What the two datasets show
- The UK denominator is shrinking. Indeed Hiring Lab's 2026 Mid-Year UK Jobs & Hiring Trends Report, published on 3 August 2026, puts overall UK job postings down 11% since the start of 2026 and 32% below the pre-pandemic baseline of 1 February 2020.
- The AI numerator is setting records. In the same series, 9.4% of UK job postings mentioned AI or related tools as of the end of June 2026 — a new high.
- The mentions are concentrated, not spread. Indeed Hiring Lab has Data & Analytics highest at 48.8% and Software Development at around half of postings, against below 1% for Beauty & Wellness, Personal Care, Cleaning and Driving.
- India's chart points the other way. NASSCOM recorded roughly 2.9 lakh AI job postings in India in 2025 and projects 3.82 lakh in 2026, a 32% year-on-year increase.
- The friction is quality, not headcount. On NASSCOM's employer data, 58% cite low applicant volume and 50% cite a skills mismatch, with the sharpest gaps in applied, production-ready work such as RAG and MLOps.
Two things are true about the same labour market and they point in opposite directions. Each half gets quoted in isolation by people arguing for opposite conclusions: one camp reads the 11% fall and declares the market finished, the other reads the 9.4% record and declares AI skills a golden ticket. Both are reading one number and ignoring the one underneath it.
One caution first. Nothing in the Indeed Hiring Lab data establishes that AI is causing the fall in overall postings. The two series move within the same window; that is correlation, and the report does not claim a causal link. Rate policy, employer costs and ordinary cyclical caution sit in the same period. Anyone telling you the 11% decline is the machines arriving is drawing an arrow the data does not support.
A record share of a shrinking market
Here is the arithmetic most coverage of the 9.4% figure skips. A share is a ratio, and a ratio can rise for two different reasons: the numerator grows, or the denominator shrinks. In the UK the denominator is demonstrably shrinking — down 11% this year and a third below February 2020. A record AI share is therefore compatible with AI-mentioning postings growing strongly, and equally compatible with them growing modestly while everything around them falls away. The published summary does not separate the two, and honest analysis has to say so.
So the framing matters. Scarcity in the UK AI market is currently relative, not absolute. Being 9.4% of a market that has contracted by roughly a third since 2020 is a materially different proposition from being 9.4% of a growing one: in the first case you are the healthiest part of a smaller room. The word "record" does identical work in a headline either way, which is why it is worth distrusting on its own.
What the record does establish is a behavioural change: nearly one in ten British job adverts now names AI or an AI tool somewhere in the requirements. That is a signal about the composition of hiring rather than its volume.
Where the AI mentions actually sit
The average conceals almost everything. Indeed Hiring Lab's sector breakdown puts the extremes a long way apart: the 9.4% is not a thin uniform layer but a heavy concentration in a few sectors and near-absence elsewhere.
| Sector | Share of postings mentioning AI | Read |
|---|---|---|
| Data & Analytics | 48.8% | Highest of any sector — roughly one posting in two |
| Software Development | Around half of postings | Effectively an expectation, not a differentiator |
| Beauty & Wellness, Personal Care, Cleaning, Driving | Below 1% each | Essentially untouched in the language of adverts |
Read the top two rows again. When around half of software development postings mention AI, naming AI on your own profile has stopped being a signal: in the sectors where the demand sits, the mention is table stakes, and differentiation has moved downstream to what you have specifically built and run.
The premium figure that has not settled
Which brings us to the number nobody should quote confidently. The UK AI wage premium is reported inconsistently across published coverage: PwC's AI Jobs Barometer 2026 is cited at 34.2%, up from 11% the prior year, in some coverage, while other coverage of PwC data cites a 56% premium for "advanced" AI skills, up from 25% — and because the methodologies and the underlying definitions of what counts as an AI skill differ between those readings, the two figures are not directly comparable and neither can be treated as the settled number.
That is not a hedge for its own sake. The gap between 34.2% and 56% is the difference between a strong premium and a transformative one, and anchoring a salary conversation on either means anchoring on a definition you have not read. Our guide to benchmarking and negotiating AI engineer pay across India and the UK takes the sensible route: benchmark against roles with your actual scope in your actual market, and use published premiums as context rather than as a claim.
Two traps here. First, do not over-read a single half-year: a 9.4% share measured at the end of June 2026 is one observation in a volatile series, and postings data reflects how employers write adverts as much as what they intend to hire — a wording fashion can move a share without a single extra role existing. Second, treat any single AI wage-premium number as contested: the 34.2% and 56% figures both trace to coverage of PwC's AI Jobs Barometer, are built on different definitions, and quoting either as settled fact anchors a negotiation on a number that will not survive scrutiny.
India is running the opposite chart
Move the same question to India and the shape inverts. NASSCOM recorded roughly 2.9 lakh AI job postings in 2025 and projects 3.82 lakh in 2026 — a 32% year-on-year rise. This is not a rising share of a falling total. It is a rising total. NASSCOM–Deloitte project India's AI talent pool reaching 1.25 million by 2027, and one 2026 analysis by Taggd puts about 11.7% of all Indian job postings as explicitly requiring AI skills, up from 8.2% a year earlier — a figure worth carrying carefully, since it rests on a single source.
