Training Industry Report: Real Spend per Learner Fell 31% in Five Years

US training expenditures reached $102.8 billion in Training magazine's 2025 Training Industry Report, the highest total in the five editions we tracked. Adjust that for inflation and the picture reverses. Workforce training spend per learner fell from $1,272 in 2025 dollars to $874, a 31% real decline since 2021, while training hours fell from nearly 64 to 40. The one line that grew fast was money paid to outside vendors.

What the Training Industry Report measures, and how it is built

Training magazine has published its Training Industry Report every year for more than four decades. The 2025 page describes the study as being in its 44th year. It is the most widely quoted source for US corporate training statistics, and it is a survey, not a census.

The method is stated on each edition's page. An outside research firm emails members of the Training magazine database between April and July, and they complete an online survey. Only US corporations and educational institutions with 100 or more employees are included. The research firm is not named.

That method shapes every number below. The respondents are people who subscribe to a training trade magazine, which means they are disproportionately the organisations that run a formal learning function. Your company may sit well outside that group.

How the survey becomes a national total

The headline dollar figure is not anyone's reported spend. It is an average training budget, weighted by company size and industry, then multiplied across a universe of US companies drawn from a Dun & Bradstreet database. The 2025 edition states the universe was 152,572 companies. The 2024 edition used 142,105.

This detail matters more than any other in the report. A change in the company universe moves the national total even if no single company changes its budget. We return to it in the section on weaknesses, because it changes how the 2025 increase should be read.

Five editions of workforce training spend in one table

The table below collects the core series from the 2021 to 2025 editions, each taken from the edition's own public page. Every edition reports on the year it is published, so "2025" means the survey run from April to July 2025. Where the public page does not state the sample size, the table says so rather than filling the gap.

All figures nominal, as published by Training magazine. Percentage changes are the report's own year-on-year figures. Sample sizes are as stated in the text of each public page; where a count appears only in an image or not at all, it is marked not stated.

Read across the rows and two things stand out. The total is roughly flat at around $100 billion for four years. The per-learner figure and the hours figure both fall almost every year. A flat total with falling per-learner spend means the money is being spread across more learners, or the count of learners is being measured differently. The public pages do not say which.

The hours series is the clearest signal in the whole report, and the one most often skipped in coverage of it. Training hours per learner fell in every edition after 2021: nearly 64, then 62.4, 57, 47 and 40. That is a drop of about 37% in four years, with no year of recovery.

Adjusted for inflation, training spend per learner fell 31%

None of the five editions adjusts for inflation, and almost no one who quotes them does either. Between June 2021 and June 2025 the US consumer price index rose about 18.7%, according to the Bureau of Labor Statistics CPI-U series. We used June because it sits in the middle of each April to July survey window.

Restated in 2025 dollars, total training spend was about $109.6 billion in 2021, $110.6 billion in 2022, $107.6 billion in 2023, $100.6 billion in 2024 and $102.8 billion in 2025. The record nominal total is, in real terms, about 6% below the 2021 level and about 7% below the 2022 peak.

Spend per learner falls further. In 2025 dollars it went from about $1,272 in 2021 to $1,314 in 2022, then $1,009, $795 and $874. The nominal decline of 18% from 2021 becomes a real decline of 31%. That is the honest version of the headline that per-learner spend "rose" to $874 in 2025.

Hours fell faster than dollars

The hours series needs no inflation adjustment, and it is the stronger evidence. Average training hours per learner fell from nearly 64 in 2021 to 40 in 2025. The 2025 edition reports the drop from 47 to 40 in a single line, without comment on cause.

My reading is that this is the number L&D leaders should be quoting in budget meetings, not the $102.8 billion total. Hours are what a learner actually experiences. A budget can be restated, reallocated or re-labelled. An hour of instruction either happened or it did not.

