AI jobs

AI Salaries

What AI jobs actually pay in 2026, from a $69,000 entry-level average to a declined $1.5 billion offer — with every figure sourced, dated, and labelled by how trustworthy the data behind it is.

Updated August 2026

Quick answer: A US machine learning engineer earns a median total compensation of $278,000 on self-reported Levels.fyi data — though the narrower “AI engineer” title medians $159,430, and which of the two you are hired under matters more than most people expect — against an audited US median of $133,080 for software developers generally (Bureau of Labor Statistics, May 2024 data). The number most people want is the top end, and it is real: OpenAI’s average stock-based compensation was $1.5 million per employee across roughly 4,000 people in 2025, and Meta offered one researcher a package reported at $1.5 billion over six years, which he turned down. The caveat that matters more than any single figure: AI salary data is unusually bad, because the audited sources are two years stale and the current sources are self-selected, so treat every number below as a reference point rather than a quote — and check the label on which kind of data it is.

This page covers the market. For one role in depth, see prompt engineer salary. For the two very different job markets hiding behind the phrase “AI jobs”, start at the AI jobs hub.

The range, at a glance

This is the whole distribution on one screen. Every row is sourced and dated below.

The spread between the bottom and top of this table is roughly four to five orders of magnitude, and both ends are in the same industry. That is the single most important fact about AI pay in 2026, and it is why an “average AI salary” is close to meaningless.

Read this before you trust any AI salary figure

Every salary page on the internet quotes numbers with equal confidence. They should not be treated equally. There are three tiers of AI pay data and they disagree with each other by enormous margins.

Tier one: audited government data. The US Bureau of Labor Statistics publishes the only figures here that are collected from employers under a statistical methodology rather than volunteered. Its median annual wage for software developers is $133,080, with the lowest ten per cent under $79,850 and the highest ten per cent over $211,450. The problem is currency: that is May 2024 data, published on a page last updated 28 August 2025, and the BLS has no dedicated “AI engineer” occupation code at all. It is accurate and out of date.

Tier two: self-reported aggregators. Levels.fyi, Glassdoor and Levels-style platforms collect submissions from employees. This is where every large AI salary figure you have seen comes from. The data is current and granular, and it is not a random sample — it skews heavily towards people at large, equity-heavy technology employers who have a reason to submit, which pushes medians well above what a typical company pays.

Tier three: job-board averages. ZipRecruiter, Indeed and similar compute averages from advertised postings. These are the least reliable of all, because advertised ranges are negotiating positions and the job-title matching is loose.

Here is how badly they disagree. For the same job title — entry-level AI engineer — Glassdoor reports an average of $128,013 while ZipRecruiter reports $69,362. That is not a rounding difference. It is one source saying nearly double the other for the same words.

Our approach on this page: every figure carries its source and its date, and we tell you which tier it came from. Where sources conflict, we show the conflict rather than picking the flattering number.

What AI and machine learning engineers actually earn

The medians

Levels.fyi data, pulled 15 August 2026, self-reported:

The gap between “machine learning engineer” at $278,000 and “AI engineer” at $159,430 is the most useful thing on this page for anyone choosing what to call themselves. The two titles describe overlapping work. The likeliest explanation for a gap that large is composition rather than skill: “machine learning engineer” remains concentrated in large technology companies with heavy equity, while “AI engineer” has spread rapidly into ordinary industries that pay ordinary salaries. Indeed’s Hiring Lab found US job titles referencing AI more than tripled from 264 in 2022 to 822 by the first quarter of 2026, with 63 per cent now outside traditional technology occupations — and those roles bring their sector’s pay scale with them.

If you are negotiating, the practical implication is direct: the same work advertised under two titles can carry a six-figure difference in expectation, and the title is often negotiable when the salary band is not.

