Two numbers landed this year that look like they cannot both be right.
PwC went through more than a billion job advertisements in 27 countries and found that roles asking for AI skills advertise a 62% wage premium, up from 57% the year before. Payscale asked employers what they actually do with pay, and found that 55% offer no premium, no bonus and no equity for strong AI skills. Only 14% put it in base pay.
If you have spent evenings learning this stuff on the promise of a raise, that second number stings. But both are measured correctly, and the gap between them is the most practically useful thing in the whole AI careers conversation, because it tells you exactly where the money is handed over and where it is not.
Here is what each study actually measured, why the premium is real and still mostly uncollectable from your current employer, and the four moves that turn it into money in your account.
What PwC measured
The 2026 Global AI Jobs Barometer, published 15 June, is the biggest dataset anyone has on this.
| Finding | Figure |
|---|---|
| Wage premium advertised for AI skills | 62%, up from 57% in 2025 |
| Dataset | 1 billion+ job ads, 27 territories |
| Growth in postings asking for specific AI skills | 69%, against 8.6% for the market as a whole |
| "Professionalised" roles vs "democratised" ones | twice the job growth, 42% faster salary growth |
| Headcount growth, most AI-exposed companies vs least | 52% vs 36% |
| Wage growth, most AI-exposed companies vs least | 24% vs 17% |
Read the first row precisely, because the precision is the point: this is what employers advertise when they are hiring. It is the price of an AI-skilled stranger on the open market. It is not a measurement of what happened to the pay of people already on the payroll.
What Payscale measured
Payscale asked the other question: when the hiring is done and it is time to set pay, what do companies actually do? Its 2026 Compensation Best Practices Report, and a preview of its AI Workforce Impact Report published 24 September, give an uncomfortable answer.
| Finding | Figure |
| Companies that rewrote job listings to ask for AI skills | over 60% |
| Offer no premium, bonus or equity for those skills | 55% |
| Offer higher base pay | 14% |
| Offer a bonus | 10% |
| Offer other long-term incentives | 9% |
| Employers who once paid a premium and now treat AI fluency as baseline | 23% |
| Say their salary structures have not kept pace with AI | 49% |
| Cannot find candidates with the AI skills they need | 41% |
| Employees who believe they should be paid more for AI skills | 56% |
Payscale's chief compensation strategist, Ruth Thomas, put the mismatch in one line: "AI skill demand is outpacing employers' ability to price it."
Notice the shape of it. Employers are rewriting what they ask for much faster than they are rewriting what they pay. Six in ten changed the job description. Fewer than three in twenty changed the base salary.
Why both numbers are true
Put the two studies side by side and the apparent contradiction dissolves.
| PwC | Payscale | |
| What it looks at | job advertisements | employers' pay practices |
| Whose pay | the person being hired | the person already employed |
| What it captures | the market price of the skill | the internal salary structure |
| The number | 62% premium | 55% pay nothing extra |
The premium is priced at the moment of hiring, and almost nowhere else. A company competing for a scarce candidate will pay well above its own internal bands to win them, then put that person into a structure that gives everyone a 3.5% annual increase. Someone sitting at the next desk who quietly taught themselves the same skills is governed by the structure, not by the market.
This is the same gap I pointed at in the AI job split, but the Payscale data sharpens it from a warning into an instruction. The skills work. The channel most people use to cash them in does not.
A correction to that earlier piece while I am here, because the numbers have moved and I quoted them loosely: the 56% figure was PwC's 2025 reading, and the jump from 25% happened over several years, not one. The current reading is 62%, up from 57%.
The clock on the premium
There is a second number in the Payscale preview that deserves more attention than it got: 23% of employers who used to pay a premium for AI skills now treat AI fluency as baseline.
That is the whole lifecycle of a skill premium in a single statistic. A scarce skill commands extra pay, then enough people acquire it that it stops being scarce, then it becomes a line in the standard job description that everyone is expected to meet and nobody is paid extra for. Spreadsheets went through this. So did basic web literacy.
Nearly a quarter of the market has already made that transition for AI, while the advertised premium is still rising. Both can run at once, because different employers are at different stages. What it means for you is a deadline rather than an opportunity without end: the premium is at its most collectable now, and the window closes employer by employer.
The four moves that actually collect it
Change the employer, or get a real offer. This is the uncomfortable one, and it is the single biggest lever in the data. The 62% is paid at hiring. If your current employer is in the 55% that has no mechanism to pay for AI skills, no amount of demonstrating them changes the structure they are bound by. An outside offer is also the only instrument that reliably moves an internal salary, and 41% of employers cannot find the AI talent they need, so offers are gettable.
