June 21, 2026

How University Libraries Are Evolving in the Digital Age

University librarian using AI-assisted research tools at a modern reference desk

Walk into the main library at most research universities today and the scene is genuinely strange if you grew up with the traditional model. There are still books. But there's also a podcast studio, a 3D printer, and an AI chatbot fielding reference questions at 2 a.m. when the desk is dark. The old stereotype of libraries as quiet warehouses for print knowledge is dissolving—and what's replacing it is messier, more contested, and far more interesting than the breathless tech coverage usually suggests.

The AI Experiment (and the 35% Problem)

According to Clarivate's 2025 Pulse of the Library survey—which covered more than 2,000 librarians across 109 countries—67% of libraries are now exploring or implementing AI tools, up from 63% in 2024. The growth sounds decisive. But dig into the regional breakdown and a more complicated picture emerges.

Asia and Europe are moving fast, with 37–40% of libraries past the evaluation stage and into actual implementation. The U.S. lags well behind: just 7% of American library workers reported feeling optimistic about AI, compared to 27–31% in Asia. That's not simple technophobia. It's skepticism rooted in experience with vendor tools that have overpromised.

One number explains much of that skepticism. The Library of Congress tested five machine learning models on roughly 23,000 ebooks and found that subject classification hit only 35% accuracy with the Annif model. Transformer-based models handled author names and titles adequately. But meaning—the actual intellectual work of organizing knowledge—remains slippery for machines.

King's College London's "KingbotGPT" (built on retrieval-augmented generation) and San Jose State University's custom reference chatbot show where AI works: routine queries about hours, locations, and policies. "Is the library open Saturday?" is solved. "What's the best database for mid-20th century African economic history?" still needs a human.

What AI handles well: routine reference queries, basic metadata generation, personalized reading recommendations
Where librarians still win: nuanced subject classification, complex research support, ethics review of AI outputs
The working model: "human-in-the-loop" workflows where AI handles volume and librarians handle judgment

ACRL published its AI Competencies for Academic Library Workers in 2025. But 56% of surveyed librarians say they need significant upskilling to work effectively with AI tools. The institutions moving fastest on adoption are doing so with workforces that, by their own assessment, aren't fully ready.

What's Replacing the Stacks

The stacks are shrinking. That's been true for two decades, but what's growing in their place at research libraries is more varied than most people expect.

Makerspaces have gone from novelty to standard at major research universities. By 2019—the most recent comprehensive count—110 academic libraries across 214 institutions had them. Virginia Tech's Prototyping Studio inside Newman Library spans 2,450 square feet and houses metal 3D printers, CNC machines capable of cutting full 4x8 plywood sheets, vacuum forming gear, and podcast booths. The University of Michigan's Design Lab adds a vintage video game archive and audio/video production studios serving the university's game development programs.

These spaces do something strategically important: they make the library relevant to any student, regardless of major. Engineering department makerspaces often restrict access to STEM students. Library makerspaces almost never do. Anyone with a university ID can show up and make something.

The University of Sheffield built its Digital Commons inside the Information Commons through the "Liberate the Library" project, using participatory design where students and staff co-planned the space (rather than administrators deciding what patrons supposedly needed). That distinction matters more than it sounds—spaces designed with users tend to get used.

The library's competitive advantage is no longer the collection alone. It's the people, the space, and the expertise in knowing which resource to reach for.

Research on library space design confirms what good architects already knew: acoustic comfort and physical comfort directly predict visit frequency and session length. Students stay longer in well-designed spaces. That sounds obvious, but it was systematically ignored in 1970s brutalist library designs now being expensively retrofitted.

The Ebook Pricing War Nobody Warned You About

Here is a slow-moving problem that most people outside academic publishing haven't noticed. And it's genuinely bad.

A single-user ebook license for an academic title regularly costs three times the cover price of the print edition. Not because the content differs, but because publishers price the institutional market separately from consumer markets. Most of these licenses allow only one patron to access the title at a time—functionally similar to print lending, but at a much higher price.

