Which AI Companies Are Hiring Right Now? We Counted 3,645 Open Jobs on 20 Boards (Aug. 22, 2026)

Saturday, 6:40 a.m. Twenty job boards, one script, one cup of coffee. By the time the coffee was cold the script had read every posting that OpenAI, Anthropic, Waymo, xAI, Figure and fifteen other AI and robotics companies had live on their own career pages, and it had a number nobody else on the first page of Google seems to have: 3,645 open roles, 2,104 of them with a pay range printed in the posting.

You have probably seen the other kind of number. “Microsoft is hiring 4,200+ AI engineers.” “Amazon, 4,000+.” They circulate on LinkedIn and on job-aggregator pages without a date, a method or a link you can click. We cannot verify them and neither can you. So this desk does the slower thing: we count what is actually posted, we date it, and we link every row to the employer. If a posting closes at noon, the link will tell you. That is the whole idea of The Jobs Desk, which starts today.

The count: 20 boards, 3,645 open roles

Here is the board-by-board count. “Open roles” is every live posting on the company’s board worldwide; “US” removes postings whose location names another country; “Remote” counts postings the employer marked remote or remote-eligible; “Pay band printed” counts postings with a salary range in the text; the last column is the median of those US ranges where at least fifteen postings print one.

Two things in that table are worth a second look. First, the frontier labs are not small employers anymore: OpenAI’s 754 and Anthropic’s 512 open roles are the hiring plans of companies that expect to be several times larger next year. Second, the robot companies — Figure, Apptronik, Agility, 1X, Physical Intelligence — have 378 roles between them, nearly all on-site, nearly all in California or Texas. That is what “physical AI” looks like on a job board: you have to be in the building.

A job board is a company’s budget with the numbers scratched out. Count the rows and you can read the budget anyway.

Where the jobs actually sit

San Francisco is still the center of gravity, with 969 of the 2,882 US roles, and 809 more are marked remote. Mountain View’s 233 are mostly Waymo. The number that stopped us was Memphis, Tennessee: 103 roles, plus a cluster across the state line in Southaven, Mississippi. That is xAI’s data-center build, and it is the reason this article has a section on jobs that do not require a degree.

What the postings say about pay

California, New York, Washington and Colorado require employers to print a pay range in the posting, and because most of these companies hire in at least one of those states, 2,104 of the 3,645 postings (58%) carry a salary band. We took the median of the low ends and the median of the high ends for each company’s US postings. Here is what the employers themselves wrote down.

Anthropic prints a range on 385 of its 512 postings and its median band is the highest on the list, roughly $305K–$405K. OpenAI prints one on 608 of 754, median about $239K–$356K, with its research and inference engineering roles reaching $555K at the top of the band. Waymo, Scale AI and Sierra sit around $200K–$260K. The robot companies print lower bands — Figure AI about $150K–$250K, 1X $170K–$225K — which tells you something true about hardware margins versus software margins.

AI jobs that don’t require a degree: the Memphis list

Most articles about AI careers begin with a master’s degree. The job boards begin somewhere else. Of the 85 postings whose titles say technician, apprentice, intern or fellow, the largest single group is xAI’s data-center build in Memphis and Southaven, and we read the qualifications sections so you do not have to guess:

• Associate Data Center Operations Technician (xAI, Memphis/Southaven): “High school diploma or equivalency certificate. 0–2 years of experience working with server, storage, compute, or network hardware (recent graduates or early-career candidates encouraged to apply).”
• Data Center Operations Technician (xAI): high-school diploma or equivalency, two or more years with hardware.
• IT Services Technician (xAI, Memphis): a degree is “preferred,” equivalent experience accepted, and the posting prints $61,200–$91,800.
• Site Operations Technician (OpenAI, San Francisco): “an Associate’s degree, technical certification, military experience, or equivalent practical experience” — and a printed band of $157,000–$221,000 plus equity. Yes, San Francisco rent is the reason that number is what it is. It is still a technician posting with no bachelor’s requirement.

