Microsoft and G42's stalled Kenya data center is a warning for the AI buildout: ambition is now running into power, finance and infrastructure limits at the same time.
The most important part of Microsoft's Kenya data center story is not that a single project has slowed down. It is that one of the world's biggest technology companies, working with one of the Gulf's most aggressive AI investors, has discovered how hard it is to plant hyperscale AI infrastructure in a market where the grid, the public balance sheet and guaranteed demand are all still catching up.
Microsoft and Abu Dhabi-based G42 announced the plan in May 2024 as part of a $1 billion digital investment package for Kenya. The headline asset was a geothermal-powered data center in Olkaria, built by G42 and its partners to support a new Microsoft Azure cloud region for East Africa. It was framed as a clean-energy cloud project and a sign that Africa could be more than an end market for AI products. It could host the infrastructure too.
That vision has now met the arithmetic. Reuters, citing Bloomberg reporting, said the East Africa data center site has been delayed after Microsoft and G42 asked Kenya to guarantee payment for a certain amount of computing capacity each year. The government could not provide guarantees at the level requested, according to the report. John Tanui, principal secretary at Kenya's Ministry of Information, told Bloomberg that talks remain active and that the project is not failed or withdrawn.
Kenya's power challenge is not a side issue. President William Ruto has warned that the full-scale facility could need about 1 gigawatt of electricity, compared with a national grid of roughly 3 gigawatts. In plain terms, one data center campus could demand close to a third of the country's installed capacity. Ruto's own description was blunt: powering it at that scale could mean switching off supply for half the country.
That is not an argument against Kenya. In many ways, Kenya is one of the more logical African markets for this kind of project. It has strong geothermal resources, a growing technology sector, improving connectivity and a government that has actively courted digital infrastructure investment. Olkaria was attractive because geothermal power offers the kind of cleaner, more reliable supply hyperscalers want as their energy use comes under closer scrutiny.
The problem is scale. A 100 megawatt first phase, which Bloomberg reported in 2024 was expected to be operational in about two years, is already a large project in a developing power market. A 1 gigawatt AI campus is something else entirely. That is the kind of demand profile usually associated with a national infrastructure plan, not just a private cloud expansion.
This is where the AI story changes. For years, cloud regions were sold as symbols of digital progress: lower latency, local data residency, more enterprise software adoption and better access to global platforms. AI turns them into heavy industrial assets. They need huge power connections, cooling systems, land, transmission investment and access to advanced chips. The cloud may feel weightless to users, but the new AI cloud is closer to a factory.
Guarantees are the new battleground
The payment dispute matters because it shows that power is only one part of the risk. If Microsoft and G42 wanted Kenya to commit to annual capacity payments, they were effectively asking the state to help underwrite demand. That makes sense from a project finance perspective. AI data centers are expensive to build, expensive to equip and risky if local demand develops more slowly than expected.
But it is a hard ask for an emerging-market government. Kenya would have to balance the prestige and long-term economic promise of an Azure region against the near-term cost of guaranteeing capacity that public agencies and local businesses may not immediately use. That is a serious fiscal decision, especially when the same country also needs investment in transmission lines, generation, schools, health systems and basic public services.
Microsoft's partnership with G42 was always more than a normal commercial tie-up. In April 2024, Microsoft announced a $1.5 billion investment in G42, with Microsoft president Brad Smith joining the company's board. The companies said they would bring AI and digital infrastructure to regions including Africa, the Middle East and Central Asia, backed by security commitments involving the United States and the United Arab Emirates. Kenya was supposed to be one of the clearest examples of that strategy in action.
Now it is becoming an example of the conditions that strategy requires. The next wave of AI infrastructure will not arrive in emerging markets simply because cloud companies want growth and governments want investment announcements. It will need bankable buyers, credible power plans and public agencies willing to make long-term commitments. Without those pieces, even a high-profile project can stall.
There is a more practical path. Smaller regional cloud projects in the 40 to 100 megawatt range may be less dramatic, but they are more likely to fit the power systems and demand curves of markets such as Kenya. They can still improve latency, support local startups, help banks and telecoms modernize, and give governments better access to secure cloud services. They just do not require the country to reorganize its electricity system around one campus from day one.
That may be where the Microsoft-G42 project lands if talks continue. A scaled-down version would not carry the same global headline, but it could prove more useful. AI infrastructure in Africa will be built by matching ambition to grid reality, not by pretending every market can absorb the same hyperscale model used in Virginia, Texas or the Gulf.
The wider takeaway is simple. AI is no longer limited by software talent or model ideas alone. It is increasingly limited by substations, sovereign guarantees and the ability to turn political excitement into durable infrastructure. Kenya's stalled data center does not mean the opportunity is gone. It means the next phase of the AI buildout will be won by countries and companies that can make the power, money and demand line up before the ribbon-cutting.
Also read: GPT-5.5 is turning AI benchmarks into an audit problem • Vapi reaches $500 million as Amazon Ring puts voice agents to work • Nvidia loses a seat at the table on Trump's China trip
