Four numbers that frame the whole market
- $173,482 — Glassdoor's February 2026 figure for the median US AI/ML engineer, with the 90th percentile at $269,611.
- £85,000 — the UK median Gen AI Engineer salary in ITJobsWatch's six-month window to 8 August 2026, up 21.4% year on year.
- ₹40–65 LPA — what senior AI engineers at Global Capability Centres in India now earn, against ₹20–35 LPA at IT services firms for the same experience.
- 3.2:1 — the global demand-to-supply ratio for qualified AI engineers, per FutureProofing's July 2026 analysis: roughly 1.6 million open positions chasing about 518,000 qualified candidates.
Salary surveys usually disagree with each other. This year they mostly do not. Between February and August 2026, five independent datasets — Glassdoor, Levels.fyi, Robert Half's 2026 Salary Guide, ITJobsWatch in the UK and the Indian GCC benchmarking reports — converged on the same shape: a market that pays a steep premium for a short list of skills, splits sharply by employer type, and is still nowhere near clearing. That last part matters most for anyone reading this from Bengaluru, Chennai, London or Manchester, because a market that cannot fill roles in 90 days is a market where leverage sits with the builder who is easy to find.
The US baseline — and the frontier-lab outlier
Start with the US, because it anchors global pricing. Glassdoor's February 2026 data puts the median AI/ML engineer at $173,482, with the 90th percentile at $269,611. Robert Half's 2026 Salary Guide lands nearby — it projects AI/ML engineer pay rising 4.4% to $170,750 this year, against just 1.6% growth for technology salaries overall. In other words, generalist tech pay has flattened; AI pay has not.
Then there is the tier that distorts every average: the frontier labs. Levels.fyi's May 2026 data puts median software-engineer total compensation at roughly $600,000 at Anthropic and $795,000 at OpenAI, with equity making up the majority of the package above mid-level. A single reported OpenAI L5 package reached $1.15 million total — $336,000 base plus $774,000 in annual stock — though that is one data point on Levels.fyi, not a median, and should be read as a ceiling rather than a benchmark. Fewer than a few thousand people worldwide sit in this tier. It is worth knowing it exists mainly because it drags the aspirational headlines upward while the working market prices very differently.
The bands by level: US, UK, India
Put the three markets side by side and the structural gap is stark — but so is the pattern within each market. These are cash-compensation bands for AI/ML engineering roles, compiled from Glassdoor (February 2026), kaam.work's June 2026 three-country comparison and ITJobsWatch:
| Level | US (total comp) | UK (base) | India (cash CTC) |
|---|---|---|---|
| Junior (0–2 yrs) | $120,000–150,000 | £30,000–38,000 | ₹6–12 LPA |
| Mid (3–6 yrs) | ~$173,000 median; $260,000 at top product firms | £45,000–60,000 | ₹12–30 LPA |
| Senior (7+ yrs) | $280,000–350,000+ | £70,000–110,000 | ₹30–60 LPA+ |
| Frontier-lab tier | $600,000–795,000 median TC | — | — |
The UK numbers deserve a closer look, because they moved fast. A year ago the ITJobsWatch median for a Gen AI Engineer sat at £70,000; the rolling window to 8 August 2026 shows £85,000, a 21.4% jump, with the 25th percentile at £61,250 and the 90th at £132,000. Central London medians run around £100,000. One honest caveat: the UK-wide sample is thin — 31 advertised permanent vacancies in the window — so these bands are directional, not gospel. The direction, though, is unambiguous, and it is up. If you are a UK builder weighing a move from a generalist engineering title into a GenAI-titled role, the title change alone is currently worth £15,000–25,000 at the median.
India: your employer type matters more than your experience
India's 2026 pay data carries one lesson worth more than any skills course: the biggest single salary lever is the kind of company you join. Benchmarks compiled by Taggd's 2026 India salary report show the same engineer, with the same years of experience, priced wildly differently across four employer types:
| Employer type | Fresher | Mid (3–6 yrs) | Senior (7+ yrs) |
|---|---|---|---|
| IT services | ₹5.8–8 LPA | ₹12–20 LPA | ₹20–35 LPA |
| Product companies | ₹9–14 LPA | ₹20–35 LPA | ₹35–60 LPA |
| GCCs | ₹10–18 LPA | ₹22–40 LPA | ₹40–65 LPA |
| Global tech (Google, Microsoft, Amazon) | ₹15–25 LPA | ₹30–55 LPA | ₹55–80 LPA+ |
Multiple Indian datasets agree that GCCs pay 40–70% more than IT services firms for equivalent experience, and that switching from services to a product company or GCC delivers an immediate 60–100% rise. That makes the Global Capability Centre channel the most underpriced career move in Indian AI right now — we covered how to actually get into one in our guide to AI roles at Global Capability Centres.