Read the “what it asks for” column the way a hiring manager would: diploma or equivalency means the door is open; preferred means the door is open if you bring experience; apprentice means they expect to teach you. None of these rows is a promise of an interview. All of them are real postings on the employer’s own site as of this morning.

The AI industry is hiring electricians, fiber technicians and facilities crews in Tennessee. Nobody puts that in the keynote.

The federal door and the national labs

Type “artificial intelligence” into USAJOBS and you get 148 announcements this morning. Most are not AI jobs: the phrase simply appears somewhere in the duties. These nine are the ones that actually are, or that open a data or research path, and every one of them is open to the public. Federal postings have something company boards almost never show — a closing date and the full pay range.

Three of these close within five days, which is normal: federal announcements are often open for a week. The two NIST roles are the ones to notice if you work in AI safety, evaluation or standards; NIST writes the AI Risk Management Framework the rest of the government cites. The NSA program is rare in a different way — a single announcement that runs from entry level (GG-7, $87,362) to expert, with many openings. The Fiscal Service and DIA data-scientist announcements are the most ordinary-looking rows here, and for that reason the most reachable.

The Department of Energy’s seventeen national laboratories are the other federal door, and they do not use USAJOBS at all: Oak Ridge, Livermore, Los Alamos, Argonne, Sandia and the rest hire on their own career sites, listed on one DOE page. Their machine-learning postdoc postings come and go in weeks — an Oak Ridge “Scientific Machine Learning” postdoc posted July 28 was already closed when we checked — so the Monday edition of this desk will track them.

For the wider market the Bureau of Labor Statistics is the calm voice in the room: the 2024 median was $140,910 for computer and information research scientists, with employment projected to grow 20 percent from 2024 to 2034 and about 3,200 openings a year; for data scientists, $112,590, 34 percent growth and about 23,400 openings a year. Those are the numbers behind the numbers above — the whole occupation, not the twenty best-funded employers in it.

How to read a posting in 60 seconds

1. The date may be a lie, innocently. Anthropic, xAI and Waymo refresh the “updated” date on every posting nightly. Our script found 512 of Anthropic’s 512 roles “updated” yesterday. A posting that has been open for five months looks brand new. Use the job ID, not the date, to judge age — lower IDs on the same board are older.
2. Look for the pay band before you read the duties. If the company hires in California or New York and the posting has no range, the range exists and they chose not to print it for this role. Ask in the first call.
3. “Remote (San Francisco)” is a sentence, not a location. On Ashby boards it means remote-eligible with a home office; the tax and the occasional on-site week are real. “Remote — US” is the cleaner phrase.
4. Company boards rarely show a closing date; federal postings always do. Treat every company posting as closing tomorrow.
5. Check the domain. job-boards.greenhouse.io, jobs.ashbyhq.com, the company’s own site, or usajobs.gov. Anything else, stop.

Before you apply anywhere

The FTC counted more than 105,000 job-scam reports in 2024, nearly three times the 2020 figure, with reported losses above $513 million. The fastest-growing variety is the “task scam” — a text offering easy money for rating products or liking videos, which then asks you to deposit your own money to unlock earnings: zero reports in 2020, about 5,000 in 2023, about 20,000 in the first half of 2024 alone. AI job seekers are a natural target because the real postings pay so well that a fake one sounds plausible.

What to do with this

• If you are a technician, electrician or IT generalist: open the xAI and OpenAI rows in the no-degree table today. Federal technician tracks at the labs come Monday.
• If you are an engineer or researcher: sort by pay band, not brand. The median bands above are what the postings print; the top of OpenAI’s and xAI’s bands run to $555K–$600K for model training and inference roles.
• If you want a federal role: the NSA development program closes Aug. 26, the NIST AI Standards Coordinator Aug. 31 and the NIST AI-risk researcher role Sep. 2. USAJOBS applications take hours, not minutes; start the resume builder now.
• If you are not sure what fits: ask the desk below. We research every request and post the answer, usually within a day.

