AI jobs

Machine Learning Jobs

Nine machine learning job titles, where the postings are listed, and which commonly recommended job boards have closed, checked on 22 September 2026.

Updated September 2026

Quick answer: “Machine learning job” is not one job — it covers at least nine posted titles, and the two most similar-sounding of them differ by $118,570 in median total compensation on Levels.fyi self-reported data, so the title you search for decides what you find. The highest-volume places to look are the employers’ own boards rather than aggregators: OpenAI listed 810 open roles, xAI 275 and Apple 301 in its machine learning and AI team on their own careers pages on 22 September 2026, while Google DeepMind runs no job board at all and posts everything into Google’s general system. Several commonly recommended boards do not carry ML jobs: ai-jobs.net no longer exists as an ML board, Kaggle’s job board has been shut since December 2020, and Hugging Face has no third-party job board.

This page is the practical one: what the roles are, and where the postings are. For the routes into this work from wherever you are starting, see AI career paths. For the full pay picture across the whole industry, including contract work and the frontier-lab outliers, see AI salaries. For the two very different markets hiding behind the phrase “AI jobs”, start at the AI jobs hub.

All listing counts, programme statuses and board checks on this page were carried out on 22 September 2026 and are labelled with that date. Counts drift daily and several of the deadlines below fall within days of it.

The nine titles, decoded

Employers use these titles inconsistently, so a search on one title returns only part of the market.

The compensation figures are Levels.fyi self-reported data pulled on 15 August 2026 and are the same figures we publish on AI salaries, where the sourcing and the caveats are set out in full. They are not a random sample: they skew towards employees at large, equity-heavy employers. Where we write “data not available”, it is because no figure we would stand behind was published for that title — not because the role pays nothing.

Machine learning engineer against AI engineer

On Levels.fyi data pulled 15 August 2026, “machine learning engineer” medians $278,000 and “AI engineer” medians $159,430 — a gap of $118,570 for work that overlaps heavily.

The working distinction: a machine learning engineer is accountable for a model, and an AI engineer is accountable for a product that calls someone else’s model. If the job description talks about training runs, datasets, evaluation harnesses and drift, it is the first. If it talks about retrieval, prompts, agents, latency and cost per request, it is the second.

Research scientist against research engineer

These two are posted side by side at the frontier labs and are not interchangeable. Research scientist roles are judged on novel contributions and normally expect a publication record; research engineer roles are judged on making large training and evaluation runs work.

At xAI — now SpaceXAI — the model training roles are posted as “Member of Technical Staff”, a label that contains none of the words a candidate would search for.

Applied scientist

Amazon posts most of its science roles as “applied scientist”. On 22 September 2026, Amazon’s science careers site listed 684 open roles, of which 393 were categorised Applied Science and 83 Machine Learning Science; its machine-learning-science category page separately stated 550.

Inference and performance engineering

Inference and performance engineering — making a trained model cheaper to serve — is now posted as a distinct title. It sits between ML and systems programming: CUDA kernels, quantisation, batching, throughput per dollar.

Prompt engineer

Prompt engineer has largely been absorbed into AI engineer as a posted standalone title. We track what remains of it, including the pay evidence and the demand trend, on prompt engineer salary rather than restating it here.

Search the set, not the title

Result counts vary sharply by title. On Wellfound on 22 September 2026, with the same remote filter on the same platform on the same day:

• machine-learning-engineer returned 302 results across 11 pages
• artificial-intelligence-engineer returned 6,701 results across 122 pages

A 22-fold difference between two labels. Neither number is the size of the market. Both are a slice of it.

The working set: machine learning engineer, ML engineer, MLE, AI engineer, applied scientist, research scientist, research engineer, member of technical staff, deep learning engineer, perception engineer, ML platform engineer, inference engineer.

Where the jobs actually are

The general boards

LinkedIn, Indeed, Glassdoor and ZipRecruiter are free to search and carry the widest net. They also carry the problems below, and we could not reproduce a defensible ML listing count on any of them — counts change with login state, location filter and query string, so no count is given here.

Duplication. Nvidia’s job pages carry a referrer map that pushes a single requisition simultaneously to Indeed, LinkedIn, Glassdoor and ZipRecruiter, including sponsored and Appcast variants. One opening, four boards, several listings.

