Why Is Olkaria’s Geothermal Power Not Yet Winning Kenya the AI Data-Centre Race?

Kenya will spend KSh 2.30 trillion servicing its debt in the year to June 2027, according to National Treasury projections. That is 79.8% of ordinary revenue, up from 72.4% the year before. Public debt stood at KSh 13.01 trillion at the end of June, or 68.5% of GDP.

Much of that money comes from home. Domestic debt makes up 56.3% of the total, and banks hold 34.6% of it. The 91-day Treasury bill paid 8.76% at its latest auction. Every shilling a bank lends to the state is a shilling it does not lend to a business.

Raising taxes has limits too. In June 2024, protests against the Finance Bill forced the government to withdraw it.

About 90 kilometres north-west of Nairobi, under the floor of the Rift Valley, sits something the world’s largest technology companies are now paying a premium for: heat that can make electricity day and night. Kenya had 980 MW of geothermal capacity at the end of 2025, the sixth-largest fleet in the world, and official estimates put its potential at up to 10,000 MW. In September, Google signed what its supplier calls the largest contract yet for next-generation geothermal power: 396 MW in Utah.

Kenya has not yet sold its heat to that market.

The Olkaria stall

In May 2024, during President William Ruto’s state visit to the United States, Microsoft and the Abu Dhabi group G42 announced a $1bn package for Kenya built around a geothermal-powered data centre at Olkaria in Naivasha. The first phase was to draw 100 MW, with room to grow to 1 GW. Kenya’s entire grid supplies about 3 GW.

Ruto described the problem in remarks in Nairobi reported in early May 2026. “To switch on that one data center, we would need to shut off power for half the country,” he said.

Power was only half of it. Microsoft and G42 wanted the government to commit to paying for a fixed amount of capacity every year, Bloomberg reported. With debt service taking most of its revenue, the Treasury could not agree, and talks stalled.

Microsoft kept building elsewhere. In August it opened its fourth Indian cloud region, in Hyderabad, part of $20.5bn it has committed to India, cooled by a system it says uses no water.

This week Microsoft told staff at its Africa Development Centre in Nairobi that the centre’s engineering work would end, putting nearly 400 jobs at risk, according to employees who spoke to TechCabal. Microsoft calls the changes proposed and part of a global restructuring, and nothing links them to the Olkaria talks. But they land five months after the Olkaria project stalled, and together the two decisions show how little of the AI economy Kenya has yet captured.

Steam is not the problem.

Kenya’s heat is proven. KenGen, the state's power generating company, runs 754 MW of geothermal plant, most of it at Olkaria. Its near-term pipeline is modest. The rehabilitation of Olkaria I, the oldest plant, lifts it from 45 MW to 63 MW, and the first rebuilt unit went back on the grid in June. Olkaria VII, approved by the Cabinet in July 2025, will add 80.3 MW by June 2027. Together they add about 98 MW, roughly the size of the stalled data centre’s first phase alone.

The way KenGen develops geothermal power helps explain the pace. Its traditional model builds generation for the national grid, finances projects through a mix of its own funds and development lenders, and sells most of the electricity to Kenya Power under long-term power purchase agreements. That approach has helped make Kenya a global geothermal leader, but it is geared to national-system expansion rather than delivering several hundred megawatts on a compressed timetable for a single hyperscale buyer. Much of the development and offtake risk also remains within state-controlled institutions, even as KenGen begins testing more direct supply models for large industrial customers at Olkaria.

How Google and Fervo did it.

Google faced the same problem from the other side. It wants its data centres to run on carbon-free power every hour of the day by 2030, and wind and solar cannot guarantee that alone. In 2021 it signed an agreement with Fervo Energy, a Houston start-up, to develop next-generation geothermal power.

Fervo’s method comes from the shale industry. Instead of searching for rock that is hot, fractured and full of water, it drills into hot, dry rock, turns the wells sideways and pumps fluid in to create its own fractures. The heat exchanger is manufactured, not found.

The progression took five years.

Proof. At Project Red in northern Nevada, Fervo landed two wells about 2,350 metres down and ran them sideways for about 990 metres through rock as hot as 190°C. A 30-day test in 2023 pushed 63 litres a second through the system, enough for 3.5 MW. By November 2023 the project was feeding the grid that serves Google’s Nevada data centres.

Payment. Corsac Station solved the question that sank Olkaria: who pays? In May 2025, Nevada’s utility regulator approved the Clean Transition Tariff. NV Energy buys 115 MW from Fervo and dedicates it to Google, and Google pays for it. The cost of a new technology stays off ordinary customers’ bills, and the state guarantees nothing.

