Keeping up with Climate Tech vol. 16

Across the U.S., utilities and technology companies are scrambling to find enough generation and grid capacity for an unprecedented buildout of artificial intelligence infrastructure. Gridless, however, starts from the other end of the equation. In parts of the developing world, some power projects have the opposite problem: power without enough paying customers.

Gridless sees an opportunity to put that surplus electricity to work powering the world’s growing demand for computing.

Overview

Gridless, founded in 2022 by Erik Hersman, Janet Maingi, and Philip Walton, has found hydro, geothermal, biomass, and solar resources in Africa with capacity to produce inexpensive electricity without enough nearby demand to fully monetize it. The company is backed by Block, Inc. and Stillmark.

Gridless’s solution to this problem sounds almost too simple: instead of moving more electricity to computers, move the computers to the electricity.

The company has spent the past four years testing that premise with Bitcoin mining. At six renewable energy sites across Kenya, Malawi, and Zambia, modular mining containers consume electricity that surrounding communities do not. When local demand rises, the computers can reduce their consumption almost instantly. Gridless says that model has helped lower community tariffs at some hydropower sites and provided a new revenue source for otherwise underutilized generation. Across its sites, Gridless has helped bring electricity to about 30,000 people.

Now, Gridless wants to put graphics processing units (GPUs) — which provide massive parallel processing power to accelerate artificial intelligence training, inference, and high-performance computing workloads — into the same architecture.

“These big assets get built and there’s no off-taker, so that capacity goes unused,” Tre Mansdoerfer, President of Gridless AI, the company’s AI compute division, and HBS ’26, told the Harvard Technology Review. “So Gridless can come in and put modular compute containers and actually monetize it.”

Gridless is working with power producers across several African countries that have stranded or underused hydro and geothermal generation. That power is low-cost because it would otherwise go unused.

If that model scales, Gridless’s proposition becomes much larger than finding cheap power for AI. The startup is betting that global demand for compute could become the customer that makes new electricity infrastructure financially viable in the first place.

Gridless’ Origins: Power Plants in Search of Customers

At a community hydro plant in Zambia, Gridless runs bitcoin miners in a modular container alongside the local mini-grid. Rather than competing with residents for electricity, the computers use generation that would otherwise go unsold.

That distinction is central to understanding Gridless’ value proposition. A power plant cannot earn money from simply possessing generating capacity. Someone has to buy the electricity.

Rural mini-grids make that problem particularly acute. More than 560 million people in Sub-Saharan Africa still lack electricity, and off-grid systems are expected to play an important role in extending access. Yet mini-grids frequently have to be sized for the potential demand a community might reach rather than what households actually consume right now. What results is a huge volume of unused capacity during a project’s early years.

A newly connected household may initially use electricity for little more than lights, phone charging, and a handful of appliances. Forbes reported that an average African mini-grid customer can consume in a month what an American household uses in 12 hours. The panels, turbines, batteries, and distribution infrastructure, however, still have to be financed upfront.

Gridless’ deployment at the Zambian hydropower plant demonstrates what happens when another customer enters that equation.

Gridless installed Bitcoin mining equipment to buy the plant’s excess generation. Forbes reported that within roughly a year, the additional revenue helped support connections for around 2,000 more people and 40 businesses, including clinics, corn mills, and barbershops.

“Bitcoin ramps up and down as community demand changes,” Mansdoerfer explained. “It can arrive quickly, consume almost anywhere, and leave when a higher-priority customer needs the extra power.”

And because the computers are modular, if local electricity consumption rises enough, or the Zambian hydropower plant gains another outlet for its power, Gridless can redeploy its equipment elsewhere.

Where AI Figures

Gridless now sees AI as the next layer of that system.

Long before artificial intelligence made electricity procurement a boardroom concern for technology companies, Bitcoin miners had already spent years searching globally for cheap and underused electricity.

