This is a question which preoccupies me at the moment as higher education signs up to enterprise AI deals, with costly implementation work packages, at the most exuberant point in the investment cycle. This paper provides a deeply relevant case study which helps us understand what this can look like in practice:
According to company representatives, roughly 80% of US institutions rely
on Google Workspace (Bonk, 2023), and more than 150 million students and teachers worldwide use Google cloud products for education (Dey, 2025). For nearly a decade, Google Workspace for Education offered members from research universities what was described as ‘unlimited storage’ (Bishop, 2014). Offering unlimited cloud storage to campus-wide organizations cannot be overestimated in the context of platform-driven work. By enabling seemingly frictionless infrastructure integration for thousands of users through their email accounts, Google asserts platform power by shaping collaboration and normalizing habitual use of their app ecosystem (Popiel & Vasudevan, 2024).The COVID-19 pandemic intensified the use of Google Workspace apps and learning products across many education sectors. For example, in 2020–2021, Google Classroom (a version of Google Workspace tailored for K-12 schools) usage grew from 40 million to over 150 million users (Perez, 2021). Reaching saturation and dominating the education space with learning tools created new thresholds of usage that resulted in new pricing and product opportunities for the platform. Google explained in its February 2021 policy update that it was discontinuing unlimited storage, ‘as we’ve grown to serve more schools and universities each year, storage consumption has also rapidly accelerated’ (Sinha, 2021).
The next year, Google introduced a pooled storage model capping institutions at 100TB, replacing unlimited capacity and rolling back storage products like Shared Drives (Google Workspace Admin, 2025). Universities were then forced to roll back storage, ration tiers of access (across faculty, students, staff, and alumni), and retrofit research data management policies to comply with superseding federal research data access mandates (Dodge, 2023; Hanchon, 2023; Shanbom, 2023; Walker, 2022). As many campus journalists and IT offices reported, graduate students with terabytes of lab data and faculty coordinating multi-university projects were suddenly facing new barriers to basic workflows that involved considering the resource limits of data manage- ment and cloud storage for research teams (Carnegie Mellon University, 2025; Stanford, 2024; University of Washington, 2023).
If we reduce headcount and reorganise processes based on the perceived affordance of enterprise AI, the dependence this embeds will be uniquely pernicious. The sector’s in a bad enough place financially that I would (reluctantly) accept the logic if (a) efficiency savings were as straightforward as a methodologically deficient literature suggests (b) The conditions of access to the technological capacity were commercially stable. But given neither of those conditions are true enterprise AI in 2026 seems like an obviously bad idea which we are going to regret as a sector.
