{"resourceId":"virginia-tech-hokieai-usage-token-costs-wdbj-20261022","versions":[{"version":"external-2f1dfd174539173c3fc31b570154ef23d73db6a22ae68e9bc2d5c68e0e37b594","resource":{"id":"virginia-tech-hokieai-usage-token-costs-wdbj-20261022","title":"Virginia Tech's HokieAI: about 10,000 active users of 75,000 eligible, roughly $40,000 in token spend since August and a $15 monthly cap per user","organization":"Virginia Tech","sector":"Public research university: campus-wide AI platform","geography":"Blacksburg, Virginia, United States","publishedAt":"22 October 2026, 6:20 PM EDT (WDBJ7)","publicationDate":"2026-10-22","eventDate":"2026-10-22","sourceName":"WDBJ7 (Gray Media)","sourceLabel":"Local television news report","sourceUrl":"https://www.wdbj7.com/2026/10/22/virginia-techs-hokieai-platform-draws-mixed-reactions-campus/","evidenceClass":"independent-reporting","outcomeClass":"mixed","topics":["knowledge-work","governance-procurement","operating-model","accessibility-workforce"],"finding":"Two months after production rollout, about 10,000 of roughly 75,000 eligible students and faculty actively use HokieAI, Virginia Tech's own platform offering ChatGPT and Claude. The NebulaONE license and support cost about $275,000 over two years, but officials call token usage the largest cost driver; it has totaled about $40,000 since August. Most users are capped at $15 a month in token spend and use only a fraction. Officials cite a usage agreement under which data never trains models. Student reactions are mixed, with one citing lack of training and a sense of not being anonymous.","sledRelevance":"One of the few public data points on actual uptake and consumption cost for a campus-wide, metered multi-model platform, useful for sizing budgets and adoption programs. This is new reporting on a platform already in the archive from its August launch.","evidence":"Reporter's account of officials' figures plus student interviews. Figures are not independently audited; active-use definition and measurement period are not given.","architectureImplications":"University-operated platform on NebulaONE brokering commercial models, with per-user token metering.","governanceImplications":"Usage agreement bars model training on user data; per-user spending caps are the main cost control.","securityPrivacyImplications":"Officials claim full custody of prompts; a student's sense of not being anonymous signals that institutional custody can itself deter use.","caveats":"Two-month snapshot; token cost will scale with use. Figures come from officials via a news report.","streamIds":["campus-operations"],"roles":{"sales":"Sales framing: Virginia Tech gives a rare look at real uptake and cost: about 10,000 active users of 75,000 eligible, and token spend of about $40,000 since August against a $275,000 two-year platform license. Interpretation: institutions buying platforms may need help forecasting variable token costs and lifting adoption beyond early users. Offer a bounded adoption and cost-forecast review. Do not cite this as proof of value; no outcomes are reported.","engineering":"Engineering framing: HokieAI runs on NebulaONE with per-user monthly token caps, so engineering choices center on metering, model routing and data custody. Interpretation: when replicating, confirm the platform reports spend by user and department, supports differentiated caps, and can route routine prompts to cheaper models. The report does not describe hosting, logging or retention, and a student's comment about anonymity suggests telemetry practices should be documented for users.","delivery":"Delivery framing: access is broad but active use is about one in seven eligible users after two months. Interpretation: delivery should track active use by role, pair access with training, and publish plainly what is logged and who can see it, since one student cited both lack of training and not feeling anonymous. Review caps against actual consumption each term so cost control does not quietly limit heavy legitimate users."},"roleDetails":{"sales":{"customerProblem":"Campus platform leaders need to forecast variable token costs and raise adoption beyond a minority of eligible users.","stakeholders":["CIO and central IT","Provost's office for teaching and learning","Budget office","Student government"],"discoveryQuestions":["What share of eligible users are active each month?","How is token spend tracked by unit?","When would caps need to rise, and who pays?","What training reaches non-users?"],"valueHypothesis":"Interpretation: a usage and cost forecast could help size budgets and caps; the report shows no outcome or savings data.","potentialEngagement":"A bounded adoption and token-cost forecasting review for one campus AI platform.","unsupportedClaims":["Productivity or learning gains","That low token spend will persist as use grows","That a $15 cap suits staff workloads","Guaranteed savings"]},"engineering":{"fit":"Reference for a metered, university-operated multi-model platform.","architectureIntegration":"NebulaONE brokers ChatGPT and Claude models with per-user token spending limits.","prerequisites":["Per-user and per-unit spend reporting","Identity integration","Model routing policy","Telemetry disclosure to users"],"constraints":["Hosting and retention details not reported","Caps may constrain heavy users","Costs scale with token prices and model choice"],"security":"Officials claim data never trains models and the university retains custody; logging visibility to administrators is not described.","proofOfValue":["Reconcile reported token spend with invoices","Test cap enforcement and overage handling","Document what usage data administrators can view"]},"delivery":{"work":"Run adoption and cost operations for a campus-wide metered AI platform.","dependencies":["Spend telemetry","Training capacity","Communications on data handling"],"ownership":"Central IT and the teaching and learning office jointly, with budget office input on caps.","skillsAdoption":"Training gaps were cited by a student; targeted training for inactive groups is needed.","governanceCheckpoints":["Termly review of active-use share","Cap review against consumption","Published statement on logging and privacy"],"acceptanceCriteria":["Proposed: active monthly users rise to a defined share of eligible users within two terms","Proposed: monthly spend forecast within 15% of actual","Proposed: plain-language logging notice published"],"risks":["Token costs rising faster than budget","Users avoiding the platform over privacy perceptions","Caps limiting legitimate heavy use"]}},"retrievedAt":"2026-10-22T21:54:36-05:00","enrichedAt":"2026-10-22T21:58:32-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Platform was justified as equalizing access; low active share and reported lack of training suggest adoption support lags access.","procurementImplications":"Platform license (about $275,000 over two years) is smaller than expected variable token spend over time.","operatingModelImplications":"Cost management treated as ongoing; caps may rise as use grows.","sourceVerification":{"openedUrl":"https://www.wdbj7.com/2026/10/22/virginia-techs-hokieai-platform-draws-mixed-reactions-campus/","referenceExcerpt":"About 75,000 students and faculty currently have access to HokieAI, but only around 10,000 are actively using it.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}