By company

Levels.fyi company medians for machine learning engineers, pulled 15 August 2026:

Two things worth noticing. Google has the widest reported range of the six, topping out above $743,000 — seniority dispersion inside one company can exceed the difference between companies. And Scale AI, a company whose entire business is AI training data, pays its own machine learning engineers the least of this group, which is a useful corrective to the assumption that proximity to AI equals proximity to the money.

The base salary reality check

Total compensation figures are dominated by equity, and equity is not cash. At senior levels, stock commonly accounts for 40 to 70 per cent of the package, and at frontier labs it is the clear majority.

For someone doing real AI work at a typical mid-to-large company that is not a big-tech equity machine, a base salary of $155,000 to $200,000 is the honest range. If you are comparing a big-tech offer against a normal-company offer, compare base against base first, then decide how much you believe the equity.

Entry-level AI salaries

This is the segment where the data is worst, so here is the honest version.

Treat the $69,362 and the $128,013 as measuring different populations rather than as one of them being wrong. Job boards capture a wide, geographically dispersed set of postings including many that are entry-level in name only. Glassdoor skews to people who work at companies worth reviewing.

One definitional trap. “Entry level” in AI engineering does not mean no qualifications. It typically assumes a computer science degree, frequently a master’s, and real project work training or deploying models. A posting that says “entry level” and pays $150,000 is describing an entry-level position on a specialist ladder, not an entry point into the industry. If you are trying to get into AI work without that background, the AI jobs hub covers the realistic routes.

The top end: what “several million a year” actually looks like

This is the part of the market that generates headlines, and it is also the part where the reporting is least reliable. Here is what is verified, what is disputed, and what is simply repeated.

What is solidly reported

OpenAI’s average stock-based compensation reached $1.5 million per employee across roughly 4,000 employees in 2025, according to the Wall Street Journal, reported by Fortune on 18 February 2026. That is an average, not a top-of-band figure. To put it in historical context, the next closest case is Google at its IPO, where average stock compensation was roughly a quarter of a million dollars in inflation-adjusted terms — about one sixth of OpenAI’s level.

OpenAI is spending about 46.2 per cent of its annual revenue on stock-based compensation. Nearly half of what the company earns goes to paying people in equity.

OpenAI completed a roughly $7 billion employee tender offer, reported by TechCrunch on 10 August 2026 at a valuation of about $852 billion. Tender offers matter enormously here, because frontier-lab equity is otherwise illiquid: OpenAI staff hold profit participation units rather than conventional stock, and those cannot be transferred without company approval. A tender offer is the mechanism that converts a paper number into money.

Research scientist stock grants run $2 million to $4 million at a Series D-stage AI startup, according to Menlo Ventures partner Tim Tully, speaking to Fortune. Top OpenAI researchers have been reported earning over $10 million annually, with retention packages above $2 million and equity grants exceeding $20 million offered to prevent departures.

Meta’s own named executive officers, including chief technology officer Andrew Bosworth, have earned total compensation between roughly $20 million and $24 million per year for years, per Meta’s SEC filings. This is the useful anchor: a nine-figure package for a senior AI leader is not absurd against a pay structure that already runs to $24 million for named officers.

The individual packages everyone quotes

The $100 million signing bonus, which mostly did not happen

This is the most-repeated AI salary claim in circulation and it is meaningfully wrong.

On 17 June 2025, OpenAI chief executive Sam Altman said on the Uncapped podcast that Meta had offered his staff signing bonuses as high as $100 million, adding that “so far none of our best people have decided to take them up on that.” It was reported everywhere.

Ten days later, on 27 June 2025, TechCrunch reported what Meta said internally. At a company all-hands leaked to The Verge, Meta CTO Andrew Bosworth said only a few people in very senior leadership roles may have been offered money at that scale, and that “the actual terms of the offer” were not a “sign-on bonus. It’s all these different things.” In other words: a multi-year package weighted towards restricted stock units vesting against tenure or performance, not a cheque on day one. Bosworth’s summary of Altman’s framing was blunt: “Altman was suggesting that we’re doing this for every single person. Look, you guys, the market’s hot. It’s not that hot.”

