The headline finding: there is no AI budget. There is a displacement decision. In the most recent institutional budget data available, 78 percent of US academic library leaders said they were likely to cancel one or more journal packages in the next licensing cycle — and the most frequently cited barrier to adopting AI tools was not cost or ethics but low prioritisation relative to other strategies (50 percent).
Read together, those two figures describe the actual purchasing environment for anything sold into a university in 2026. Every number below is attributed with its source, sample and field dates.
The source, and why it is unusually usable
| Ithaka S+R, US Library Survey 2025: Under Pressure | |
|---|---|
| Published | 14 May 2026 |
| Fieldwork | 3 November – 19 December 2025 |
| Population | Library deans and directors, four-year not-for-profit US institutions |
| Responses | 483 complete, valid |
| Response rate | 36 percent |
| Sample by Carnegie type | Doctoral 183 (37.9%) · Master’s 119 (24.6%) · Special focus 94 (19.5%) · Baccalaureate 75 (15.5%) |
| Sample by sector | Private not-for-profit 255 (52.8%) · Public 216 (44.7%) |
| Series | Seventh iteration; triennial since 2010 |
Two things make this stronger than most figures circulating on this topic. The 36 percent response rate is high for a survey of senior leaders, and the fieldwork is recent — late 2025 rather than the spring 2024 that most AI-in-education evidence still rests on, a gap we set out in what the staff-side AI data actually shows.
Its limits should be stated with equal clarity. It covers US four-year not-for-profit institutions, so applying it to a UK, German, Dutch or Spanish institution is an extrapolation. It surveys library leaders, who hold a large share of institutional content and platform spend but not all of it. And respondents skew 84 percent White, 61 percent women and 44 percent aged 55 or over, with 57 percent in their current role for five years or fewer.
The money is contracting, not expanding
Financial constraint is the survey’s dominant theme, and the cancellation figures give it a number.
- 78 percent said they were likely to cancel one or more journal packages in the next licensing cycle.
- 55 percent reported a high expectation of cancellations — up from 45 percent in 2022 and 47 percent in 2019.
- Public institutions were slightly more likely to anticipate cancellations (57 percent) than private ones (54 percent).
That trend line is the important part. Expected cancellations rose across three cycles spanning six years, which makes this a structural condition rather than a single bad year. A vendor arriving in 2026 with a new subscription is arriving in the middle of a subscription reduction.
What gets funded, and what gets cut
The survey asked leaders to allocate a hypothetical 10 percent increase, and separately to absorb a hypothetical 10 percent reduction. The answers are more informative than a spending total.
| With a 10% increase, invest in… | With a 10% cut, reduce… | ||
|---|---|---|---|
| New or redefined staff positions | 48% | Digital journals and databases | 64% |
| Salary increases | 39% | Print monographs | 39% |
| Digital journals and databases | 34% | Print journals | 33% |
Three readings worth taking to a business case.
People beat platforms. The top two uses of new money are both staff. A pitch that promises to replace staff effort is arguing against the stated preference; one that makes existing staff effort go further is arguing with it.
Digital subscriptions appear on both lists. They are the third-largest growth priority and the largest cut target. That is not inconsistency — it is a category under active review in both directions, which means it is the most contestable line in the budget and the one where a new entrant is most likely to be assessed against an incumbent.
Print is no longer the easy cut. Reductions in print monographs and print journals fell to 39 and 33 percent from 54 and 45 percent in 2022. The traditional shock absorber has been substantially used up, which is precisely why digital subscriptions have become the cut target.
One more figure belongs here: 83 percent said maintaining access to subscription-based digital resources is a priority in their library, while only 41 percent agreed their library will become increasingly dependent on externally-provided electronic resources — down from 63 percent in 2022 and 57 percent in 2019. Institutions want to keep what they have and are actively resisting deeper external dependence.
Why AI adoption stalls — and it is not the price
Asked to identify the primary barriers to employing AI tools, leaders gave answers that should reshape how these products are pitched and how internal cases are written:
- 50 percent — low prioritisation relative to other library strategies
- 48 percent — limited staff skills in AI
- 46 percent — lack of time or staff capacity
- 46 percent — ethical or moral opposition to AI
The top three are all about attention and capability rather than money. An institution that has not adopted an AI tool has generally not rejected it on price; it has not reached it. That reframes the entire adoption problem: the competitor is not another vendor, it is the twelve things ahead of it on the strategy list.
The fourth is worth stating without euphemism, because vendors routinely pretend it does not exist. Nearly half of respondents cited ethical or moral opposition. That is a real, principled and widely-held position among senior staff, and a business case that does not address it will fail in the room rather than on the page.