Multi-location splitting. Apple publishes one requisition as separate location-specific listings with separate URLs, so aggregators ingest the same role several times over.

Stale listings. On Wellfound on 22 September 2026, the first page of machine learning engineer results carried live apply buttons on postings marked four and five years old.

The specialist boards that are alive

80,000 Hours curates rather than mirrors. Its Anthropic entry showed 39 open roles, explicitly safety, policy and security only, against roughly 621 roles on Anthropic’s own board; its OpenAI entry showed 27 against 810.

aijobs.ai’s homepage stated “20,000+ Jobs”; its machine learning category showed 551.

Built In’s ML listings are mostly established US employers — ServiceNow, Optum, Capital One, JPMorganChase, General Motors, Datadog, Citadel — and it carried no frontier-lab roles.

Boards that have closed or changed

Every entry was checked on 22 September 2026.

Three other commonly suggested channels carry no job listings: Latent Space states that it has no job board, the Weights & Biases community forum has no jobs category, and the fast.ai forums have none either. The PyTorch forums do have a jobs category, but it carried roughly five topics across all of 2026, several with no replies and one spam post.

Employers’ own boards

Several employers post roles that appear on no aggregator.

Five points on that table.

Google DeepMind has no job board. Its careers page routes into Google’s general application system.

Microsoft has four separate hiring surfaces, and they do not share a job store. The roles on microsoft.ai/careers — pre-training, large-scale reinforcement learning, frontier risk and alignment — use different identifiers from the main Microsoft portal and will not appear if you search it.

Apple’s two research programmes are applied for entirely outside its main board, at machinelearning.apple.com/work-with-us. They appear in no requisition count, including the 301 above.

Palantir. Of 280 roles on its board, six had AI in the title, and no title contained “machine learning”, “research scientist”, “data scientist”, “computer vision” or “LLM”.

Anduril. Its postings require applicants to be a US person because of access to export-controlled information, with clearance-eligibility questions as mandatory application fields.

Two further notes. xAI became SpaceXAI when SpaceX acquired it, announced 2 February 2026, and its board has absorbed X Platform hiring; its ML roles are a small minority of the 275. Posting URLs do not persist: closed Anduril requisitions redirect to the listing index rather than returning an error, old Scale posting URLs redirect, and dead Hugging Face job identifiers bounce to a not-found parameter.

Residencies and fellowships: what is actually running

Every status below was checked on 22 September 2026.

Careful with the Nvidia fellowship page. It displays a 15 September deadline, which reads as this month. It is the 2025 deadline for the 2026 to 2027 cycle, and that cycle is closed.

The residencies that are gone

Google AI Residency. Its page still loads, frozen in the 2018 application cycle — it invites applications for 2019, references a Google+ account, and the last content Google published about the programme was in November 2019. It does not appear on Google Research’s current careers page.

Meta AI Residency. Its page carries a 2026 footer over 2023 content, states that applications for the 2023 cohort are closed, says new residents will start in autumn 2023, and lists a legacy contact address.

Microsoft Research’s Redmond AI residency is absent from both the academic programmes page and Microsoft Research Redmond’s own opportunities page, while Microsoft Research Cambridge’s FAQ still links to it and dead-ends.

A claim circulating since March 2026 that Google and Meta are launching new AI residencies is not supported by either company’s own pages. It traces to an HR-trends aggregator with no primary source, no dates and no link. Both companies’ residency pages remain frozen.

The Fatima Fellowship has been renamed to the Fatima Institute for Global AI Research; its old site is frozen at 2023 and the new one could not be read, so its current cohort status is data not available.

What is open right now

Four dated items, because they expire.

Google internship, 25 September. Google’s software engineering internship for summer 2027 at bachelor’s level states an application window open until 25 September 2026, with Google’s own caveat that it may close earlier if all projects fill. Note that Google’s levels differ: its PhD research internship for summer 2027 in Canada runs to 26 February 2027.

MATS results land soon. Stage one for the winter 2027 cohort closed on 6 September 2026 and offers go out in early to mid November 2026. Mentor applications are open now if you are on the other side of that.