The poached researchers said the same. Lucas Beyer, who left OpenAI for Meta with two colleagues from its Zurich office, posted publicly: “1) yes, we will be joining Meta. 2) no, we did not get 100M sign-on, that’s fake news.”

What to take from this: multimillion-dollar annual packages for frontier AI researchers are real and well evidenced. Nine-figure total packages for a very small number of senior leaders are plausible and consistent with Meta’s existing executive pay. Instant $100 million cash bonuses as a general practice are not a thing, and any salary guide presenting them as a market rate is repeating a headline rather than reporting a fact.

Why the top end is so extreme

Three structural reasons, and none of them is that these people are a thousand times more productive than a good engineer.

1. Scarcity of a specific track record. The people commanding these packages are a few hundred individuals worldwide who have personally led the training of a frontier model. That credential cannot be acquired by study; it requires having been in the room.
2. The cost of being wrong is asymmetric. Against Meta’s reported capital expenditure of $125 billion to $145 billion for 2026, a $250 million hire is roughly two tenths of one per cent of the budget. If that person improves the outcome of the spend even marginally, the package is rational.
3. It is equity, and equity is a bet. Almost all of these figures are multi-year grants whose value depends on the company’s valuation holding. A $250 million package at a valuation that halves is a $125 million package.

The AI pay premium for everyone else

You do not have to be an AI engineer to be paid more for AI skills, and this is where the numbers apply to the largest number of people.

PwC’s Global AI Jobs Barometer, built from more than one billion job advertisements across 27 countries, puts the wage premium for AI skills at 25 per cent in 2024, 57 per cent in 2025 and 62 per cent in 2026. Within the US data the premium reaches 84 per cent for chief executives and managing directors and 49 per cent for lawyers, and falls to near zero for manual roles such as freight handling.

Lightcast, analysing 1.3 billion job postings, found a smaller effect: postings requiring AI skills advertise salaries 28 per cent higher, about $18,000 more per year, rising to a 43 per cent premium where two or more AI skills are required.

The two figures are not in conflict — they measure different things. PwC compares what AI-skilled roles pay within an occupation. Lightcast measures what employers advertise when a posting lists AI skills as a requirement.

Both carry the same critical caveat, and it is one most coverage omits. These are cross-sectional comparisons of advertised salaries. Neither tracks the same worker before and after learning AI, and neither fully controls for seniority, employer or location. The premium reflects what the market pays for roles that demand AI skills. It is not evidence that adding a skill to your CV produces a 62 per cent raise. Roles demanding AI skills also tend to be more senior and at better-paying employers, and that composition effect is doing an unknown share of the work.

The contradiction nobody reconciles: record pay, record layoffs

Both of these are true in 2026, simultaneously, sometimes at the same company.

• US technology employers have cut close to 140,000 jobs so far in 2026, with Amazon, Oracle, Meta and Microsoft accounting for nearly 50,000 between them.
• Meta cut about 8,000 roles, roughly ten per cent of its workforce, while moving around 7,000 employees into AI-focused roles and committing $125 billion to $145 billion in 2026 capital expenditure.
• Meta paused AI hiring after roughly 50 senior hires, following its 2025 recruitment spree.
• Tech salaries are broadly flat against 2025, with AI engineering among the few exceptions.

At the same time:

• Technology unemployment sits at 2.9 per cent against a US national rate of 4.2 per cent, with more than 280,000 new technology job postings in June 2026, a sixth consecutive month of growth (CompTIA).
• LinkedIn recorded 1.3 million new AI-related roles globally over two years, plus more than 600,000 AI-enabled data centre jobs.

The reconciliation is that this is a reallocation, not a contraction. Payroll is being moved out of general engineering and support functions and into AI infrastructure and AI-adjacent roles — partly as headcount, overwhelmingly as capital expenditure on data centres. If you are inside the reallocation you are being paid more than ever. If you are outside it, the same employer is cutting.

