Dylan Goldblatt & Laila Harrison
94% of higher ed professionals use AI for work ¹
86% of students globally use AI for academics ²
66% of institutions leveraging AI — up from 49% last year ³
¹ UPCEA/EDUCAUSE 2026 · ² The Conversation / Digital Education Council · ³ Ellucian 3rd Annual AI Survey
92% of institutions have articulated an AI strategy ¹
13% evaluate ROI ¹
20% have formal AI policy ²
10% have governance to support effective use ³
¹ UPCEA/EDUCAUSE 2026 · ² Engageli · ³ AI and Education Studio / Medium
66% of business leaders won't hire graduates who lack AI literacy ¹
70% of job skills will change by 2030 ¹
Workers with AI skills command a 56% wage premium ²
¹ AACSB/Microsoft via Open Access Government · ² Globe and Mail / KPMG
AI in education: $9.58B today → $136.79B by 2035 ¹
34.52% compound annual growth rate ¹
OECD education systems: $310–496B in potential savings from AI ²
¹ Yahoo Finance / Industry Analysis · ² Journal de Québec / TVA Nouvelles
95% of AI pilots fail ¹
25% of AI initiatives hit expected ROI ¹
66% of orgs report productivity gains ²
29% of executives can measure AI ROI ¹
¹ IBM / MIT · ² Deloitte State of AI in the Enterprise
Seamless model work at any scale, edge to cloud
One unified, governed data lakehouse for machine intelligence
Self-service research compute and inference
A consistent open-source stack for open science
Office of Research Agent on Sharepoint
Office of Research Chrome Extension on Cayuse
vHPC cluster: 15 petaFLOPS
High-availability flash storage (Ceph)
400 Gbit/s networking (NSF CC*)
AI Garages: portable NVIDIA GPU, on-prem, on-demand
Inference APIs: guardrailed, secure, metered, BYOK
Serverless GPU on demand
Recommendation: decommission CPU clusters
portkey.ai
keys.theparley.org
KSU AI Policy
KSU AI Guidance Document
Public Goods
Build People
certify leaders and teams
Buy Use Time
rent GPUs; only build on-prem with >70% average utilization
Build Software
pursue wins where ROI ≤ 36 months; AI changes the SaaS moat
tokens-per-second-per-watt
median user token burn per day
utilized high-bandwidth memory (HBM) and supercomputer power (PFLOPS)
Research workflows that accumulate capability over time
Each refinement cycle compounds across the group
Readiness = Transfer × Velocity
ngoldbla@kennesaw.edu
lharr260@kennesaw.edu