Levels.fyi tells LDS the AI pay premium has stopped growing
Levels.fyi prepared a compensation analysis for Lets Data Science from its US salary submissions, and the findings cut against the prevailing story. The median AI premium is real but has stopped growing: an L3 ML/AI engineer earned about 1.34 times a non-AI peer in late 2024 and about 1.29 times by mid 2026. ML engineers have held a stable gap of roughly 40 percent over data scientists for two years, with no convergence. What is growing is concentration: the 90th-percentile-to-median ratio for ML/AI engineers widened from 2.06 to 2.23 while the broad software population barely moved, and the extreme numbers are employer effects. The median recruiter at OpenAI earns about 580,000 dollars on a sample of seven, which Levels.fyi flags as directional, against 155,000 dollars in the broad market.
Ask whether AI skills still command a pay premium and the honest answer, according to data Levels.fyi prepared for Lets Data Science, depends entirely on where in the distribution you look. At the median, the premium is real and has stopped growing. At the top, it has rarely been more extreme, and the thing doing the work is not the skill. It is the logo on the badge.
The analysis, prepared by Hakeem Shibly of Levels.fyi from the platform's US salary submissions in answer to our questions, controls for level and role rather than comparing averages, which is what makes its three findings worth taking seriously.
The median premium is real, and frozen
Holding role and level constant for US software engineers, the ML/AI specialization leads every level: at L3 the trailing-year median is $367,398 against $243,768 for data-focused engineers, with security, front-end and DevOps in between. An ML/AI engineer earns roughly a third more than a non-AI peer at the same level.
But tracked over time, the gap has stopped widening. In late 2024 an L3 ML/AI engineer's median total compensation of $375K was about 1.34x a non-AI peer's $279K; by mid 2026 it was about 1.29x, at $383K against $297K. Levels.fyi's own reading of the monthly series: "plateaued / mild compression. ML/AI flat, non-AI catching up."
The same stillness shows up between titles. Over four full half-year periods, ML engineers have sat roughly 40 percent above data scientists at the same level, about $370K to $400K against $255K to $290K at L3, and the ratio has not drifted. Because the narrow ML Engineer title and the broad ML/AI specialization tag agree so closely, Shibly calls it "a durable, skill-based gap rather than a labeling quirk."
One methodological choice deserves notice. The much-hyped "AI Engineer" title was excluded from the comparison entirely: with under about 30 US L3 data points per half-year, heavily junior, Levels.fyi concluded that "putting a number on it would misrepresent what we actually know." A data company declining to print a weak number is rarer than it should be.
What is actually growing: concentration
The second finding is the one that reframes the first. Over the last two years, the ratio of the 90th percentile to the median for ML/AI engineers widened from about 2.06x to 2.23x, while the same ratio for the broader software population barely moved, from 1.99x to 2.06x. In Shibly's words, "the gains are pooling into an elite rather than lifting everyone."
Where the elite sits is specific: a short list of frontier labs. Median engineer total compensation runs about $635K at OpenAI and $665K at xAI, roughly 2.5x the broader US software market, and across the L4 to L5 range individual packages run from about $1.5M into $3M territory. Levels.fyi is careful about what those numbers are made of: a large share is equity marked at private valuations, "paper, not liquidated cash," drawn from submissions rather than a full view of any employer's payroll.
Meanwhile, at big established firms the pure skill premium is modest. At Google, Meta and Nvidia, the ML/AI tag adds only low to mid single digits over an already high baseline, though frontier units inside those companies, Google DeepMind or Meta's superintelligence group, pay in frontier-lab bands because they compete for the same people.
The recruiter test
The cleanest evidence that the elite is employer-defined rather than skill-defined comes from a role with no AI specialization at all. The median US tech recruiter earns about $155K on 1,367 data points. At Google, the median recruiter earns about $217K. At OpenAI, the median recruiter earns about $580K, roughly 3.7x the market, "a bigger premium than the AI-skill premium itself," as the analysis puts it.
The sample is seven people, and Levels.fyi flags it plainly: treat exact multiples as directional, and Anthropic's recruiter data was too sparse to report at all. But the base-salary cut confirms the premium is not only paper equity: OpenAI recruiter base salaries run roughly $215K to $240K against a $140K market median. Put in ratio terms, a recruiter in the broad market earns about 60 percent of what an AI engineer earns; inside OpenAI, about 90 percent of what an OpenAI AI engineer earns. "The frontier lab lifts all roles under the same logo," Shibly wrote.
What this means if you work in data
For the practitioners who read us, the distribution matters more than the headline. The specialization still pays: a third more at the median is not nothing, and it has held. But it is no longer compounding, and the distance between the median and the top of the same specialization, in Shibly's phrasing, "has rarely been wider."
That points the career decision somewhere uncomfortable. The biggest lever on compensation in this data is not adding the AI label to your title, which is worth low single digits inside a big firm, and not even the specialization itself, whose premium has plateaued. It is which employer you land at, because a handful of them pay top of market for every role in the building. The label follows the market; the outlier outcomes follow the logo.
Key Points
- 1The median AI premium has plateaued. Holding level constant, an L3 ML/AI engineer earned about 1.34x a non-AI peer in late 2024 ($375K against $279K) and about 1.29x by mid 2026 ($383K against $297K). The gap between ML engineers and data scientists has been similarly frozen at roughly 1.4x for two full years.
- 2The gains are concentrating, not lifting the field: the 90th-percentile-to-median ratio for ML/AI engineers widened from 2.06x to 2.23x in two years while the broader engineering population barely moved, and the extreme packages sit at a short list of frontier labs whose median engineer pay runs about $635K at OpenAI and $665K at xAI, much of it equity marked at private valuations.
- 3The elite is defined by employer, not specialization: the median recruiter at OpenAI earns about $580K, roughly 3.7x the market recruiter median of $155K, a bigger premium than the AI-skill premium itself, though on a sample of seven that Levels.fyi itself labels directional. Inside a frontier lab a recruiter earns about 90 percent of what an AI engineer earns; in the broad market, about 60 percent.
Scoring Rationale
Original compensation analysis prepared for Lets Data Science by Levels.fyi from its salary-submission data, answering our questions in writing with level-controlled cuts that exist nowhere else, including a two-year premium time series, a title-gap comparison, and a dispersion analysis. Compensation is among the highest-interest subjects for our practitioner audience, and the findings cut against the prevailing narrative.
Sources
Original reporting, with the public references used alongside it.
LDS Exclusive
Reporting based on written answers given directly to Let's Data Science by Hakeem Shibly, Levels.fyi.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