Cost per training hour jumped in 2025

Divide spend per learner by hours per learner and you get a rough cost per training hour. It is a derived figure, because the two averages may not be weighted identically, so treat it as directional. It sat close to $16 to $17 an hour in 2021, 2023 and 2024, touched about $19 in 2022, and jumped to about $22 in 2025.

That is a rise of about 33% in one year. It is consistent with organisations buying more expensive content or platforms and delivering fewer hours through them. It is also consistent with a measurement change. The data cannot separate the two, and I would not build a budget case on this ratio alone.

Where the training budget moved: outside vendors nearly doubled

Training magazine splits the national total into three buckets. Training payroll is the cost of internal L&D staff. Outside products and services covers vendors, content libraries, platforms and contractors. Other expenditures covers travel, facilities and equipment.

Two of the three buckets are flat or falling across five years. The third almost doubled. Outside products and services went from $8.1 billion in 2021 to $16 billion in 2025, while the total grew by $10.5 billion in nominal terms. About three quarters of all nominal growth in workforce training spend went to vendors.

Payroll and facilities tell the other half

Training payroll was $68.7 billion in 2021 and $64.7 billion in 2025. The 2022 and 2023 editions report the direction of payroll but not a national dollar figure, so we derived it as the total minus the other two buckets: about $65.1 billion and $63.0 billion. The 2024 edition reports $60.6 billion.

Other expenditures doubled from $15.5 billion in 2021 to $28.3 billion in 2022 as in-person training resumed, then fell each year to $22.1 billion in 2025. The broad pattern is clear. Organisations are spending less on their own trainers and rooms, and more on things they buy from vendors.

That shift lines up with the period in which training vendors added AI features to their products. The 2025 edition reports that 29% of the training solution providers it surveyed launched AI-related products. The report does not split vendor spend by product type, so it cannot tell you how much of the $16 billion paid for AI.

How much of the L&D budget is AI training

The question behind this post is whether workforce training budgets grew, or were relabelled as AI. Training magazine gives four AI signals across the five editions, and none of them is a dollar figure. The table sets them side by side.

Usage figures are shares of responding organisations; hours figures are shares of total training hours. The 2025 budget allocation by content area appears only in a chart image on the public page, so the AI share for 2025 is not transcribed here.

Adoption of AI tools quadrupled in two years

The share of organisations reporting AI among their learning technologies went from 9% in 2023 to 25% in 2024 and 37% in 2025. Large companies led. In the 2024 edition, 46% of large companies used AI, against 25% of midsize and 17% of small companies.

Delivery tells a different story. AI delivered 2% of training hours in 2025, up from 0.8%. Classroom instruction held at 28% and online or computer-based learning at 34%, the same as the year before. AI is being bought and switched on much faster than it is carrying instruction.

The 4% line, and the slot DEI vacated

AI training first appeared as its own budget category in the 2024 edition, at 4% of budget allocation. That is an average share across respondents, not a share of national dollars. Read loosely as a share of the $98 billion total, it would be roughly $3.9 billion. That is an illustration of scale, not a measured figure.

The funding-intent question moved more sharply. In the 2024 edition, 24% of organisations expected diversity, equity and inclusion training to receive more funding. In 2025, the share reporting an increase in DEI funding fell to 3%, and AI training took second place on the list at 25%. Those are answers to different questions, so this is a coincidence of percentages, not proof that money moved from one line to the other. It is the closest thing to a relabelling signal anywhere in the five editions.

Grew or relabelled: the verdict on AI training budgets

On this evidence, workforce training budgets did not grow in real terms, and they were not relabelled as AI either. They were redirected. The shift that the data shows is from internal delivery to purchased products, alongside a steady fall in the hours each learner receives.

The relabelling theory needs AI to account for a large slice of the budget. It does not. A 4% allocation in 2024 and 2% of hours in 2025 cannot explain a five-year story in which real per-learner spend fell by almost a third. Whatever happened to training budgets, AI is a passenger in it so far, not the driver.

