a retrieval pipeline serving 40,000 queries a day at a 93% eval pass rate" is assessable in seconds. "Experienced with LLMs" transfers the entire assessment cost back to the reader, and a reader with three open requisitions will not pay it.
• Treat geography as a variable, not a constraint. The UK's mobility friction and India's depth shortage are, from the right angle, the same opportunity viewed from two sides. Remote and distributed hiring is how the imbalance gets arbitraged, and our guide on how Indian and UK AI engineers land remote global roles covers how to position for it without pretending it is frictionless.

There is a longer-horizon point too. Salary growth in the Indian market has averaged 15 to 20% annually, and it is expected to continue through 2030. If that holds, the compounding advantage of being in the top quartile of a widening band is very large indeed — and the difference between quartiles, in a market that struggles to assess capability, is substantially a difference in legibility rather than in ability.

The bottom line

The demand story was true and is now insufficient. Demand growth of roughly 40% a year in India against 15 to 20% supply growth, and an advert increase of 1,133% in the UK against a base that cannot expand at anything like that rate, describe a market where supply has become the binding constraint. Markets constrained by supply clear through search rather than through filtering. When the buyer is doing the searching, discoverability is not a marketing nicety layered on top of competence — it is the mechanism by which competence gets converted into an offer. Build the proof, publish it where the search happens, and let the market find you.

Figures in this article are attributed in the sentences that use them and are dated to July 2026. UK advert growth is from IT Jobs Watch as reported by VIQU, and is single-origin; the directional corroboration is PwC's 2026 AI Jobs Barometer, from which no specific figure is cited; Indian funding figures are from Inc42; Indian demand, supply and salary figures are drawn from published market analyses of the Indian AI hiring market. Compensation data in this field re-prices quickly — treat all bands as a snapshot.