Why infrastructure decides whether CBE and AI are real
I coordinate e-learning work and teach in the Faculty of Computing and Informatics at Makerere University Business School (MUBS). When NCHE talks about Competence-Based Education (CBE) and when campus conversations turn to Artificial Intelligence (AI), I look at the lab first. Policy can demand observed competence. A lab with eight working PCs for forty students, a projector that dies mid-demo, and a power cut at 11 a.m. will quietly rewrite that policy into attendance theatre.
From the 2027/2028 academic year, higher education institutions in Uganda are expected to admit first-year students only to CBE-aligned programmes. NCHE set that direction in its January 2026 consultative meeting on CBE reforms. Assessment, labs, and faculty support sit inside that reform. Related: The 2027 NCHE mandate for computing faculties. This post is a practical infrastructure checklist for a computing faculty, grounded in constraints I see on campus and in public evidence about Uganda’s research and education networks.
What “infrastructure gap” means on a Tuesday morning
In consulting and teaching support work, the gap is not one missing GPU rack. It is a stack of ordinary failures:
Labs: machines that take ten minutes to boot, missing RAM, shared logins, no standard image for hardening exercises.
Power: outages that erase a timed practical; UPS units that cover servers but not student benches.
Bandwidth: classroom Wi-Fi that collapses when a whole cohort opens the LMS; uneven last-mile quality between campuses.
LMS: platforms that work for file drops but fail when everyone uploads video evidence in the same hour; weak offline fallback.
Peer-reviewed work on Ugandan higher education documents the same pattern at national scale. A 2025 TOJET study comparing Makerere University and Kampala International University reports disparities in labs, connectivity, devices, and engagement with national ICT agencies, and notes power interruptions and outdated machines as lived barriers in under-resourced settings: The Role of RENU and NITA-U in Providing Infrastructure and Online Access in Ugandan Higher Education (Ojulong, Kaddu, and Magara). Public universities often sit closer to RENU and NITA-U support. Private and thinner faculties still fight commercial ISP costs and weaker device pools. Faculty planners should not pretend those differences are only “student attitude.”
National gains are real. Local bottlenecks remain.
Uganda’s research and education network has moved. RENU reports a core backbone upgrade to 200 Gbps (announced around its twentieth anniversary milestone): RENU Marks 20 Years with Major Network Milestone, Upgrading Backbone to 200 Gbps. Metro eduroam expansion, including solar-powered access points in some deployments, aims to keep identity-based Wi-Fi usable beyond a single building: Metro eduroam deployment for greater impact.
In March 2026, Makerere University and RENU commissioned a National AI Research Cloud to provide GPU-backed compute and dataset storage for research and innovation cohorts: Makerere University and RENU partner to commission a National AI Research Cloud. That is genuine national capacity. It does not automatically mean every undergraduate Information Security lab at every university has stable machines for observed assessment tomorrow morning. National cloud and campus lab are different layers. Faculties need a phased plan that starts with teachable rooms, then evidence capture, then shared AI access.
Phase 0: Baseline in two to four weeks
Before buying speeches or servers, measure.
Working seats vs class size. Count PCs that boot into a usable image within five minutes. Divide by the largest scheduled lab group. If the ratio is worse than about 1:2 without rotation plans, CBE observed labs will fail.
Power map. Note UPS coverage (servers only vs benches), generator runtime, and typical outage windows. Schedule high-stakes practicals away from known weak hours where possible.
Bandwidth truth. Measure lab and classroom throughput at peak, not at 7 a.m. Record whether LMS, package installs, and video calls share one congested link.
LMS reliability. List last semester’s outages, slow upload windows, and whether staff can fall back to USB or printed checklists without losing grade integrity.
Dead kit list. Projectors, switches, VGA/HDMI adapters, keyboard stock, and VM hosts. Small failures cancel big policy.
Write the baseline as a one-page faculty note with dates and owners. Treat it like planning evidence in an audit: scope, risk, and what you will test next. If that language helps your team, see Part 1 of the IT/IS audit series for how planning separates claims from checks.
Phase 1: Make one lab teachable
Do not renovate every room at once. Pick one lab that serves required computing modules.
Image a standard desktop (or offline VM set) for Unix, networking, and security practicals so every seat starts from the same baseline.
Keep paper or local PDF checklists that still prove competence when the LMS is unreachable.
