Glossary
Plain-language definitions of technical terms used across this tracker, with concrete comparisons to help you judge scale.
IT Load (MW)
The electricity consumed by the IT equipment inside a data centre — servers, storage, and networking gear — at a given capacity level. It does not include the power used by cooling systems, lighting, or building infrastructure.
In context
1 MW of IT load runs 24 hours a day, 365 days a year — it never switches off.
1 MW ≈ 2,900 South African households (based on average consumption of ~250 kWh/month per household).
The Cavaleros Group Cape Town facility announced 360 MW — equivalent to about 1,044,000 households, or 21.5% of Cape Town's entire peak electricity demand.
Total Facility Power
The total electricity draw of the entire data centre complex, including IT equipment, cooling systems, lighting, security, and all other building loads. This is always larger than the IT load.
If a facility has an IT load of 100 MW and a PUE of 1.4, its total facility power is 140 MW — the extra 40 MW goes to cooling and overhead.
Published capacity figures in press releases are typically IT load (the useful computing power), not total facility power. This tracker uses IT load where specified.
Power Usage Effectiveness (PUE)
The ratio of total facility power to IT load power. A perfect score of 1.0 would mean every watt goes directly to computing, with zero overhead. In practice:
- PUE 1.0 — theoretical perfect (impossible in practice)
- PUE 1.2 — hyperscaler-grade efficiency
- PUE 1.4–1.6 — industry average
- PUE 2.0+ — older or poorly managed facility
Cooling is the main driver of PUE. In hot climates like South Africa, maintaining low PUE requires more energy or advanced cooling approaches (liquid cooling, free-air economisation).
Colocation (Colo)
A data centre that rents out space, power, and cooling to multiple tenants, who install and own their own servers. The facility operator provides the building, power supply, cooling, physical security, and connectivity — tenants provide everything else.
Think of it as a high-security, climate-controlled warehouse where companies keep their own hardware rather than building their own facility.
Carrier-neutral
A colocation facility that allows multiple competing telecommunications companies and internet service providers to interconnect inside the same building, without giving any single carrier preferential access or exclusivity.
Carrier-neutral facilities become natural internet exchange points (IXPs). Equinix's Johannesburg campus hosts JINX (Johannesburg Internet Exchange) — one of Africa's busiest traffic hubs — precisely because multiple carriers can meet there.
Hyperscaler
The largest cloud computing companies — AWS, Microsoft Azure, Google Cloud, Meta, and a handful of others — that operate at planetary scale. They build or occupy hundreds of data centres globally, each consuming tens to hundreds of megawatts.
Their arrival in a country typically unlocks cloud regions (local availability zones), which reduces latency for local businesses and enables compliance with data residency laws.
AI Training vs AI Inference
Two very different computational jobs that both happen in data centres:
AI Training
Creating an AI model from scratch by processing vast datasets. Extremely power-intensive, done periodically, requires high-end GPU clusters (Nvidia H100s etc.). GPT-4 training reportedly used ~50 MW for weeks.
AI Inference
Running an already-trained model to answer a query ("generate this image", "translate this text"). Lower power per query, but done billions of times per day — total power demand at scale rivals training.
Most facilities in Africa currently run inference workloads. Training data centres require specialised GPU infrastructure and very large power connections (100 MW+) that are still rare on the continent.
Wheeling
In South Africa, wheeling is the process of buying electricity from a private renewable energy generator (e.g. a solar farm in the Free State) and having it delivered to your facility via the national Eskom grid — even if your facility is hundreds of kilometres away.
The generator feeds electricity into the grid; the buyer draws an equivalent amount at their end. Eskom (or a municipality) charges a wheeling tariff for use of the transmission infrastructure. This allows data centres in Cape Town or Johannesburg to claim renewable energy from remote solar or wind farms. Africa Data Centres uses this mechanism with a 12 MW solar farm near Bloemfontein.
Power Purchase Agreement (PPA)
A long-term contract (typically 10–25 years) between a power generator and a buyer, fixing the price per kWh in advance. Data centres sign PPAs with renewable energy developers (solar, wind) to lock in costs and secure green energy credentials.
In South Africa, PPAs often involve wheeling (see above) since solar resources are concentrated in the Northern Cape while demand is centred in Gauteng and Western Cape. Vantage Data Centers secured 87 MW from SolarAfrica via PPA for its Waterfall City facility.
Water Stress Region
A classification from the WRI Aqueduct tool indicating how much demand exists on water resources relative to available supply. Ratings: low → medium → high → extremely high.
- Low: demand is a small fraction of available supply
- High: demand is 40–80% of available supply — seasonal shortfalls likely
- Extremely high: demand exceeds supply in dry periods
Cape Town is rated high. The city came within weeks of running out of water in 2018 ("Day Zero"). Large data centres that use evaporative cooling can consume millions of litres per year, making water stress a critical siting and design factor here.
Tier Certification (I–IV)
The Uptime Institute's four-tier rating system for data centre reliability, based on redundancy and fault tolerance:
| Tier | Uptime SLA | What it means |
|---|---|---|
| I | 99.671% | Single path, no redundancy — any maintenance requires downtime |
| II | 99.741% | Partial redundancy in power and cooling |
| III | 99.982% | Concurrently maintainable — no downtime for planned work |
| IV | 99.995% | Fault tolerant — any single failure is handled automatically |
Tier III is the standard for carrier-neutral colocation campuses serving enterprise and cloud workloads. Most facilities in this tracker target Tier III.
Data Residency
A legal or contractual requirement that data must physically remain within a specific country or jurisdiction. Relevant laws include South Africa's POPIA (Protection of Personal Information Act), Kenya's Data Protection Act, and Nigeria's NDPR.
For governments and regulated industries (banking, health), data residency rules often require in-country data centre capacity, not just a regional cloud point of presence. This is a key driver of data centre investment across Africa.
Announced MW
The declared planned power capacity of a facility at full buildout. Large facilities are typically built in phases over 5–10 years, so the announced figure represents a long-term target, not current or near-term capacity.
This tracker records announced MW because it captures the infrastructure commitment and grid impact — even if Phase 1 is only 10% of the total. Where a phased breakdown is available, it is recorded separately in IT load — current.
Confidence Levels
Every data field in this tracker is assigned a confidence level indicating how well-supported it is by primary sources.
Directly stated in a primary source with no ambiguity.
Stated in a source but not independently cross-checked.
Calculated or inferred; methodology is noted on the record. Treat as approximate.
Conflicting sources exist. Both positions are documented. Do not treat as settled fact.
Source Trust Tiers
Sources are tagged by type, ranked roughly from most to least reliable.
Government filings, planning applications, official gazette notices.
Press releases, investor relations filings, and product pages from the operating company.
Specialist outlets (DatacenterDynamics, DataCenter Knowledge, etc.) with editorial standards.
Commercial satellite imagery used to verify construction progress.
Local news outlets; variable editorial standards, used for leads and corroboration.
Analyst reports (CBRE, JLL, Synergy Research, etc.) — often behind paywalls.
LinkedIn posts, job listings, tweets — lowest trust, used only as a lead to find primary sources.