Indian CTC figures and US total-compensation figures are not the same unit. US "total comp" includes equity that may or may not vest at its headline value; Indian CTC often includes gratuity, employer PF and one-off components. When you benchmark yourself, compare fixed cash to fixed cash first — otherwise every cross-border comparison flatters the US number by 20–40%.
Which skills actually price the premium
Robert Half's 2026 Salary Guide found that 87% of technology leaders pay a premium for specialised AI and ML skills. But "AI skills" is doing a lot of work in that sentence. The premium concentrates on a short list, and the 2026 data lets us rank it:
| Specialisation | Premium over generalist AI role | Why it prices high |
|---|---|---|
| LLM fine-tuning / post-training | 25–40% | Scarce, hard to verify from a CV, directly tied to model quality |
| MLOps / inference infrastructure | 20–35% | Every deployed model needs it; cost pressure makes it a P&L skill |
| Agent engineering & evals | Fastest-growing demand; premium still forming | Reliability work on agents is where 2026 production budgets went |
| NLP / computer vision (classical) | 15–25% | Mature; premium eroding as tooling commoditises |
The fine-tuning figure is the steadiest across sources: Indian benchmarks put GenAI/LLM specialisation at 25–40% over a generalist profile at the same experience level, and US market surveys land in the same range — some go as high as 40–60% for post-training specialists, but 25–40% is the range multiple datasets support. The agents-and-evals row is where the demand curve is steepest even though salary surveys have not yet crystallised a premium: agent-related postings and interview activity grew triple-digit percentages into 2026, and reliability roles around agents barely existed eighteen months ago. If you want to position for that wave, start with the emerging agent reliability engineer role — the job titles are new enough that early, provable experience beats years-of-experience arithmetic.
Premiums attach to evidence, not claims. "Fine-tuned LLMs" on a CV is noise; a public write-up with your eval deltas, training cost and a before/after comparison is signal a hiring manager can price. One documented project moves you further up these bands than a certificate ever will.
The arbitrage: remote India, US-linked pay
The most interesting 2026 number for Indian builders is not a domestic band at all. Senior engineers working remotely for US companies from India report ₹60–80 LPA — above every domestic band short of global-tech senior roles — while the employer still saves dramatically: kaam.work's June 2026 analysis puts the cost of a mid-level ML engineer hired in India through an employer-of-record at $35,000–50,000 a year, against a $220,000–300,000 all-in cost for equivalent US talent. Both sides of that trade win, which is why it is growing.
The same logic runs at smaller scale in the UK, where contract and fractional arrangements let builders price closer to project value than to a salary band — we broke down the trade-offs in contract vs permanent vs fractional AI engineering. The common thread: every one of these higher-leverage arrangements starts with a hirer finding you, not with you joining an applicant queue of four hundred.
Why the gap will not close soon — and what to do about it
FutureProofing's July 2026 analysis puts global demand at 3.2 qualified candidates' worth of open roles for every one candidate available, and senior AI positions take 90–120 days to fill against roughly 25 days for a generic software role. Robert Half's data says employers have already accepted they must pay up. Fresh capital keeps arriving on the demand side — the AI teams funded in August are hiring right now — and university pipelines cannot mint post-training specialists at the rate inference budgets are minting vacancies.
For builders, the practical reading of all this data comes down to three moves:
- Pick the right employer type. In India, services-to-GCC or services-to-product is worth more than any single skill certification. In the UK, a GenAI-titled role at a product firm now clears the generalist median by £15,000+.
- Specialise where the premium is durable. Fine-tuning, inference infrastructure and agent evals price 20–40% above generalist work and show no sign of commoditising in 2026. The path from senior to staff runs through exactly this kind of legible specialisation — see our senior-to-staff promotion guide.
- Be findable with proof. A 90–120-day fill time means hiring managers are searching, not sifting. The builders capturing the remote-for-US premium and the GCC premium are, almost without exception, the ones whose work is public, specific and verifiable.
The market pays a premium for builders it can verify. Are you one it can find?
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, and early profiles carry the Founding Builder badge while spots remain.
Become a Verified Builder →The honest caveats
Three qualifications before you renegotiate anything. First, survey windows differ: Glassdoor's figures are February 2026, Levels.fyi's May 2026, ITJobsWatch's a rolling window to August — a fast market makes even six-month-old numbers stale. Second, sample sizes at the specialised end are small; the UK Gen AI median rests on 31 advertised vacancies, and frontier-lab medians describe a tiny population. Third, the 3.2:1 demand ratio is one firm's analysis — it is echoed across several talent reports, but nobody audits it. None of that changes the direction of the data. It just means you should benchmark against bands, not single numbers, and against your market, not San Francisco's.
Sources: Glassdoor, Levels.fyi, Robert Half 2026 Salary Guide, ITJobsWatch, FutureProofing, Taggd India Salary Report, kaam.work.