Translate your work into money before you ask. "I use AI a lot" is not a case. "I rebuilt the monthly reporting process, it now takes four hours instead of three days, and that is roughly 200 hours a year" is a case, and it survives contact with a finance director. Employers are paying for measurable output, not for tool familiarity, which is exactly what the frontier labs concluded when they spent billions hiring people who redesign workflows.
Benchmark before the conversation, not after. Because employers price AI skills so inconsistently, the spread between two offers for the same work is unusually wide right now. Half of employers admit their salary structures have not kept pace and fewer than half say their benchmarking reflects the skills they ask for. In a market this badly priced, the person who has looked up real numbers for their specific role and location has an advantage over the person quoting a global average.
Stand on the professionalised side of the line. PwC's most useful distinction is between work AI professionalises, where the tool raises what one skilled person can produce, and work it democratises, where the tool lets anyone do what used to need a specialist. The first group saw twice the job growth and 42% faster salary growth. Prompting ability alone pushes you toward the second group, because anyone can prompt. Prompting plus a domain nobody can pick up in a weekend keeps you in the first. The operator skill stack is the practical version of that combination.
If you are early in your career, the trap is different
The entry-level finding in the Barometer is the harshest number in the report, and it has nothing to do with premiums.
PwC looked at 2.4 million US entry-level roles. Those with high AI exposure are seven times more likely to demand skills traditionally expected of senior staff, such as judgement and client handling. Those "seniorised" entry roles have grown 35% since 2019. Every other entry-level role shrank 10%.
So the junior market has not simply contracted. It has been rewritten to ask for things that juniors have not historically had, while the roles that used to teach those things disappear. Advice to "just get AI skills" misses the actual bar, which is judgement.
The usable response is to arrive with evidence instead of potential: two or three finished projects where you took something from messy to working, with the before, the after and the numbers. That is a portfolio argument, not a certificate argument, and it is why I keep steering people away from certificate shopping in the honest guide to AI courses. The material itself costs nothing, as the free resources list shows; what is scarce is proof that you finished something. More on this market in the class of 2026 analysis.
What this means for you
If you are employed and have been learning AI on your own time, test which kind of employer you have before you invest another six months of evenings. Ask HR directly whether the salary structure has any mechanism to recognise AI skills. A straight answer either way is worth having: if the answer is no, you now know the premium in your case is only collectable by moving, and you can plan accordingly rather than feeling cheated at your next review.
If you are hiring or setting pay, the Payscale numbers are a warning rather than a comfort. Your competitors advertise 62% premiums while 56% of your own staff think they deserve more for the AI skills they have built. The people who close that gap are the ones who leave, and they are the ones you cannot replace, because 41% of employers already cannot find this talent. Rewriting the job description without touching the salary band is the cheapest way to lose exactly the people you just said you needed.
If you are deciding whether learning this is still worth it, yes, and the arithmetic has not changed much. A few hundred hours against a premium that is still rising at the hiring line is one of the better returns available to a normal worker, as long as you plan to collect it in the market rather than at a performance review. The nine business models in how people actually make money with AI are the other route, where you set your own price and nobody's salary band applies.
What to watch
Payscale's full AI Workforce Impact Report in November. The September release was a preview. The full version should show whether the share of employers treating AI fluency as baseline is still at 23% or climbing, which is the single best indicator of how much runway the premium has left.
Whether the advertised premium starts falling. PwC's next Barometer is a year away, but job-ad trackers move faster. The premium rising from 57% to 62% while a quarter of employers stop paying it is an unstable combination, and one of those two lines will give.
The entry-level reversal, if it comes. Some talent leaders expect AI to increase junior hiring rather than cut it. If "seniorised" entry roles start being filled by people who were trained in them rather than by mid-career hires, the bottom of the ladder repairs itself. If not, the 35% versus minus 10% split hardens into a permanent shape.
The honest take
What this pair of studies really exposes is that a skill premium is not a property of the skill. It is a property of the moment you negotiate. The same AI fluency is worth 62% to someone being hired and nothing at all to someone with a good annual review and a 3.5% increase, and the difference between those two people is often nothing but which conversation they are in.
That is unfair, and it is also completely actionable, which is the part worth holding on to. You cannot make your employer's salary structure recognise what you have learned. You can make sure you are in the conversation where it is priced properly, with evidence in hand, while 41% of employers still cannot find what you have.
So the question worth answering this week: if the market pays 62% more for what you can already do, and your employer's pay structure has no line for it, which of those two facts have you actually tested?
Sources: PwC 2026 Global AI Jobs Barometer; Payscale: the AI skills vortex, 24 September 2026; CBS News on Payscale's Compensation Best Practices Report.