Scale. Fervo listed on Nasdaq in May 2026, raising about $1.9bn. On 1 September it signed the 396 MW deal with Google for a Cape Station expansion due in 2028, with an option for about 600 MW more by June 2030. Four weeks later the first 33 MW block of Cape Station’s first phase entered commercial operation, 23 months after construction began.

Drilling like a factory.

Fervo treats geothermal as manufacturing rather than exploration, and its costs are falling as a result.

It drills several wells from one pad and builds power plants in standard blocks. By February 2024 its fastest well at Cape Station took 21 days, 70% less time than its first horizontal well, and the cost of its first four Cape wells had fallen from $9.4m to $4.8m each. It now aims to build each block in 18 months.

Fibre-optic cables cemented into the wells let engineers see where water moves and where rock is cooling, so they can adjust flows before a fracture short-circuits. On the surface, binary plants pass the hot fluid through a heat exchanger, send it back underground and cool the working fluid with air, so they consume little water.

The Rift valley’s advantage

Fervo has to drill deep to find enough heat. At Project Red it reached about 190°C at roughly 2,350 metres. At Olkaria, temperatures pass 200°C at about 1,000 metres, according to studies of the field’s wells. Kenya also has what Fervo has to manufacture: natural fractures and water.

Kenya does not need enhanced geothermal to find heat. It can borrow Fervo’s industrial habits, pad drilling and standard blocks, to build faster, and Nevada’s tariff to make the buyer carry the cost. What Kenya lacked in May 2026 was not steam. It was a way to build at the buyer’s speed and a contract that kept the bill off the Treasury.

Four moves

First, one door. Singapore’s Economic Development Board, set up in 1961, gave foreign investors a single office for land, permits and incentives. The presidency should create a single-window Sovereign Compute Authority on that model and ring-fence the Rift Valley’s geothermal zones under it. Steam rights, land, environmental approval and transmission would sit in one office instead of several ministries, and new steam fields would be reserved for AI compute clusters, sovereign cloud regions and export-grade digital infrastructure paid for by the companies that use them.

Second, build in blocks. Nairobi should instruct KenGen and the Geothermal Development Company to move from developing one plant at a time to building standard, modular blocks. Joint-development and technology-transfer agreements with horizontal-drilling leaders such as Fervo Energy and the big oilfield service contractors would bring multi-well pad rigs, polycrystalline diamond compact (PDC) drill bits and air-cooled binary plants that use almost no water to Olkaria and Suswa. Fervo builds Cape Station in 100 MW phases made of 33 MW blocks. Kenya should standardise around repeatable blocks of that kind and set itself a public target: 1,000 MW of dedicated compute power under construction within 36 months.

Third, a Kenya Clean Transition Tariff. The Energy and Petroleum Regulatory Authority should write a K-CTT modelled on the tariff Nevada approved in May 2025. Instead of asking a debt-burdened Treasury for annual capacity guarantees, Kenya would offer any hyperscaler, whether Google, Microsoft, Amazon, Meta or G42, fast-tracked land and wellhead leases at Olkaria on one condition. The company signs a 15-year, US-dollar power purchase agreement that pays 100% of the cost of its geothermal blocks and of the dedicated high-voltage lines that serve them. Those private lines would run outside the 3 GW national grid, so Kenyan households carry neither the blackout risk nor the tariff increases, and a gigawatt-scale campus would bring tens of billions of dollars of computing hardware onto Kenyan soil.

Fourth, keep the engineers. Kenya should hold on to the talent leaving Dunhill Towers this week. Rather than watch nearly 400 engineers from Microsoft’s Westlands centre scatter to London, Dubai or Amsterdam, local pension funds, Kenyan banks and the Sovereign Compute Authority (Infrastucture fund? Anyone? Tag James Mworia) should seed a Kenyan reservoir telemetry and cloud venture. Its first hires would be the people who spent the past seven years building global cloud identity, security and machine-learning infrastructure. Their first jobs: writing the fibre-optic sensing algorithms, automated well controls and sovereign cloud software for the Rift Valley.

The clock

HE. President William Ruto has promised to lift Kenya’s generating capacity to 10 GW by 2030. Google’s option on another 600 MW in Utah runs to June 2030. Kenya has the steam. It still needs the contract.

Our team has compiled the technical, geological, financial and regulatory groundwork for this plan: thermal-gradient maps of the Rift Valley, capital models for modular geothermal blocks, draft statutory text for a Kenya Clean Transition Tariff and a blueprint for absorbing the Westlands engineers. We are ready to work with any leadership prepared to stop waiting for foreign cost centres and start building Kenya’s Singapore moment.