How many jobs has AI created in 2026?

Labor · LinkedIn · 2026

There is no official global census of “jobs created by AI.” There are hiring graphs, title inventiveness, and a lot of slides. The number that survived contact with a real dataset this year is LinkedIn’s, not a think-tank round number.

In January 2026 LinkedIn told Davos that 1.3 million new AI-enabled jobs had shown up globally over two years — AI engineers, forward-deployed engineers, data annotators — and that more than 600,000 AI-enabled data-center jobs appeared on the platform in the previous year alone. That is a platform count, not an ILO census. It is still the cleanest public figure we have.

What “created” actually means

LinkedIn is counting new roles that would not exist in that shape without the model boom: people who train, eval, deploy, annotate, or keep a cluster alive. It is not counting every marketer who typed “AI” into a job description. It is also not netting out the junior copywriter who was not replaced so much as never hired.

The lab headcount is still small in world-labor terms — tens of thousands at the named model companies, not millions. The bodies sit around the labs: construction, high-voltage, networking, evaluation, vendor management, and the unglamorous work of putting a model behind a ticket system. That is why the data-center number can be huge while “OpenAI employee count” stays a rounding error in a G7 labor force.

The forecast people quote, and should label as a forecast

The World Economic Forum’s Future of Jobs work still gets compressed into a bumper sticker: on the order of 170 million roles created and 92 million displaced by 2030, a net of about 78 million. That is a 2025-era employer-survey forecast to the end of the decade. It is not a 2026 headcount. Treat it as a direction: more rotation than apocalypse, with the pain loaded onto tasks that look like call-center scripts and first-draft copy.

Call-center and junior copy roles did not vanish in a year. They stopped growing. The hiring went to QA, vendor management, and whoever owns the failure cases.

Where the seats actually opened

AI Engineer is LinkedIn’s #1 US role for a second year. Head of AI titles rose in double digits in a single year at companies in Australia (32%), Canada (31%), India (30%), Germany (30%), the UK (30%), and the US (28%). That is not “everyone hired a researcher.” That is companies admitting the model is now a line of business.

Forward-deployed engineers — sit with the buyer, ship the retrieval, eat the outage — are the difference between a logo on a slide and revenue. Data annotators did not disappear because models got smarter; eval sets got larger. Prompt engineer as a standalone title peaked and was absorbed into whoever writes the product.

Displacement without a funeral

The honest pain is at the bottom of knowledge work. Companies froze graduate intakes in customer support and marketing production, then hired a smaller number of people who can run an eval harness or tune a deflection agent. PwC’s AI Jobs Barometer has, for years, shown a wage premium on the order of 25% for roles that list AI skills versus neighbors that do not. That premium is a price signal: the work changed shape faster than the org chart.

If you need a sentence for a slide: high-single-digit millions of workers worldwide now list AI as a core skill; the new-net creation we can count is in the buildout and in a relatively small set of titles; anyone quoting a precise global total for “jobs AI created in 2026” is rounding LinkedIn or inventing a WEF year that has not happened yet.

Sources

LinkedIn, A New World of Work (14 Jan 2026). World Economic Forum write-up of the same LinkedIn series, 15 Jan 2026. WEF Future of Jobs 2025 (2030 forecast, not a 2026 count). PwC AI Jobs Barometer (wage premium). US BLS computer-and-information research scientists outlook (26% projected 2023–2033) is a related occupation, not a synonym for “AI engineer.”

Labs are small. The buildout is not.

Named model companies still employ on the order of tens of thousands of people, not millions. That fact gets used as a gotcha (“see, AI creates no jobs”) and as a boast (“see, a tiny staff runs the economy now”). Both are lazy. LinkedIn’s 600,000 data-center figure is the tell: electricians, cooling techs, network engineers, security, and the project managers who live in a trailer next to a substation. Those are AI-enabled jobs in the same way 19th-century rail jobs were “steam-enabled.” They do not require a transformer paper.

The geographic lumpiness is the political problem. A county that gets the campus gets overtime and a strained housing market. A city that used to hire 200 graduates into a call center gets a hiring freeze and a vendor slide. Both can be true in the same quarter. National “net jobs” averages wash that out, which is why the average is a bad communication tool and a fine spreadsheet cell.

Early-career is the bruise

The 2024–25 freeze on junior copy, research-assistant, and first-line support roles is the part of the story that is not a LinkedIn press release. Tasks that look like “take this pile and make a first draft” got cheaper. Companies did not always fire the people doing them. They stopped backfilling. A 22-year-old who can already run an eval set and write a retrieval eval is being hired; a 22-year-old who can only do the first draft is waiting. That is a skills split inside a cohort, not a verdict on a generation.

If you run a university careers office: stop sending students to “prompt engineer bootcamps” as a plan. Send them to measurement, domain knowledge, and one production stack. If you run a company: a freeze on juniors plus a slide about “AI creating jobs” is a reputation problem you will pay for in 2028.

How to quote this without lying

• Use 1.3 million AI-enabled roles / two years and attribute LinkedIn, January 2026.
• Use 600,000+ data-center jobs / one year the same way.
• Use WEF’s 170 million / 92 million only with the year 2030 and the word forecast.
• Do not add those numbers together. They are different objects.
• Do not convert “AI-enabled” into “created by ChatGPT.” A crane operator at a campus is in the first set and not the second.