NeurIPS 2026’s careers board is live, free, and needs no conference ticket. It opened on 1 September 2026, accepts applications until 8 December 2026, and closes on 12 January 2027. On 22 September it carried seven postings from three organisations. NeurIPS 2026 is held in Sydney from 6 December, with satellite sites in Atlanta and Paris (AI Weekly).

The other three major conference boards are shut. ICML 2026 was held in Seoul on 7 to 9 July and its board closed on 16 August 2026; ICLR’s board has rolled over to 2027 and is closed, with ICLR 2027 on 26 to 28 April; CVPR 2026 was held in Denver on 5 to 7 June and its careers site closed on 7 July 2026, with CVPR 2027 in Seattle on 20 to 25 June.

Other internship rounds confirmed open and rolling, with no stated deadline: Amazon’s software development engineer internship for summer 2027 (12 weeks, advertised at the equivalent of $109,395 a year), Scale AI’s software engineering internship for summer 2027 (roughly $60 an hour, requiring a graduation date in late 2027 or early 2028), Two Sigma’s summer 2027 programme (10 weeks in New York at $3,800 a week for bachelor’s, $3,900 for master’s, $4,200 for PhD), and Netflix, whose own guidance is that it recruits from late summer through to the end of March.

For tracking the rest, the community-maintained Summer 2027 internships repository is updated daily and carries 47,500 stars. It is an aggregator, so verify anything you act on at the employer.

Hiring seasonality

No primary data source shows a seasonal hiring peak for machine learning. The US Bureau of Labor Statistics JOLTS series for the Information sector and Indeed’s Hiring Lab job postings index are both seasonally adjusted, which means seasonal variation has been deliberately removed from them. Quoting swings in either as a seasonal peak misrepresents the method.

The widely repeated claims — that January to March and September to October are the peaks, that success rates are 45 per cent higher in peak months, that July and August are 40 to 60 per cent quieter — trace to aggregator blogs with no stated methodology.

What the JOLTS data does show is a contraction rather than a cycle. Job openings in the Information sector fell from 157,000 in July 2025 to 96,000 in July 2026, a job openings rate falling from 5.2 to 3.3 per cent, with a rise from the May 2026 low of 72,000 (Bureau of Labor Statistics). Information is a broad sector, not a measure of machine learning specifically.

Most employers checked recruit on a rolling basis until roles fill, with no published deadline; Google’s bachelor’s-level internship window is the main dated exception.

Anthropic’s Fellows Program requires existing work authorisation in the US, UK or Canada, with no visa sponsorship.

Best route for each situation

You are a working software engineer moving into ML. Go direct to employer boards and search the full title set, not “machine learning engineer”. Read AI career paths for the lateral move in detail.

You want a normal, well-paid ML job at a recognisable company. Built In is the best-fit board: it shows salary bands inline and is heavy on US enterprise employers. It carried no frontier-lab roles on 22 September.

You are targeting a frontier lab. The labs’ own boards carry the most roles — OpenAI, Anthropic and xAI all publish counts in the hundreds. Add 80,000 Hours for the safety-specific subset, and check machinelearning.apple.com/work-with-us and microsoft.ai/careers because those roles appear nowhere else.

You have no ML track record and need a structured way in. The live options with no PhD requirement are the Anthropic Fellows Program (open now), Ai2’s predoctoral programme (rolling, and it sponsors visas), Apart Research (no essays, monthly entry) and SPAR (unpaid, remote, opens around December). Cohere Labs Scholars explicitly does not require prior research experience and is expected to reopen this autumn.

You are a student. Google’s bachelor’s-level internship window closes on 25 September 2026, and most others are rolling. The DeepMind Student Researcher programme is one application covering several Google AI teams, but requires four days a week on site.

You want AI work without an engineering background. That market is covered at best AI jobs for beginners and AI training jobs, and it pays by the hour rather than by salary.

You need the job to be remote. Wellfound’s remote filter returned the most remote ML listings of the boards we checked, including some stale postings. Remote AI jobs covers how remote this market actually is.

Frequently asked questions

What is a machine learning job?

It is not one job. The machine learning family covers at least nine posted titles: machine learning engineer, AI engineer, ML or AI software engineer, applied scientist, research scientist, research engineer, MLOps or ML platform engineer, data scientist, and inference or performance engineer. They differ in what you build, what qualifications they expect and what they pay — on Levels.fyi data pulled 15 August 2026, machine learning engineer medians $278,000 in total compensation while AI engineer medians $159,430. A working dividing line is that a machine learning engineer is accountable for a model while an AI engineer is accountable for a product built on somebody else’s model.