What leaders expect AI to do to them
Asked about the most significant impacts within the next three years:
- 83 percent — increased demand for AI literacy instruction
- 74 percent — integrating AI tools into discovery systems
- 55 percent — staffing changes or reskilling needs
- 51 percent — heightened scrutiny of research integrity
Note what the top answer is. The largest anticipated impact of AI is demand for teaching about it — a service and workload consequence, not a technology purchase. That is consistent with the faculty-side finding that top-down guidance is more wanted than more tools, and it points at the same conclusion: the cheapest high-value intervention available is clear institutional guidance, whose components are in what a university AI policy should include and whose curriculum form is in how to build an AI literacy curriculum.
The procurement number: vendor data access concern is at a record high
One legacy item tracks concern about the extent to which third-party vendors or partners have access to individual-level data from library users. Its trajectory:
- 2019: 40 percent concerned (first measured)
- 2022: 30 percent
- 2025: 50 percent — the highest point recorded
The report links the increase to heightened attention to data privacy, including issues associated with AI use. For anyone selling into this market, or writing a case to buy, this is the single most operationally relevant number in the survey: half of the people evaluating you are actively worried about exactly the thing your product does with student text. Answer it before you are asked — the residency and sub-processor evidence to have ready is set out in where your students’ text is actually processed.
One structural finding for anyone building a case
Only 31 percent of leaders agreed that they are involved in key decision-making processes at campus level, and while most feel confident articulating the library’s value proposition, fewer believe that value is recognised by senior administrators.
That has a practical consequence. A well-evidenced case from a service that is not in the room does not automatically reach the decision. Establish who is in the room before the case is written, and make sure the evidence answers their question rather than the service’s. On what leaders do routinely share upward, utilisation data leads at 77 percent, followed by narrative evidence at 73 percent, user experience data at 66 percent and external recognition such as awards and grants at 64 percent — a reporting mix weighted heavily towards volume and story rather than measured effect, which is a gap worth closing using the method in how to measure whether a writing support intervention actually worked.
What to take from this into a 2026 decision
- Write the displacement explicitly. Name what the new spend replaces. A case that does not will have the question asked for it, less favourably.
- Lead on capacity, not on features. The top barriers are attention, skills and time. A proposal that reduces demands on all three is arguing against the actual obstacle.
- Answer the data question unprompted. Half your audience is already concerned about vendor access to individual-level data, and that share has risen sharply.
- Budget for the literacy demand, not just the licence. 83 percent expect increased demand for AI literacy instruction. That workload arrives whether or not you buy anything.
- Test with a pilot rather than a purchase. Where the barrier is prioritisation rather than price, the winning move is the one that consumes no budget and no governance cycle — the design is in how to run a departmental pilot.
If you would like to scope an evaluation that produces its own local evidence rather than borrowing sector figures, request an institutional evaluation.
Frequently asked questions
Are university budgets for AI tools growing?
The available evidence says budgets are under sustained constraint. Financial resources were the most widely cited barrier to change, and 78 percent of leaders expected to cancel one or more journal packages in the next licensing cycle.
How reliable is this data?
It is among the more reliable figures in this area: 483 valid responses, a 36 percent response rate, and fieldwork from 3 November to 19 December 2025. It is limited to four-year not-for-profit US institutions.
What is the biggest barrier to AI adoption?
Low prioritisation relative to other strategies, at 50 percent — ahead of limited staff AI skills (48 percent), lack of time or capacity (46 percent) and ethical or moral opposition (46 percent).
Is cost the reason institutions have not adopted AI tools?
Not primarily. The top three barriers concern attention and capability. Cost enters through the displacement question rather than as a stated objection.
Where would new money go?
To people. New or redefined staff positions (48 percent) and salary increases (39 percent) lead, with digital journals and databases third at 34 percent.
What gets cut first?
Digital journals and databases, at 64 percent — the same category that ranks third for investment.
Is print still the easy saving?
Less so. Print monograph and print journal reductions fell to 39 and 33 percent, from 54 and 45 percent in 2022.
What impact do leaders expect from AI?
Chiefly a teaching-demand impact: 83 percent expect increased demand for AI literacy instruction, ahead of discovery-system integration (74 percent), reskilling (55 percent) and research integrity scrutiny (51 percent).
How concerned are institutions about vendor data access?
Fifty percent expressed concern about third-party access to individual-level user data in 2025 — the highest recorded, against 30 percent in 2022 and 40 percent in 2019.
Do these figures apply outside the United States?
Not directly. The population is four-year not-for-profit US institutions, and any application elsewhere is an extrapolation that should be labelled as one.
Does this cover total institutional edtech spend?
No. It covers library leaders, who control a significant share of content and platform spend but not all institutional technology budgets. Treat it as a well-measured slice rather than a sector total.
What is the most useful single number here?
The 50 percent barrier figure for low prioritisation. It says that most institutions have not decided against these tools — they have not yet reached them.