How Clean Cement Startups Are Riding the AI Boom

For the last few years, there’s been just one story dominating the economy, Silicon Valley, and much of the climate tech world too: artificial intelligence. It has consumed investor’s time and money, leaving relatively little for the rest of the startup ecosystem. But for companies that can hitch themselves to the AI boom and tie their value proposition to the data center buildout, this narrow funding focus can be a tailwind.

The most obvious beneficiaries so far have largely fallen into two camps: startups using AI to build cheaper, better products or those developing technologies to cleanly power data centers themselves. But what about the companies actually manufacturing the physical materials behind these facilities? The data center buildout is ultimately an investment in the physical economy, which largely means an investment in concrete — the most widely used man-made material on Earth.

Cement, the key ingredient that binds concrete together, accounts for 8% of global CO2 emissions, and is a major driver of hyperscaler’s scope 3 emissions. Microsoft and Google’s recent sustainability reports, for example, reveal that their largest emissions category isn’t electricity but “capital goods,” which includes the embodied carbon in their physical assets and infrastructure such as the concrete, steel, server racks, and silicon used to build data centers.

Cement is a big part of that picture because producing it typically requires burning limestone in kilns at extremely high temperatures, a process that both uses large amounts of fossil fuels and releases CO2 through the underlying chemical reaction itself. So if hyperscalers are serious about decarbonization, one might expect them to be pretty interested in startups such as Brimstone, Sublime Systems, and Fortera, each of which is pursuing a different approach to reducing cement’s carbon footprint.

And they are interested. But that alone won’t fill these company’s orderbooks or offset the headwinds generated by the Trump administration rescinding previously obligated grants. That challenge has only been compounded by climate tech’s broader fall from favor as investors chase flashier, more explicitly AI-centric bets.

Still, Cory Waltrip, Sublime’s VP of business development, told me that data centers make a fantastic beachhead market for the company’s low-carbon cement, which it produces through an electrochemical process that eliminates the need for high-temperature kilns. Hyperscalers, he said, have both the market power and financial runway to think long-term about “the way that they’re signing agreements” and “how you can structure those agreements.” Of course, “the balance sheet and the amount of capital that they allocate towards sustainability commitments” doesn’t hurt either.

Last May, Microsoft signed an offtake agreement with Sublime to purchase up to 622,500 metric tons of cement from the company’s future demonstration plant in Holyoke, Massachusetts, as well as a yet-to-be-sited full-scale facility. The deal is unique because it doesn’t require Microsoft to actually use Sublime’s cement in its data centers. Since cement is expensive and impractical to ship long distances, what Microsoft really purchased is the cement’s so-called “environmental attributes,” allowing Sublime to sell the physical product to local customers while Microsoft gets to claim the associated emissions reductions.

It was one of the first deals in the cement industry to decouple the physical product from its environmental benefits. But that good news was quickly overshadowed. Just eight days later, Energy Secretary Chris Wright announced the cancellation of 24 awards from the DOE’s Office of Clean Energy Demonstrations, including a $87 million grant for Sublime and a $189 million grant for Brimstone. That sent Sublime into a tailspin: In December, it paused plans for its demo plant, and in March it laid off roughly two-thirds of its workforce. The company has since filed a suit in the court of federal claims, alleging that the DOE breached its contract with Sublime, but a resolution could take years.

All the cement-hungry data centers in the world would struggle to make up for the loss of that federal funding. Hyperscalers want to buy low-carbon cement from companies that already have a credible pathway to commercial production, not foot the bill for a first-of-a-kind plant.

So Sublime is now pursuing “alternative scale up plans” that don’t involve the Holyoke facility, with Microsoft remaining “a committed customer,” Waltrip said. The most promising option involves co-locating with existing but underutilized standard cement plants in North America or Europe. Doing so could reduce capital costs by roughly 20% to 40%, Waltrip told me. “We can use all of the existing crushing, grinding, finishing, and storage equipment that an existing cement plant already has.”

Building in Europe — something Sublime has yet to commit to but is certainly considering — could also open the door to other non-dilutive public financing, such as the bloc’s roughly €40 billion EU Innovation Fund, which regularly backs industrial decarbonization projects such as low-carbon cement.

In the meantime, the company also says it’s made significant process improvements that could drastically change the scale at which it builds plants. While former CEO Leah Ellis described Sublime’s future commercial facility as a “megaton-scale plant,” Sublime now thinks it could economically produce the material in 50,000 to 250,000 metric tons-per-year facilities. These smaller plants would be far easier to finance without relying on large government grants, Waltrip told me.

Sublime is exploring multiple other undisclosed data center engagements as well, as Waltrip revealed that “we’ve completed materials testing with at least one hyperscaler. We’ve completed a concrete demonstration pour with another hyperscaler,” and “we’ve negotiated or are in the process of negotiating commercial agreements with other hyperscalers beyond Microsoft.”