What is the difference between a machine learning engineer and an AI engineer?

A machine learning engineer owns a model in production — training pipelines, evaluation, serving, monitoring and retraining. An AI engineer builds products on top of existing models, working with prompts, retrieval, evaluation and orchestration rather than training runs. The work overlaps and employers use the labels inconsistently, but the reported pay differs: Levels.fyi data pulled on 15 August 2026 shows a median total compensation of $278,000 for machine learning engineer against $159,430 for AI engineer, a gap of $118,570.

Where can I find machine learning jobs?

Employers’ own boards carry the largest counts. On 22 September 2026, OpenAI listed 810 open roles, Anthropic roughly 621, xAI 275, Apple 301 in its machine learning and AI team, Amazon 684 on its science careers site, and Microsoft Research 126. Among third-party boards, 80,000 Hours curates frontier-lab and AI safety roles, Built In lists mostly US enterprise employers and shows salary bands inline, and Wellfound lists startup roles. Note that Google DeepMind has no job board of its own and posts into Google’s general system, and that several employers — Apple’s research programmes and Microsoft AI among them — post roles that appear on no aggregator.

What is the best job board for machine learning jobs?

There is no single best one, because each covers a different slice of the market. For frontier labs and AI safety work, 80,000 Hours curates roles and showed 963 listings on 22 September 2026, but read its per-employer counts as a filtered subset rather than a total — its Anthropic entry showed 39 safety and policy roles against roughly 621 on Anthropic’s own board. For established US employers, Built In lists them and displays salary ranges. For startups, Wellfound. The general boards — LinkedIn, Indeed, Glassdoor, ZipRecruiter — carry the widest range but include duplicate postings, because some employers syndicate a single requisition to several of them.

Which job boards for machine learning jobs are out of date?

Several commonly recommended ones have closed or changed, checked individually on 22 September 2026. ai-jobs.net no longer exists as a machine learning board — every URL redirects to foorilla, a general developer careers platform, and its machine learning category pages are gone. Kaggle’s job board has been shut since 22 December 2020. Hugging Face has no job board for third-party employers; the product at that name is a paid GPU compute service. Papers With Code and its jobs board were reportedly sunset on 24 July 2025. Triplebyte closed in March 2023 and Hired.com reportedly in May 2024. aijobsdb.com still loads but has ingested nothing since late March.

Do I need a PhD for a machine learning job?

For most of this family, no. Research scientist roles at frontier labs normally expect a publication record and in practice a PhD. Research engineer, machine learning engineer, ML platform, AI engineer and inference engineer roles generally do not,. Several structured entry programmes state the point explicitly: MATS, LASR Labs and SPAR carry no degree requirement, the Anthropic Fellows Program asks for a bachelor’s degree or equivalent, Cohere Labs Scholars states that prior research experience is not needed, Microsoft Research Cambridge’s residency FAQ says a PhD may not be required for some roles.

How much do machine learning jobs pay?

On Levels.fyi self-reported data pulled 15 August 2026, the US median total compensation is $278,000 for machine learning engineer, $245,000 for ML or AI software engineer and $159,430 for AI engineer. Those figures include equity, which commonly makes up 40 to 70 per cent of a senior package, so base salary alone is typically far lower — around $155,000 to $200,000 at an employer that is not an equity-heavy large technology company. The data is self-reported and skews towards large employers rather than being a random sample. We publish no figure for applied scientist, research engineer, MLOps or inference engineer because no source we would stand behind gives one. The full picture, including the frontier-lab outliers and the contract market, is on AI salaries.

Can I get a machine learning job with no experience?

Structured programmes are one route in. On 22 September 2026, the Anthropic Fellows Program was open at $3,850 a week with no PhD requirement, Ai2’s predoctoral programme accepts bachelor’s and master’s holders on a rolling basis and sponsors visas, Apart Research uses monthly hackathons as its entry point with no application essays, and SPAR is unpaid, remote and open to anyone aged 13 and over. If what you want is paid AI work you can start this month rather than a career track, that is a different market — see best AI jobs for beginners.