The interesting part is that growth has not resolved the shortage. Around 16% of Indian IT professionals are AI-skilled, according to the Ministry of Electronics and Information Technology, and NASSCOM's employer-side data shows 58% of employers citing low applicant volume and 50% citing a skills mismatch. A talent pool heading for 1.25 million and employers who cannot fill roles are not a contradiction: supply is growing in headcount faster than it is growing in demonstrable capability.
| United Kingdom | India | |
|---|---|---|
| Overall postings trend | Down 11% since the start of 2026; 32% below the February 2020 baseline (Indeed Hiring Lab) | Growing: AI postings from roughly 2.9 lakh in 2025 to a projected 3.82 lakh in 2026 (NASSCOM) |
| AI share of postings | 9.4% of all postings mention AI as of end-June 2026, a record (Indeed Hiring Lab) | About 11.7% explicitly require AI skills, up from 8.2% — one 2026 analysis by Taggd |
| Direction of AI demand | Rising share of a falling total (Indeed Hiring Lab) | Rising share of a rising total, projected up 32% year on year (NASSCOM) |
| Binding constraint | Concentration — AI mentions cluster in Data & Analytics and software and sit below 1% across several trades (Indeed Hiring Lab) | Supply quality — 58% of employers cite low applicant volume, 50% a skills mismatch (NASSCOM) |
| Best-sourced figure | 9.4% AI mention share at end-June 2026 (Indeed Hiring Lab) | 3.82 lakh projected AI postings in 2026 (NASSCOM) |
Where the two genuinely differ
It would be lazy to flatten these into one narrative. The UK story is a two-speed market inside a contraction: total demand is well down, AI-adjacent demand is holding up conspicuously well within it, and sector choice therefore matters more than it did. The Indian story is growth with a supply-quality gap: demand is expanding in absolute terms, the pool is expanding too, and the constraint sits in what that pool can evidence. We argued in The AI Talent Gap Is a Supply Problem Now, Not a Demand One that supply quality is the through-line across both, and the mid-year data has not changed our view.
One structural feature connects them. Capability centres run by international firms hire against global standards — global capability centres account for a large and growing share of India's AI hiring — so a UK-headquartered employer's requirements land in Bengaluru or Hyderabad regardless of what the UK posting count does. Postings and hiring are related but not identical, and a contraction in one country's adverts is not automatically a contraction in its employers' hiring.
Every article here is written by a Verified Builder. Want your name on the next one?
AI Tech Connect lists AI engineers, founders and researchers across India and the UK — and the people hiring browse it to find them. Adding your profile is free.
Become a Verified Builder →The constraint is evidence, not headcount
Put NASSCOM's two employer-side numbers next to each other and something specific falls out: 58% report low applicant volume; 50% report a skills mismatch. Those are close to the same magnitude, so the problem is not simply that too few people apply — if it were, a pool heading towards 1.25 million would be fixing it. The second number describes a different failure: people apply, and the employer cannot establish that they can do the work.
NASSCOM puts the sharpest part of that mismatch in applied, production-ready skills, naming RAG and MLOps. That specificity is useful. It is not that employers cannot find people who understand transformers or have finished a course; it is that they cannot find demonstrable experience of retrieval systems that behave under real query distributions, or of the machinery that keeps a model running once the demo is over. Those are the least examinable skills on a CV and the most examinable in an artefact.
Which reframes the scarcity discussion. If the binding constraint were headcount, the answer would be more graduates and conversion courses, resolving over a few years. If it is evidence, the answer is individual: what moves you across the gap is not another credential but a legible record of production work. In a market where nearly half of software postings already mention AI, the credential has been commoditised and the evidence has not.
Write your evidence in the vocabulary the mismatch is measured in. Not "experienced with RAG" but "rebuilt a retrieval layer over 400k support documents, moved from naive chunking to hybrid search, cut unanswered queries measurably, and here is the evaluation set". Not "familiar with MLOps" but "own the deployment pipeline for three production models, including rollback and drift monitoring". Two entries of that quality beat ten lines of tooling keywords — our proof-of-work portfolio guide covers the format.
In a contracting market, being findable beats being applied
The last piece follows from the first two. When posting volumes fall, employer behaviour changes shape. Advertising in a thin market generates a large, poorly targeted applicant pile that costs more to filter than it saves, and organisations cautious about headcount prefer a targeted approach over an open call. So they search more and post less — the pattern we would expect in a soft hiring market, and the UK is in one by any reading of the Indeed Hiring Lab series.
The consequence is asymmetric. Applications are linear: each one costs time and returns at most one outcome. A public, structured, specific record of what you have built works the other way — it gets found by searches you never knew were run, and improves as you add to it. In a market with 11% fewer adverts, the ratio between those two activities should shift. Not to zero applications, but towards making sure a recruiter searching for someone who has shipped a retrieval system under load can actually find you.
That is the honest case for a Builder profile, and its limits matter. A profile is not a job. It does not guarantee an outcome, will not compensate for thin experience, and no directory can promise you a hiring manager's attention. What it removes is one specific failure mode: being qualified, in a market where employers report they cannot find qualified people, and being invisible to the search that would have found you.