What this means for your expectations: the extraordinary numbers on this page describe a narrow, contested segment of a market that is simultaneously shedding tens of thousands of jobs. Both facts should inform your planning. Neither should be planned around alone.

What to expect, by situation

You are a software engineer moving into AI. This is the shortest path to the medians on this page. Expect $245,000 to $278,000 total compensation as a realistic mid-level target at an equity-paying employer, and $155,000 to $200,000 base at a normal one. Ship something using models in production before you apply.

You are early-career. Expect wide variance and do not anchor on a single source. Big-tech and AI-startup new-graduate offers cluster at $140,000 to $180,000 total compensation; the broader market average is far lower. Negotiate the title as well as the number.

You have deep domain expertise outside technology. The PwC and Lightcast premiums apply to you, and this is the fastest-growing category — nearly two thirds of new AI job titles now sit outside technology occupations. You are competing in a much thinner field than a junior engineer.

You are aiming at a frontier lab. Compensation is dominated by illiquid equity, so ask two questions before the headline number: what is the tender offer history, and what is the vesting schedule. A $1.5 million average means nothing if you cannot sell.

You are considering AI training contract work. This is a different market with different economics, and the figures on this page do not apply. See get paid to train AI and the AI jobs hub for what that work actually pays.

Frequently asked questions

How much does an AI engineer make in 2026?

On Levels.fyi self-reported data pulled in August 2026, the median total compensation is $278,000 for machine learning engineers and $245,000 for ML or AI software engineers, while the narrower “AI engineer” title shows a median of $159,430. For context, the audited US median for software developers generally is $133,080 according to Bureau of Labor Statistics data from May 2024. Total compensation figures include equity, which commonly accounts for 40 to 70 per cent of a senior package, so base salary alone is typically far lower — around $155,000 to $200,000 at a company that is not an equity-heavy large technology employer.

What is the highest-paying AI job?

Frontier model research is the highest-paying role in the industry by a wide margin. OpenAI’s average stock-based compensation reached $1.5 million per employee across roughly 4,000 employees in 2025 according to the Wall Street Journal, with top researchers reported earning over $10 million annually. The individual outliers go far higher: Meta reportedly offered Thinking Machines Lab co-founder Andrew Tulloch up to $1.5 billion over six years, which he declined, and reportedly paid 24-year-old researcher Matt Deitke about $250 million over four years. These packages go to a few hundred people worldwide who have personally led frontier model training.

Do AI jobs really pay millions of dollars a year?

Some do, for a very small number of people, and the structure matters more than the headline. Multimillion-dollar annual packages for frontier AI researchers are well evidenced — OpenAI averaged $1.5 million in stock compensation per employee in 2025. Nine-figure total packages exist for a handful of very senior leaders and are consistent with Meta’s existing executive pay of $20 million to $24 million a year for named officers. But almost all of it is multi-year equity rather than cash, its value depends on the valuation holding, and at frontier labs it is illiquid until a tender offer. The typical AI engineer earns a fraction of these numbers.

Did Meta really offer $100 million signing bonuses?

Not in the way it was reported. Sam Altman said on the Uncapped podcast on 17 June 2025 that Meta had offered OpenAI staff signing bonuses as high as $100 million. Meta’s chief technology officer Andrew Bosworth told a company all-hands ten days later that only a few very senior leadership roles may have been offered money at that scale and that the terms were not a sign-on bonus but a multi-year package of “all these different things”, adding that “the market’s hot. It’s not that hot.” Researcher Lucas Beyer, who did move from OpenAI to Meta, posted publicly that he did not receive a $100 million sign-on and called the claim “fake news”.

What is the entry-level AI engineer salary?

The sources disagree by roughly a factor of two, so treat any single figure with caution. Glassdoor reports an average of $128,013 with a typical range of $96,010 to $177,937, while ZipRecruiter reports $69,362. Reported new-graduate offers at big technology companies and AI startups cluster at $140,000 to $180,000 total compensation with base salaries of $90,000 to $135,000, and entry-level base offers in San Francisco and New York routinely start at $115,000 to $135,000. Note that “entry level” in AI engineering normally assumes a computer science degree, often a master’s, and demonstrable project work.