If you want to go deeper

• How often will this count update? Every morning. The company-board scan runs daily; the federal and lab edition runs Mondays; entry-level and no-degree roles get Wednesdays; part-time and remote get Thursdays with a scam check.
• Why only twenty companies? Because these twenty publish machine-readable boards we can read without guessing. Microsoft, Google, Amazon, Nvidia and Tesla use systems that do not, so their counts in other articles are estimates. We would rather give you a smaller number you can click.
• Is this a good time to switch into AI work? The honest answer is in two articles we already wrote: when agents beat assistants (what the work is becoming) and what the tools actually cost (what it takes to practice).

Frequently asked questions

Which AI companies are hiring right now?

On Aug. 22, 2026 the twenty company boards we read held 3,645 open roles. By count: OpenAI 754, Anthropic 512, Waymo 366, xAI 259, ElevenLabs 255, Scale AI 208, Sierra 193, Cohere 150, Figure AI 124, Cursor 113. Every one of those numbers comes from the employer’s own job-board feed, and every link in this article goes to the employer’s page.

Are there AI jobs that don’t require a degree?

Yes, and they are concentrated in the physical side of AI. xAI’s data-center operations technician postings in Memphis and Southaven ask for a high-school diploma or equivalency, not a degree. OpenAI’s Site Operations Technician posting accepts an associate’s degree, a technical certification, military experience or equivalent practical experience and prints a $157K–$221K band. Figure AI lists an apprentice robot service technician role in San Jose. The table in this article links to each posting.

How much do entry-level AI jobs pay?

It depends on which door you walk through. A technician path prints $61,200–$91,800 at xAI in Memphis and $157K–$221K at OpenAI in San Francisco. Research and engineering roles at frontier labs print medians of roughly $239K–$356K (OpenAI) and $305K–$405K (Anthropic) on their US postings, but those are not entry-level bands. For the wider market, BLS puts the 2024 median at $112,590 for data scientists and $140,910 for computer and information research scientists.

Are AI companies still hiring, or is it a freeze?

The boards say hiring. 3,645 open roles across twenty companies, with OpenAI alone posting 91 roles in the last seven days that we could date. Where we cannot say “new” honestly — Anthropic, xAI and Waymo refresh every posting’s date nightly — we say so rather than inflate a number.

Which AI companies are hiring remote?

By remote-marked postings: OpenAI 496, ElevenLabs 210, Cohere 139, Replit 70, Anthropic 58, Cursor 50, xAI 45, Perplexity 41. The robot companies (Figure, Apptronik, 1X, Physical Intelligence) list essentially no remote roles — you cannot fix a humanoid over Zoom.

How do I know an AI job posting is legitimate?

Four rules stop almost every scam: never pay for a job, training or equipment; never deposit a check and send money back; treat a job offer by text from a number you never contacted as a scam; and apply only on the employer’s or the agency’s own domain. The FTC counted more than 105,000 job-scam reports in 2024 with over $513 million lost. Every link on this page goes to the employer or to USAJOBS.

Sources

• Company job-board feeds read Aug. 22, 2026: Greenhouse boards (Anthropic, xAI, Scale AI, Figure AI, Apptronik, Agility Robotics, Waymo, Nuro, Google DeepMind) and Ashby boards (OpenAI, Mercor, Perplexity, Cohere, Sierra, ElevenLabs, Cursor, Cognition, Replit, 1X, Physical Intelligence). Counts, locations, dates and pay ranges are as published by each employer.
• USAJOBS search, “artificial intelligence,” sorted by open date, read Aug. 22, 2026.
• Department of Energy, Jobs at the National Labs.
• BLS Occupational Outlook Handbook: Computer and Information Research Scientists; Data Scientists (May 2024 wages, 2024–34 projections).
• FTC, Dec. 2024: new data on game-like online job scams.
• Wikimedia Commons photographs as credited in each caption; Caillebotte painting public domain.

Keep reading

• National laboratory jobs, no degree: what all 126 Oak Ridge postings actually say
• When to use agentic AI, and when not to: a decision matrix built on measured runs
• The real monthly cost of AI, by the job you actually do
• AI scams in 2026: the safe-word kit and the first 60 minutes
• The Physical AI toolbox: the free software the robot companies train on