The company also conducted a small test pour of its low-carbon concrete last year with STACK Infrastructure, a data center developer that leases out its facilities. But while the material has exceeded performance standards, STACK is unlikely to become a customer anytime soon. “If we had a commercial plant ready to go, I think we would be having no issues with finding customers for that product,” Waltrip told me. The challenge is that developers outside the major hyperscalers typically lack the financial flexibility to sign long-term offtake agreements for a product that may not reach meaningful scale until the mid-2030s.

So for now, Google, Microsoft, Meta, and Amazon remain the most sought-after buyers.

Brimstone, another low-carbon cement company, also landed a major hyperscaler deal last year. The company, which still uses kilns but replaces limestone with carbon-free calcium silicate rocks in its production process, agreed to supply Amazon with an undisclosed amount of cement and supplementary cementitious materials, which can partially replace cement in concrete. CEO Cody Finke told me he couldn’t share any additional details, including the volume of materials reserved or when he expects deliveries to begin, though he readily acknowledges the impact of the data center boom.

“There’s no question that the data center buildout has increased the demand for these materials,” Finke told me. Early last year, the company announced that it’s also figured out how to adapt its process to produce alumina — the refined material that smelters turn into aluminum. Data centers also use this metal throughout their operations in structural panels, server racks, and cooling systems. Eventually, the company says it will be able to make additional critical minerals and materials including steel, magnesium, and titanium.

For now though, Brimstone is working to complete construction of its demo plant in Reno, Nevada, which the company recently said it expects to be operational in 2028. Finke was somewhat more cautious, however, telling me only that it should come online by “the end of the decade.” The company’s first full-scale plant, the location of which it’s yet to announce, is slated to begin operations around 2034, producing 350,000 metric tons of alumina and an undisclosed amount of cement and other materials.

But like Sublime, Brimstone also lost a major source of federal support when the Trump administration rescinded its $189 million DOE grant, which was intended to finance construction of the demo plant. Finke, however, insisted this hasn’t altered the company’s timeline because Brimstone, having netted over $80 million to date, “had effectively raised the money that we needed, regardless of the grant.”

Finke isn’t relying on the goodwill of hyperscalers either, even though many do appear willing to pay a green premium in order to align with their ambitious, if flailing, decarbonization agendas. “To be frank, I don’t think that it’s that important to the transition whether or not those climate policies exist, because the companies that really matter are going to be cheaper anyway,” he told me.

Brimstone, he argues, is one of those companies. By co-producing multiple products at once, each can effectively offset the cost of the others, and Finke expects even the cement produced at the Reno demo plant to sell at standard market rates. Ultimately, while he sees growth in the data center industry as a tailwind, he doesn’t think Brimstone depends on that market, noting these facilities still only account for a small sliver of global cement demand. The company’s primary customers, he said, will ultimately be traditional buyers: concrete producers purchasing cement and aluminum smelters buying alumina.

Yet data centers willing to negotiate multi-year contracts still represent uniquely valuable first customers in an industry where such agreements are exceedingly rare. Instead, producers typically sell cement into a merchant spot market, where buyers purchase from whatever supplier meets their myriad requirements at the time. But that leaves low-carbon materials startups in a bind, Fortera’s CEO Ryan Gilliam told me. “When you’re trying to bring a new technology to market like us, you typically use offtake agreements to get project financing to justify building up big projects,” he explained. Potential investors simply want to see demonstrated future demand.

Fortera, which has raised about $150 million and has an operational pilot plant in California, captures the CO2 emitted from conventional cement production and converts it into a mineral form that then becomes part of the cement itself. Last year, it secured a strategic investment from Microsoft’s Climate Innovation Fund to help finance its first commercial-scale facility, expected to produce 400,000 tons of cement per year. In return, the tech giant secured the right to procure Fortera’s low-carbon cement and its associated environmental attribute certificates — more of a reservation than the binding offtake contract it signed with Sublime.

Just one plant of this size “would meet all the hyperscalers’ needs easily,” Gilliam told me, underlining Finke’s point that data centers will by no means represent a cement company’s largest buyer long-term. “Most hyperscalers, you’re talking maybe upwards of 100,000 tons a year of requirements around cement, and that might even be at the upper end,” Gilliam explained. By comparison, standard cement plants typically produce about a million tons of product annually.

So while Gilliam and others are happy to ride the AI boom, they also recognize that data centers are likely more valuable as an early market signal than a long-term source of demand. Even now, it remains unclear whether the boom is even a net positive for the sector as a whole.

“The number of AI startups and the amount of money that’s been diverted into that space definitely changed the pool of investors that you can go to right now,” Gilliam told me. And that’s the core paradox. The data center boom has become one of the clean cement industry’s most promising early markets and one of its fiercest competitors for capital. Welcome to the AI economy.