Insight, Energy Systems
Firm clean power has spent a decade without a buyer willing to pay a premium. It has one now, and it is the data centre industry. That solves the hardest problem in energy innovation and quietly hands the direction of technical change to a single class of customer.
Two announcements this week describe the same structural change from different ends.
SB Energy filed for a United States listing on 1 September, running a business that combines power generation with hyperscale data centre development, reporting 8.8 gigawatts of capacity contracted or under construction and a backlog figure in the hundreds of billions of dollars, with OpenAI among its customers (as reported).
Fervo Energy agreed to supply close to 400 megawatts of enhanced geothermal power for a future Google data centre in Utah, with an option to expand toward a gigawatt by 2030, from a project that uses drilling techniques adapted from oil and gas (as reported).
The usual reading is that AI is straining the grid. The more interesting reading is the opposite. For the first time, first-of-a-kind firm clean generation has a customer with an enormous balance sheet, a genuine need for round-the-clock carbon-free electricity, a willingness to sign fifteen-year contracts, and no particular sensitivity to a price premium.
That is a very good thing, and it comes with a consequence that nobody is naming.
Enhanced geothermal, advanced nuclear, long-duration storage and clean firm generation generally have failed for a decade on the same obstacle. They are expensive at first-of-a-kind scale and get cheaper only by being built repeatedly. Building them repeatedly requires somebody to buy the expensive early units.
Utilities are poorly placed to be that buyer. They are regulated on prudence, which means paying above market for an unproven technology invites a disallowance. Their customers are politically sensitive to tariff increases. And in most systems, as we have described in Kenya, the utility's balance sheet is in no condition to take a technology risk.
A hyperscaler has none of those constraints. Electricity is a small share of its cost base relative to the value of the compute, a carbon-free supply has strategic and reputational value, and it can sign a corporate power purchase agreement without a regulator's permission.
So the missing link in clean energy commercialisation has appeared, and it did not come from energy policy. It came from an unrelated industry's demand for compute.
Daron Acemoglu's account of directed technical change sets out the mechanism precisely. Two forces shape which direction innovation takes. The price effect pushes innovation towards scarce factors. The market size effect pushes technical change towards whatever serves the larger market.
The market size effect is the one operating here, and it is powerful because the market that has appeared is not general. It is one class of customer with a very particular load.
A data centre wants power that is firm, continuous, co-locatable, quick to deploy and available in blocks of hundreds of megawatts at a single point.
Rank clean technologies against that specification and the ordering is not the same as the one a decarbonising grid would produce.
Enhanced geothermal fits almost perfectly: firm, baseload, dispatchable, sited where the resource is, scalable in large blocks. Gas with carbon capture fits. Advanced nuclear fits, if it ever arrives. Large solar paired with multi-hour storage fits reasonably.
Now the technologies that do not fit. Demand response and flexibility, which we have argued is the cheapest and least remunerated resource in most systems, is of no interest to a customer whose entire proposition is not being interrupted. Distribution-level solutions serving many small loads have no champion here. Long-duration seasonal storage, useful to a grid managing winter, is irrelevant to a constant load. Heat networks, industrial process heat and building decarbonisation are not in the frame at all.
None of that means the neglected technologies are better. It means the selection is being made by a buyer whose requirements are unrepresentative of the system as a whole, and the selection will persist, because the technologies that get built get cheaper and the ones that do not, do not. That is the ordinary path dependence mechanism, running on a demand signal rather than a policy one.
Location follows the buyer, not the system.
Fervo's resource is in Utah because that is where the geology is. The data centre is going there because the power is. That is efficient for both parties and it means new firm generation is being sited where a private customer wants it rather than where a grid is constrained. A gigawatt built next to a load that consumes it entirely relieves nothing elsewhere.
Talent and supply chain follow too.
Enhanced geothermal draws on drilling crews, rigs and subsurface engineering. So does conventional oil and gas, and so, increasingly, does carbon storage. The adjacency that makes geothermal attractive as a labour transition also means these sectors compete for the same finite pool of people, and the sector paying data centre prices will win.
The premium may not survive contact with cost pressure.
Corporate clean power procurement has been generous while compute margins have been extraordinary. If those margins compress, the willingness to pay above market for carbon-free firm power is the first line item to be examined, and the technologies that scaled on that premium will discover their real cost curve at an awkward moment.
The instinct to be pleased about this is correct. Private demand is doing what a decade of public procurement failed to do, and the resulting cost declines in enhanced geothermal will eventually be available to everybody.
But if the direction of technical change is now being set by one customer class, then public energy innovation funding has a clearer job than it did, and it is not to co-fund what hyperscalers are already buying.
Fund the technologies the anchor customer does not need. Flexibility, demand response, distribution-level solutions, seasonal storage, heat. These now have no private champion at all, and they are the ones a decarbonised grid serving households and industry actually requires.
Attach system conditions to the connection, not to the subsidy. Where new firm generation is built for a private load, the question is whether the interconnection, the grid upgrade and any surplus capacity serve the wider system or only the campus. That is decided in connection agreements, and it is the same wheeling or islanding question that is hollowing out South African electricity.
Watch the cost curve, not the announcement. The value of this to everybody else depends entirely on whether repeated construction actually lowers unit costs, and whether those lower costs become available outside the corporate procurement market. That is measurable, it will take five years, and it is the only thing that determines whether this was an energy transition or a private supply arrangement.
The Lab works on this in energy access and on how technologies behave when they meet a real system through entering a new context.
If you are funding clean firm generation and want to know which part of the system it will actually serve, tell us what you need to know.
This is an independent insight piece by Transitions Lab. For the Lab's applied work, see Energy Access & Systems. See also Paying for Power You Curtail on why flexibility is valuable everywhere and remunerated almost nowhere, The Customers Who Can Leave on what happens when large loads build their own supply, and The Lock-In Runs Both Ways on how early deployment decides what gets cheap. To discuss a study, see Contact.