Then, in February 2025, Clarivate—parent company of ProQuest and Ebook Central, the platform at the center of how most research libraries purchase digital content—announced it was phasing out perpetual purchase licenses in favor of subscription-only access. The backlash from the library community was sharp enough that Clarivate extended perpetual purchase options through June 30, 2026. That's not a solution. It's extra time before the same problem.

Access Model	Library Owns Content?	Budget Predictability	Main Risk
Perpetual license	Yes	Medium	Publisher failure
Subscription	No	High	Collection vanishes if budget cut
Evidence-based acquisition	After trial period	Medium	Requires usage infrastructure
Open access flip	N/A	One-time cost	Not yet at scale

The shift to subscription-only matters because a collection built on subscriptions can evaporate when a budget cycle goes wrong. Several open-access alternatives are gaining traction—"Direct to Open" from MIT Press, "Fund to Mission" from Cornell University Press—but they're still demonstrating viability rather than replacing the existing market.

Librarians consistently favor evidence-based acquisition models, which let institutions preview content, collect usage data, and make informed permanent purchases before committing. Demand-driven acquisition, by contrast, has largely been abandoned due to unpredictable spending spikes and high maintenance overhead.

The Budget Squeeze

The financial picture for academic libraries right now is rough. Since 2020, 75% of academic libraries have experienced budget cuts, according to ACRL data, with some exceeding 10%. Library spending per student at four-year institutions has dropped 20%.

UCSF Library's 2025–2026 update makes it concrete: a 15% cut to collections funding—the budget line covering journal subscriptions, databases, and ebooks—at a moment when institutional growth was increasing demand. Federal cuts to NSF, NIH, and IMLS funding compound the pressure on libraries that depended on those grants for digital preservation and data management projects.

The enrollment cliff adds another dimension. Some states are projecting declines of more than 30% in high school graduates, driving predictions of up to 80 college closures between 2025 and 2029. When institutions merge or close, library collections rarely survive intact—and the patrons those collections served lose access with no warning.

One detail that stings: the IPEDS Academic Libraries survey section was officially retired in 2025. That sounds like bureaucratic housekeeping. It eliminated the primary benchmarking tool smaller institutions used to make budget cases to administrators. Less data makes advocacy harder—and administrators know it.

Research Data and Digital Preservation

One area where libraries are genuinely expanding rather than defending is research data management. The old model was consultative: a graduate student shows up with questions about a data management plan, a librarian helps them fill out the form.

The emerging model is different. Libraries are embedding into research workflows from the start—advising on metadata standards, helping design data documentation, connecting researchers to appropriate repositories before data collection begins. The NIH's Generalist Repository Ecosystem Initiative (GREI), which standardizes institutional connections to repositories like Zenodo and Figshare, has pushed libraries into this earlier, more active role.

2025 produced an unexpected example of library data work. As concern mounted over at-risk federal government datasets, data librarians organized the Data Rescue Project—a grassroots effort to identify and preserve data before potential removal from public servers. Nobody had that in a strategic plan three years ago.

Digital preservation carries a quieter crisis underneath it: technical debt in specialized collections. Formats go obsolete and software dependencies break, and no one wants to pay for maintenance. The total cost of technical debt in the U.S. runs to an estimated $1.52 trillion annually across industries. Academic libraries with proprietary file formats and aging digital archives are part of that figure.

The Two Unglamorous Priorities

Nobody builds a library career dreaming about cybersecurity audits. But after ransomware attacks hit libraries in Seattle, Toronto, and across Michigan in 2024–2025, the threat is concrete. Generative AI has made social engineering attacks cheaper and more effective—sophisticated phishing, voice cloning, and deepfakes no longer require expensive production. Libraries sit at intersections of student data, financial systems, and research repositories, which makes them attractive targets.

Accessibility is the other underdiscussed priority. A 2025 audit of 100 university library websites found 80% contained accessibility errors—missing alt text, keyboard navigation failures, color contrast problems. The DOJ's 2026 final rule requires WCAG 2.1 Level AA compliance for all public-entity digital content. For institutions serving populations over 50,000, the deadline is April 2026.