“The Bitcoin community plays an integral role in the AI infrastructure space,” Mansdoerfer said. “When mining Bitcoin, they faced the same constraint: energy and, more specifically, its cost. They’ve been building sites for 10 years because that’s just what their computing required.”

In his view, the crypto-mining industry spent the past decade solving an early version of the problem AI developers now face. Compute has to move toward power rather than power toward compute.

Rather than building every GPU into a handful of enormous centralized data center campuses, the company imagines a network of computing sites located around available electricity. Some could be relatively small installations attached to mini-grids. Others could be larger installations beside major geothermal or hydroelectric resources.

That does not mean Gridless can simply swap Bitcoin miners for GPUs. Its existing mining sites were designed precisely because Bitcoin tolerates conditions that commercial AI workloads often cannot. AI customers may have service-level agreements, latency requirements, uptime guarantees, and expensive jobs running continuously. GPUs themselves can also cost orders of magnitude more than Bitcoin-mining equipment, making physical security, redundant connectivity, cooling, and reliable electricity much more important.

Remote sites use a mix of fiber and satellite connectivity. In markets like these where conditions are not yet suited to AI workloads, bitcoin remains the flexible load that keeps plants viable.

Gridless is now planning sites built for AI inference.

The challenge is that local AI demand alone is unlikely to support the scale Gridless hopes to build initially. Its proposed solution is to use global demand as the anchor.

A hydroelectric megawatt in Africa cannot easily be exported to a customer in San Francisco. But that megawatt can power a GPU there, which can process an American customer’s workload and send the result back across fiber.

The company’s strategy is therefore almost geographically bifurcated. Gridless’s African team develops relationships with power producers, while Mansdoerfer leads Gridless AI’s compute partnerships from San Francisco.

Gridless’ model therefore links two groups that rarely intersect directly: energy projects in search of large, dependable customers and technology companies in search of inexpensive electricity. Gridless’s wager is that bringing those markets together can support a new geography of computing organized around the availability of electricity rather than the location of its end users.

Community First, AI Second, Bitcoin Third

At small sites, Gridless plans to give community consumption first priority.

“If we don’t lean into the community, if we don’t prioritize that and have that in the back of our mind, then we’re doing the wrong thing,” he said. “We care about the local communities alongside which we are building.”

A portion of power may support relatively firm AI workloads. Bitcoin miners then absorb whatever generation remains and turn down as either of the first two categories grows.

At larger sites, the arrangement would look different. An AI facility cannot be expected to disappear every afternoon when electricity demand rises. Instead, Mansdoerfer imagines new energy projects being built with enough generation to support both a contracted block of data-center demand and additional electricity for the surrounding grid.

The goal is for each AI site to add new generation for the surrounding community, not only to consume it.

What’s Next

Gridless’s near-term focus is bringing its first AI capacity online with compute partners.

“As soon as you have some meaningful compute running in Africa, I think this is a J-curve and things start moving quickly for us,” Mansdoerfer said.

That first deployment would serve as proof of something considerably larger than whether GPUs can technically operate in Africa. Gridless needs to show that global customers will place meaningful workloads on infrastructure located there; that its larger sites can meet the reliability, connectivity, and security requirements those customers expect; and that cheap power can offset the additional complexity of operating outside established data-center markets.

“We want to build AI capacity at scale and host that compute in Africa,” Mansdoerfer said.

The ambition also extends beyond simply attaching data centers to existing generation. Gridless ultimately wants compute contracts to help bring new energy projects into existence. Some of the hydro and geothermal opportunities in its pipeline have not been developed precisely because they lack a large enough customer to support financing. A sufficiently durable AI offtake agreement could change that equation: guaranteed compute demand creates guaranteed electricity demand, which in turn can make a new generation project easier to finance.

Over the next five years, Gridless aims to scale AI capacity significantly and bring new electricity to many more communities.

But the company’s wager remains searingly straightforward: go where the power is, bring computing demand with you, and use the relationship between the two to make both infrastructures more valuable.