Why Kenya's Data Centre Market Estimates Disagree

Ask five research firms for the size of Kenya's data centre market and you will get answers that cannot all be true at once. For the 2025 base year, published estimates run from $266 million to $606 million, a factor of 2.3 between the smallest and the largest. Growth forecasts stretch from 4.67% to 32% CAGR, a seven-fold spread. And against all of them, the most methodologically transparent public source sizes Kenya's commercial colocation revenue at roughly $25 million a year. Somebody is measuring something very different from everybody else. This article is the one nobody else in this market writes: we fetched the actual report pages, every one on the same day, 8 September 2026, and explain, definition by definition, why the numbers disagree, and how a serious reader should use each one.

This matters beyond curiosity. These figures get cited in investor decks, government policy briefs, and journalist headlines, usually without the base year, the definition, or the firm's method attached. An investor reading "$606 million market" and an operator reading "$25 million colocation revenue" will make very different decisions about the same country. Both would cite "research". Only one of them would be looking at the market as it physically exists, the one you can verify facility by facility in our DC Directory.

The estimates, side by side

Here is every Kenya data centre market estimate we could verify as of 8 September 2026, with the firm, the figures, and the date, the way they should always be cited:

Look closely at the table and several things jump out. Ken Research's headline numbers are identical to Arizton's, same $266M, same $805M, same CAGR to the decimal, which means they are one data point wearing two logos, whether through licensing, a shared upstream dataset, or circular citation. Credence and DCMarketInsights publish no verifiable dates on their free pages. Mordor is not even measuring the same quantity: it sizes capacity (15 MW growing to 25 MW) rather than revenue. Only Xalam's numbers, roughly 15 MW live, $25 million a year of colocation revenue, 55 MW under construction and ~25 MW more in pipeline, roughly five-fold growth at 32% CAGR, come from a study whose facility-level method is publicly visible, because it was published openly by the EU's Digital Investment Facility under the D4D Hub in January 2026.

Reason 1: Nobody agrees on what "the market" is

The single biggest driver of divergence is definitional, and almost none of the summaries state their definition. At least four different things get called "the Kenya data centre market":

• Colocation revenue, what operators earn from renting racks, cages, suites, and power. This is the tightest, most defensible definition, and it is what Xalam measures: roughly $25 million a year. You can sanity-check it: ~15 MW of live commercial capacity at even a generous $150–200 per kW per month yields $27–36 million at full occupancy, and actual occupancy runs around 60–65%.
• The total data centre economy, colocation plus enterprise self-build capex, connectivity and managed services, cloud on-ramps, and professional services. This definition gets you into the hundreds of millions, which is where Arizton's $266M and 6Wresearch's $369M plausibly live.
• Investment value, the capital cost of building facilities, sometimes cumulative over multiple years. DCMarketInsights' $606M and Credence's $509M read like this: a single 100 MW AI campus would blow past them, but a market measured in "everything spent on data centres, recently" fits.
• Capacity, megawatts, not dollars. Mordor's 15 MW→25 MW is honest about being this; capacity estimates at least have the virtue of being checkable against facilities.

None of these definitions is wrong. But adding them together in a headline, or worse, comparing them against each other, produces exactly the confusion the Kenyan market suffers from. When we carry market numbers in our own market-in-numbers piece, we label each figure with firm, date, and what it appears to measure, because the label is most of the information.

Reason 2: Circular citation and the illusion of consensus

Market research has a supply chain, and Kenya sits at its thin end. The global firms maintain deep models for the US, Europe, and China; African markets get modelled from fewer inputs, and smaller research houses often license or adapt figures from larger ones. The result is a phenomenon worth naming: two reports citing each other's number is not independent confirmation. Ken Research quoting Arizton's exact $266M→$805M at the same CAGR is the clearest documented case in this market, but the pattern is general, search for any Kenyan market stat and you will find the same figures echoing across aggregator sites with the original source stripped away.

The practical rule for readers: count sources, not reports. Before you describe a number as "confirmed by multiple studies", check whether the studies share a lineage. In our coverage, we try to keep at least one methodology-visible source (currently Xalam via the D4D Hub) and at least one definitional outsider (a capacity-based count like Mordor's) in the same conversation, because agreement between different methods is evidence, while agreement between copies is just repetition.

Reason 3: Base years, forecast windows, and the moving target

The Kenyan market is small enough that a single facility launch moves it by double digits. Digital Realty's 6.4 MW NBO2, opened 7 September 2026, increased the country's verified commercial colocation capacity by roughly 40% in one day, on paper, at least, until occupancy fills in. A model with a 2024 base year (Credence), a 2025 base year (Arizton, 6Wresearch), and a mid-2025 cut-off (Xalam's January 2026 publication of its Kenya briefing) are therefore measuring visibly different markets. Add different forecast windows (2030 vs 2031 vs 2032 vs 2035) and identical underlying data would still produce different CAGRs and headline "future sizes".

This is also why the same firm's numbers move between editions. Arizton's Kenya page was updated in June 2026; its earlier editions carried different totals. That is not falsification, it is a small market repricing as real capacity lands. But it does mean any citation without a date is close to meaningless, and it is why our methodology attaches a fetch date to every claim we publish, including the ones in this article.

Reason 4: Paywalls, summaries, and what journalists actually read

Almost none of the people citing these reports have read them, because almost all of them are behind paywalls costing thousands of dollars. What circulates publicly is the free executive summary (typically one paragraph and a chart) plus secondary reporting that quotes the summary. The definitional subtleties that explain the divergence (what revenue is counted, what is excluded, how the base year is constructed) live in the paid body of the report, not the summary. So the public conversation runs on numbers whose definitions are invisible by construction.

The honest response is not to buy six reports; it is to read the free layer for what it is (directional and named) and to anchor on what can be verified for free. PeeringDB's facility records, operator disclosures, Uptime's certification directory, and registry data cost nothing, which is why our directory builds on them and why a publicly released study like the Xalam/D4D Hub briefing is worth so much: it is the only estimates document in this stack whose method a stranger can audit.

How to read any Kenya market number (the working checklist)

After walking through the divergence, here is the checklist we apply before repeating any market figure, and that we suggest you apply to ours:

1. Name the firm and the date. "Arizton, June 2026 edition", not "research shows".
2. Identify the quantity. Revenue, investment, or capacity? Colocation-only or the total economy? If the summary does not say, the number is decoration.
3. Check for siblings. Identical figures at two firms usually mean one shared source, not two confirmations.
4. Prefer the auditable. A study whose method is public (Xalam/D4D Hub) beats a bigger number you cannot inspect. Free registry data beats both when it exists.
5. Reconcile with physical reality. Any revenue claim should roughly cohere with live megawatts, occupancy, and pricing; any capacity claim should cohere with the facility list. Our directory exists precisely so this check takes minutes, not weeks.

Run the $266M and $606M figures through that checklist and they stop contradicting each other: they were never measuring the same thing. Run the "$25 million colocation market" through it and it becomes the most credible single number in circulation, the one consistent with fifteen-odd megawatts of live commercial capacity at regional pricing. That is not a smaller story than "$606 million"; it is a true one, and for an emerging market the true version is more useful: it tells you where the real upside is, from a small, verified base growing five-fold, not from a large number that was never measurable to begin with.