Both priorities share a pattern: they're infrastructure work, invisible when done right and catastrophic when ignored. The library director who spends capital on these gets no ribbon-cutting ceremony. But they also don't appear in a local newspaper after a ransomware attack locks students out of course reserves during finals week.

What's Worth Being Optimistic About

I want to end on something other than a crisis inventory, because the picture isn't all bleak.

The library's self-concept is genuinely expanding in useful ways. The shift from "we hold the collection" to "we build this institution's research capacity" is more than a rebrand. It's a different value proposition that's harder to cut and easier to defend in budget hearings. Libraries that have made this shift are finding new relevance even inside constrained budgets.

Open access is moving—if slower than advocates hoped. The flip-to-open models are imperfect, but they represent a structural shift away from libraries paying ever-escalating subscription costs for content their own faculty created. The writing is on the wall for the current journal pricing model.

And the AI conversation, despite warranted skepticism, is pushing libraries to articulate something important: human expertise in information curation is not a legacy cost to be automated away. It's the thing that catches what a machine classifies wrong 65% of the time. That argument, made clearly, is a stronger case for libraries than any claim about collection size.

Bottom Line

Ebook perpetual licensing is changing: Clarivate's extension expires June 30, 2026. Libraries without a clear position on subscription-versus-ownership are already behind.
AI in libraries works at the margins right now—routine reference, basic metadata, patron recommendations. It still fails at nuanced subject classification. Invest in staff AI literacy before tool adoption.
Physical spaces are assets, not overhead. Makerspaces and learning commons drive usage in ways digital portals alone don't. Space renovation budgets are justified by measurable changes in patron behavior.
Research data management has moved from optional consultation to core service. If your library isn't embedded in grant proposal workflows, it's missing where the need actually lives.
The 35% accuracy ceiling on library AI subject classification is, I'd argue, the single most important number in this conversation. It's the evidence that human judgment in information systems isn't nostalgia. It's necessity.

Frequently Asked Questions

Are physical university libraries being replaced by digital ones?

No—and the renovation activity argues against it. Makerspaces, learning commons, and collaborative studios are replacing traditional stacks, not the buildings that house them. Research consistently shows that acoustic comfort and physical design directly affect how frequently students visit and how long they stay. The function is changing; the space is not disappearing.

Why do ebooks cost so much more for libraries than for individual buyers?

Publishers price the institutional market separately from consumer markets, often charging three to five times cover price for single-user library access. The rationale is that libraries enable multiple uses from one purchase. Many contracts allow only one patron at a time—functionally similar to print lending, but at a dramatically higher cost and without the ownership rights that print provides.

Is AI actually replacing reference librarians?

Not in any near-term timeframe. Current AI tools handle routine queries well but hit a ceiling fast—Library of Congress testing found subject classification accuracy topped out at 35% with leading models. The practical outcome is hybrid workflows where AI handles volume and librarians handle judgment, research consultations, and quality control. It's augmentation, not replacement, and the gap in classification accuracy explains why.

What is a library makerspace and who can actually use it?

A library makerspace is a hands-on space with tools for prototyping and creating—3D printers, laser cutters, sewing machines, audio/video production gear. Unlike department-based makerspaces that restrict access to specific majors, library makerspaces are typically open to all students, faculty, and staff regardless of discipline. Most offer free equipment use; materials costs vary by institution.

How are university libraries responding to budget cuts?

Through a mix of canceling high-cost journal subscriptions, shifting to evidence-based ebook acquisition models, pursuing open-access publishing partnerships, and consolidating services. Some libraries are cutting liaison librarian positions to preserve collections budgets. A few, like UCSF, have published detailed budget impact statements that name specific databases and services being cut—a transparency approach that creates external pressure for restoration.

What is the Data Rescue Project and why does it matter for libraries?

The Data Rescue Project emerged in 2025 as data librarians organized to identify and preserve federal government datasets perceived to be at risk of removal. It's significant because it illustrates how academic libraries have extended their preservation role beyond their own collections to public information more broadly. It also reflects the grassroots capacity within the library profession to mobilize quickly around a shared mission when institutions are slow to act.