Align lab slots with power reality. A brilliant rubric is useless if the bench dies at minute twelve of a timed task.
Use eduroam or institutional Wi-Fi where RENU membership exists; document guest access rules for industry visitors.
Assign one technician owner for image refresh and cable hygiene. Human capacity is infrastructure.
At MUBS, the difference between a chaotic week and a calm week is often whether the image was ready on Monday. Students notice. Assessors notice. Industry guests notice.
Phase 2: Capacity for CBE evidence
CBE asks students to show competence over time. That needs storage, observation space, and predictable upload paths.
Observed stations: clear sight lines, enough room for an assessor to watch without blocking peers.
Evidence store: a shared drive or Git classroom for config dumps and short reports, with retention rules that match exam policy.
Low-bandwidth assessment windows: quizzes and text uploads first; large media only when the link can take it.
Portable fallback kit: spare AP, projector, and a few Ethernet cables in a labelled bag.
One industry touchpoint: a practitioner who reviews whether a lab outcome matches a real ticket, not a ceremonial speech.
Assessment redesign and infrastructure travel together. If you are rewriting essays into viva and log evidence, you need seats that boot: AI as a catalyst, not a crutch, for university assessment.
Phase 3: AI-ready without vanity projects
AI readiness for a computing faculty is not a logo on a banner. It is a stable path to compute, clear data rules, and staff who can teach with the tools.
Confirm how your institution connects to NREN services and whether research cloud access is available through national partners such as the Makerere-RENU AI Research Cloud model.
Prefer shared GPU or cloud quotas for final-year and research cohorts before buying idle hardware that overheats in a small server room.
Write simple rules for what student data may enter public generative tools. Governance belongs beside bandwidth: internal AI governance policies for Ugandan universities.
Train faculty on AI literacy before pushing mandatory student AI tasks. Untrained staff produce unfair marking.
Review cost per demonstrated competence each semester. A tool that burns bandwidth but never appears in an observed outcome is a distraction.
MUBS leadership has publicly pushed AI fluency across programmes. New Vision covered Principal Prof. Moses Muhwezi’s call for AI programmes and systems across courses: Universities challenged to introduce AI programmes. Fluency still needs sockets, seats, and sober scheduling.
Bandwidth and LMS: design for congestion
Even where backbone capacity rises, the last metres fail first: access points, switch uplinks, and authentication queues. Practical habits that survive peak load:
Pre-stage lab ISOs and package caches on a local mirror.
Stagger cohort uploads; never set one deadline minute for 200 video files.
Keep a “LMS down” protocol: signed paper checklists photographed later, or USB drop to a controlled folder.
Separate assessment traffic from recreational streaming where policy and network gear allow.
These are unglamorous. They protect formative evidence when continuous assessment grows under CBE.
Power: plan the outage, do not deny it
Unstable electricity is a national constraint, not a personal failure of the lab tech. Faculty responses that work in practice:
Prioritise UPS for the image server and instructor machine.
Use offline VMs so a short outage does not require a full internet resync.
Build 10-15 minute recovery buffers into timed practicals.
Where solar-assisted access exists (as in some eduroam deployments), treat it as resilience for connectivity, not a full substitute for bench power.
Equity is an infrastructure question
Gender and equity mainstreaming appears in NCHE’s CBE focus areas for a reason. Evening students, students without personal laptops, and students on weaker home links need the campus lab to work. Device loan schemes and extended lab hours matter more than motivational posters. The TOJET study’s warning about uneven private/public access should push consortium thinking: shared regional labs, joint image libraries, and pooled technician training beat isolated shopping lists.
A 90-day faculty action list
Week 1-2: finish Phase 0 baseline and name owners.
Week 3-6: complete Phase 1 on one lab; run two observed sessions with checklists.
Week 7-10: stand up Phase 2 evidence store and LMS-down protocol; trial one industry-reviewed outcome.
Week 11-13: map Phase 3 access (NREN, cloud quotas, AI literacy workshop for staff).
Week 14: report to the faculty CBET committee with numbers (working seats, outage hours, failed boots), not slogans.
If we skip that discipline, the 2027 mandate and AI enthusiasm will produce a two-speed story: programmes that can show competence on working kit, and programmes that only rewrite handbooks. Closing the infrastructure gap is not a press release. It is a phased checklist a computing faculty can finish, measure, and defend.