Build1 publisher2 min readPublished
Lightcast prices the AI wage premium at less than half of PwC's 62%
AI skills appear in 2.5% of US job postings, roughly a million of them, and the two most-cited estimates of what they add to pay within the same occupation differ by more than two to one, with only one of them expressed in dollars.
The Engineer · Build desk
What happened
- Lightcast data in the Stanford AI Index 2026 puts AI skill requirements in 2.5 percent of US job postings for 2025, up 55 percent in a year and roughly a million postings.
- PwC's AI Jobs Barometer 2026 puts the wage premium for AI skills within the same occupation at 62 percent, up from 56 percent the year before.
- Lightcast's own method produces a much lower premium for the same phenomenon: 28 percent, which it puts at about 18,000 dollars a year.
- Bain projects more than 1.3 million US AI jobs by 2027 against about 645,000 people available to fill them, leaving half of the roles open.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- contradiction PwC and Lightcast disagree by more than two to one on the same quantity, so whichever premium a company budgets is a choice of method, and the two choices support different band structures.
- cost A premium measured within occupations lands on roles already in the budget, and it lands on people who never changed jobs, so the bill arrives at annual review rather than at offer stage.
- decision With half of AI postings outside IT, the reopened bands belong to managers who do not think of themselves as hiring for AI, and someone has to decide which of those job families gets the money.
- exposure If Bain's 2027 gap holds and half of US AI requisitions stay open, employers planning to hire their way out are exposed to counteroffers on staff they already employ.
Lightcast's 18,000 dollars a year at 28 percent implies an average affected base of about 64,000 dollars [4][1]. Put PwC's 62 percent on that same base and the premium is close to 40,000 dollars, roughly 22,000 dollars a head more [3][2]. Across fifty roles the two published figures are about 1.1 million dollars apart while describing the same quantity [10].
The posting counts, at least, hold together. A million postings at 2.5 percent implies a US universe near 40 million [1][4]. The agentic AI skill went from 0.06 to 0.23 percent of postings in a year, and 0.23 percent of 40 million is about 92,000, against the roughly 90,000 the compilation reports [10][5].
The title data covers a small share of those postings. AI Engineer is first on LinkedIn's Jobs on the Rise in the US two years running, at about 1,550 new US postings a week and a median of 176,000 dollars [7]. Annualized, that is roughly 80,600 postings, about 8 percent of the million that require AI skills [6]. The roundup's author, who publishes as ilinmaks, wrote that "half of all AI postings sit outside IT departments" [9] and that "Prompt engineer is fading as a job title" [8]. A within-occupation premium is, by construction, paid inside titles a company already has [3].
Data Scientist shows how far a headcount forecast sits from a price. The WEF has it as the fastest-growing job of all, up 113 percent by 2030, while postings are up 15 percent over three years and pay is down 3 percent; the compilation says the role is being redefined toward ML and LLM work [11].
The frontier tier is priced on a different scale. Levels.fyi data puts median research scientist pay at Anthropic near 746,000 dollars and OpenAI L5 around 1.15 million [12]. Meta's summer 2025 poaching packages ran 200 to 250 million dollars over four years [13], which annualizes to at least 50 million, about 67 times the Anthropic median [9]. Fortune and Metix estimates put the number of people who can build frontier models at 500 to 1,000 worldwide, with acceptance rates below 1 percent at OpenAI, Anthropic and DeepMind [14].
For the 62 percent to transfer to a specific payroll, two things have to be true. The occupation mix has to resemble PwC's sample, and the premium has to show up in salaries paid, not only in what gets posted. The roundup does not set out the method behind either premium figure [20]. PwC's other 2026 number counts postings for AI specialists and reports 68.9 percent growth in 2025 against 8.6 percent for the market as a whole [2].
What to watch
- A revision to Lightcast's dollar premium in its next update would move any band built on it.
- Whether PwC's next barometer extends the 56-to-62 climb or the within-occupation premium flattens.
- Whether Bain's 2027 shortfall shows up as unfilled requisitions or as existing staff re-titled into AI roles.