Is an AI engineer paid more than a machine learning engineer?

No, and the gap runs the other way by a surprising margin. Levels.fyi data pulled in August 2026 shows a median of $278,000 for machine learning engineers against $159,430 for the “AI engineer” title. The likeliest explanation is the composition of who holds each title rather than the difficulty of the work: machine learning engineer remains concentrated in large equity-paying technology companies, while AI engineer has spread rapidly into other industries that pay their own sector’s rates. US job titles referencing AI more than tripled between 2022 and early 2026, and 63 per cent of them now sit outside technology occupations.

How much more do AI skills pay in a non-AI job?

PwC’s Global AI Jobs Barometer, based on more than a billion job advertisements across 27 countries, puts the AI skills wage premium at 62 per cent in 2026, up from 57 per cent in 2025 and 25 per cent in 2024. Lightcast, analysing 1.3 billion postings, found a smaller 28 per cent premium, roughly $18,000 a year, rising to 43 per cent where a posting requires two or more AI skills. Both are cross-sectional comparisons of advertised salaries that do not track the same worker before and after learning AI and do not fully control for seniority, employer or location, so they describe what AI-demanding roles pay rather than proving that adding a skill produces that raise.

Why are AI salaries so high if tech companies are laying people off?

Because it is a reallocation of payroll rather than a general expansion. US technology employers cut close to 140,000 jobs in 2026, with Meta alone cutting about 8,000 roles while simultaneously moving around 7,000 employees into AI-focused roles and committing $125 billion to $145 billion in capital expenditure. Technology unemployment remains low at 2.9 per cent against a national 4.2 per cent, and tech salaries are broadly flat against 2025 with AI engineering among the few exceptions. Money is moving out of general engineering and into AI infrastructure and AI-adjacent roles.

Can you trust AI salary data?

Only with the label attached. Audited Bureau of Labor Statistics figures are collected under a statistical methodology but are from May 2024 and have no AI engineer occupation code at all. Levels.fyi and Glassdoor are current and granular but self-selected, skewing towards employees of large equity-paying technology companies. Job-board averages from sites like ZipRecruiter are the least reliable, because advertised ranges are negotiating positions. The scale of disagreement is easy to demonstrate: for entry-level AI engineer, Glassdoor reports $128,013 and ZipRecruiter reports $69,362 for the same job title.

What does a prompt engineer earn compared with an AI engineer?

Prompt engineering is a distinct and much narrower market with its own pay dynamics, and it has not tracked AI engineering upwards. We cover the current figures, the demand trend and whether the role still exists as a standalone job on our dedicated prompt engineer salary page rather than restating them here.

How much do AI data annotators and AI trainers get paid?

This is a completely separate market from salaried AI engineering, and no reliable survey of it exists for 2026. The verified individual datapoints span an enormous range: contributors on major platforms have reported all-in effective rates of around $4 to $7 an hour once unpaid onboarding and training time is counted, while credentialed specialists have been paid $150 an hour on OpenAI’s financial-modelling work and $300 or more an hour for medical degrees and PhDs on other platforms. The major platforms publish no rate cards. See the AI jobs hub and get paid to train AI for the detail.

Related pages

Pay and careers: Prompt engineer salary · AI jobs hub · Remote AI jobs

Contract and gig work: Get paid to train AI · AI training jobs · Data annotation jobs · AI side hustles

Build AI fluency: Best AI models · Best AI for coding

All figures verified against their named sources on 15 August 2026, and each is labelled by data type: audited government statistics, self-reported aggregator data, or press-reported individual cases. Self-reported compensation data is not a random sample and skews towards large equity-paying employers. Individual pay packages described here are press-reported, some are disputed by the parties involved, and none has been independently confirmed by us. Equity figures depend on valuations that change.