How long an AI doctorate takes, and when to reach the people finishing one

What is being measured, and what it leaves out

Enrolment dates are rarely public, and rarely accurate when they are. A first indexed paper carries a date, a name and a venue, and anyone can check it. That is what the clock starts on here.

The consequence is that this gap runs shorter than the doctorate itself. Most people publish something a year or two in, so add a year at the front if you want the full length of the degree. For working out when to send a message, the shorter figure is the useful one, because the first paper is also the moment the person becomes visible to you at all.

Gaps above ten years are dropped. A researcher whose first paper is dated 2009 and whose defence is dated 2024 has a bad date on one side or the other, usually an ORCID entry filled in from memory.

Subject by subject

The striking thing here is how little movement there is. Vision, language, theory and applied work all land within a year of each other. Whatever separates these communities, it stops at the laboratory door and leaves the calendar alone.

What does change between subjects is the size of the pipeline. A subject with three thousand measured researchers and one with six hundred produce candidates at very different rates, even though each individual takes the same time to arrive.

When doctorates actually end

The year-by-year distribution below comes from public declarations rather than from an estimate. Someone stated an end date on ORCID, or a thesis deposit carries one. It is a floor rather than a census, because plenty of people never fill that field in, and the recent years are the most complete because ORCID adoption keeps rising.

Read the years ahead of today as commitments rather than facts. A declared 2027 end can slip by two terms, and often does. That works in your favour: someone who tells you in March that they are twelve months out is describing a window that usually opens rather than closes.

What is in the pipeline

67 857 researchers in this index have entered a doctorate on ORCID without a completion date in the past. They are the people arriving on the market over the next few years, and no public list gathers them. The largest single subject is Topic Modeling, with 2 521 doctorates under way.

This table is the one that repays watching over months. A subject climbing here today is a subject where candidates become available in three or four years, which is roughly the planning horizon of a research team that intends to still exist then.

Turning the number into a date

The practical rule is short. Find the first paper, add five years, subtract nine months, and write then. For someone whose first paper appeared in 2022, that points at the first half of 2026.

Two things sharpen it. A declared end date, where one exists, beats the estimate outright, and this index carries those. Publication rhythm helps too: a candidate whose output jumped in the last eighteen months is usually assembling a thesis rather than starting one.

The competing offers set the deadline. Academic hiring runs on a calendar, with letters going out between January and April, so an industry offer arriving in May is answering a question the person already settled. Our guide on finding and hiring AI researchers covers what that offer has to contain once you are in the window.

Common questions

How long does a PhD in AI take?

Measured on this index, between 4 and 5 years separate a researcher's first indexed paper from the end of their doctorate, and that range holds in every subject we measure. The spread matters more than the median for hiring: a first paper dated three years ago tells you someone is probably mid-doctorate rather than close to done.

Does the subject change how long a doctorate takes?

Barely. Across the 25 subjects with enough people to measure, the medians run from 4 to 5 years. Vision, language and method work land in the same place. What changes between subjects is the number of people in the pipeline, not the time they spend in it.

How many AI researchers finish a doctorate each year?

This index dates the end of 9 998 doctorates from public sources, with 899 falling in 2026 and 301 already declared for 2027. Those counts cover people who stated a completion date or whose thesis deposit carries one, so they are a floor rather than the whole cohort.

When should I contact someone who is finishing a doctorate?

Six to nine months before the defence. Earlier and the honest answer from them is that it is too soon to say. Later and the postdoc letter or the industry offer is signed. Competing academic offers go out between January and April, so the useful months for a 2027 finisher run from roughly March to July 2026.

Can I be told when someone in my field is about to finish?

That is what the alerts do. Describe the subject in your own words, and the index tells you when a researcher working on it has a doctorate ending in the months ahead, before the search becomes crowded. Access is by request.

Why measure from the first paper rather than from enrolment?

Enrolment dates are rarely public and rarely accurate when they are. A first indexed paper is dated, attributable and verifiable by anyone. It arrives a year or two into a doctorate in most fields, which makes the gap we measure shorter than the doctorate itself and more useful for timing a message.

How many doctorates are in progress right now?

67 857 researchers in this index have entered a doctorate on ORCID with no completion date in the past. They are the supply arriving over the next few years, and no public list gathers them.