The growth theory fails on inflation. A nominal record of $102.8 billion is a real decline from 2021. I think the honest summary for a board is three sentences. Training spend per employee is lower than four years ago. Employees get fewer hours. A bigger share of the money now leaves the building.

That fits a pattern we have documented elsewhere. Public commitments to train people in AI are large and loosely defined, as the audit of the three big vendor AI upskilling pledges found. The budget data here is the other side of that coin: the pledges grew while the per-head spending behind formal training did not.

Where this analysis is weakest

This post leans hard on one survey. Its weaknesses are the post's weaknesses, and some of them are serious enough to change the conclusion at the margin.

Small samples, mostly unstated

Only two of the five public pages state a respondent count: 260 in 2022 and 251 in 2024. Those samples are weighted up to more than 140,000 companies. A shift in which 250 organisations happen to answer can move a national total by billions, and year-to-year changes of a few percent sit well inside any plausible margin of error. The report does not publish one.

The AI hours series shows the problem. It reads 3%, 1%, 0.4%, 0.8% and 2%. A real decline from 3% to 0.4% while AI adoption rose is implausible. The more likely explanation is that a small sample makes small percentages noisy. I would read the AI hours figures as "small" and nothing more precise.

The 2025 increase is partly a bigger universe

This is the finding I would most want a reader to take away. The 2025 universe was 152,572 companies, up 7.4% from 142,105. The national total rose 4.9%. Divide each total by its universe and the average spend per company fell from about $689,600 to about $673,800, a decline of about 2.3%.

The size-class figures agree. Average training budgets fell in 2025 for large companies, from $13.3 million to $11.7 million, for midsize companies from $1.7 million to $1.6 million, and for small companies from $374,207 to $333,305. Every size class spent less on average. The headline "increase" came from counting more companies.

There are also limits to our own adjustment. We deflated with consumer prices, but training is mostly wages and services, which inflate differently. A wage-based deflator would change the size of the real decline, and I have not tested one. The direction, a real fall, holds under any plausible deflator because the nominal per-learner figure itself fell 18%.

Finally, the report does not define a learner consistently on its public pages. If organisations began counting more casual users of a learning platform as learners, per-learner spend and per-learner hours would both fall with no change in actual training. That is a genuine competing explanation, and the published data cannot rule it out.

How to use the report as an L&D budget per employee benchmark

Most people search for these numbers to defend or size a training budget. The report can help, provided you compare like with like and avoid the national total, which tells you nothing about your organisation.

Start with company size, because the spread is wide. In 2025, small companies (100 to 999 employees) spent $1,091 per learner, midsize companies $782 and large companies $468. Small companies spend more per head because fixed costs of content and platforms are shared across fewer people.

Then check hours as well as dollars. A budget that matches the per-learner benchmark while delivering half the hours is buying something more expensive, and you should know what. That pairing is more useful than any single figure, and it is the test I would apply before approving a new learning platform contract.

Use the AI figures with care. If your team plans to fund AI training, the closer question is behavioural: whether the training changes how people work. That is examined in the review of what AI fluency training actually changes. For organisations selling into the EU, the legal floor is set out in the evidence on the AI Act's Article 4 literacy duty.

Finally, keep the training line separate from pay. Some organisations now treat AI skills as a pay question rather than a training question, a shift tracked in the analysis of merit budgets and the AI skills premium. If that is your plan, budget it as compensation, not as L&D. And when a vendor shows you an L&D survey, apply the checks in the guide to reading AI ROI survey methodology first.

Frequently asked questions

Where to start this week

Pull your own last three years of training spend and divide it by headcount and by delivered hours. If you can produce cost per learner and cost per training hour for your organisation, you can compare both against the size-class figures above. If you cannot produce the hours number, that gap is your first finding.

Then ask your two largest learning vendors one question: what share of what you pay them is for AI features, and what usage do those features record. The national data says vendor spend nearly doubled while hours fell. Your contract is the only place you can see which half of that story is yours.

Related reading