Are machine learning jobs remote?

Partly. Wellfound’s remote filter returned 302 machine learning engineer roles and 6,701 AI engineer roles on 22 September 2026, so remote listings exist in volume. But several of the most sought-after routes are explicitly on-site: the OpenAI Residency is based in San Francisco, DeepMind’s Student Researcher programme requires a minimum of four days a week in a Google office, and the Anthropic Fellows Program runs out of Berkeley and London workspaces while permitting remote participation from the US, UK and Canada. See remote AI jobs for more.

Which companies are hiring machine learning engineers in 2026?

As of 22 September 2026, the largest published role counts were OpenAI at 810, Anthropic at roughly 621, Amazon at 684 science roles, Apple at 301 in its machine learning and AI team, xAI at 275 and Microsoft Research at 126. Meta, Nvidia, Google DeepMind, Mistral and Anduril publish no total on their boards. Two notes: well over half of xAI’s 275 roles are datacentre and physical infrastructure positions rather than machine learning, and Palantir listed 280 roles of which only six had AI in the title and none used “machine learning”, “research scientist” or “data scientist”.

Are AI residency programmes still running in 2026?

Some are; others have stopped. Running or recently open on 22 September 2026: the Anthropic Fellows Program, the Anthropic Institute fellowship, Ai2’s predoctoral programme, Apart Research, Microsoft Research Cambridge’s residency and DeepMind’s Student Researcher programme. Closed but recurring: OpenAI’s Residency, MATS, SPAR, Cohere Labs Scholars and Apple’s AIML Residency. Gone: Google’s AI residency page has not been updated since the 2018 application cycle and Google has published nothing about it since November 2019; Meta’s residency page still says its 2023 cohort is closed; and Microsoft Research’s Redmond AI residency has disappeared from its own programme listings. A claim circulating since March 2026 that Google and Meta are launching new residencies is not supported by either company’s pages.

What job titles should I search for to find machine learning jobs?

Search a set rather than one term, because the label changes the result count dramatically — on Wellfound on 22 September 2026, the same platform with the same remote filter returned 302 results for machine learning engineer and 6,701 for AI engineer, a 22-fold difference. A working set is: machine learning engineer, ML engineer, MLE, AI engineer, applied scientist, research scientist, research engineer, member of technical staff, deep learning engineer, perception engineer, ML platform engineer and inference engineer. Two specific cases: Amazon posts most of its machine learning work as “applied scientist”, and frontier labs including xAI now post model training roles as “member of technical staff”, a title that contains no ML terms.

When is the best time of year to apply for machine learning jobs?

No primary data source shows a seasonal hiring pattern in this field: the two best data sources, the Bureau of Labor Statistics JOLTS series and Indeed’s Hiring Lab postings index, are both seasonally adjusted, so seasonal variation has been removed from them by design and cannot be read back out. The commonly quoted claims about peak months and success rates trace to blogs with no stated methodology. Most major employers checked recruit on a rolling basis until roles fill, with few published deadlines.

Is machine learning still a good career in 2026?

The data points in two directions. Median total compensation for a machine learning engineer was $278,000 on August 2026 self-reported data, and the frontier labs were advertising hundreds of roles each in September 2026. At the same time, job openings in the US Information sector fell from 157,000 in July 2025 to 96,000 in July 2026 on Bureau of Labor Statistics data. AI career paths sets out the alternatives and how well each is evidenced.

Related pages

Roles and pay: AI salaries · AI career paths · Prompt engineer salary · Remote AI jobs

Starting out: AI jobs hub · Best AI jobs for beginners · AI training jobs · Data annotation platforms

Build the skills: Best AI for coding · Best AI models · Best AI agents · What is agentic AI

Every board, careers page and programme status on this page was checked on 22 September 2026, and listing counts are snapshots from that date which will drift. Counts marked as stated on the page come from the employer’s own board; Anthropic’s total and Palantir’s were counted from their boards rather than published. Compensation figures are Levels.fyi self-reported data pulled 15 August 2026, which is not a random sample and skews towards large equity-paying employers. Where a figure or status could not be verified at source, it is labelled as data not available or as third-party reporting rather than stated as fact.