AI Is Being Built on Concrete. We’re Ignoring Its Biggest Carbon Problem.

As spending on AI infrastructure soars toward a trillion dollars a year, the carbon locked into data center concrete is permanent. The tools to cut it in half already exist.

The numbers behind the AI boom have stopped sounding real. The five largest technology companies spent roughly $300 billion on capital projects in 2025. Analysts expect that figure to approach $750 billion this year and to pass $1 trillion in 2027. Each forecast has been revised upward, and every new estimate has meant more construction.

Much of that money will become buildings: data center campuses spanning millions of square feet, resting on vast foundations and slabs of concrete. The debate over AI’s environmental footprint has focused almost entirely on electricity. That matters. But the carbon in a data center’s concrete is released before a single server is switched on, and nothing done later can take it back.

Concrete is the most widely used building material on Earth after water. Its binder, cement, accounts for roughly 8 percent of global carbon dioxide emissions. Most of that comes from heating limestone in kilns, a chemical reaction that releases carbon no matter how clean the power supply. A greener grid cannot fix it.

This is where I believe the climate conversation has its priorities backward.

A great deal of attention and investment is flowing to carbon capture: technologies that pull carbon dioxide from smokestacks or from the air and store it underground. We will need some of these tools for emissions we cannot avoid. But capture is expensive, energy-intensive, and still early in its development. Every captured ton must be separated, compressed, transported, stored, and monitored, often for decades.

A ton of carbon that is never emitted needs none of that. It costs nothing to store, never leaks, and needs no verification decades from now. Avoiding emissions at the mixer is not a lesser climate strategy than capturing them later. For a material produced on concrete’s scale, it is the most efficient strategy we have.

The frustrating part is that the industry has known how to avoid much of this carbon for decades. Industrial byproducts such as slag cement can replace a large share of conventional cement at a fraction of the emissions. Yet most concrete still uses only modest amounts. The barrier is not chemistry. It is strength and schedule.

Mixes with high replacement levels gain strength more slowly in their first days. On a job site, that means forms stay up longer, and schedules slip. When a hyperscale campus is racing to come online, carbon loses that trade-off every time.

Nanotechnology is now changing that equation. Materials engineered at the scale of billionths of a meter, added to concrete in very small doses, give cement’s chemistry more places to grow and bind. The result is a denser, stronger internal structure, and more strength from every pound of cement.

At Pipedream, we first tested this approach in mixes where slag replaced half of the cement. The concrete reached the conventional mix’s 28-day strength in seven days and finished about 40 percent stronger than the same mix without the additive. Those results led a major hyperscaler to push for a pilot at 70 percent replacement, a level many engineers would consider impractical for everyday structural concrete. In our trials at that level, strength still rose by roughly a third. Results vary by mix and materials, and every application should be validated. But the direction is clear.

That extra strength becomes a resource. Producers can spend it on speed, on further cement reduction, or on a balance of both, depending on what a project values most. At 70 percent replacement, with the added strength used to trim total cement content further, embodied carbon reductions of more than 60 percent compared with conventional all-cement concrete come within reach. That is territory rarely reached in mainstream structural concrete, and it is available now.

The cement industry is working on its own solutions, from cleaner kiln fuels to new low-carbon cements to capturing emissions at the plant. That work matters, but it is slow and enormously expensive. It also runs up against a vast base of existing plants, built at great cost and designed to operate for decades.

That is why solutions that work inside today’s plants matter so much. There is a lot of foundational infrastructure already built in this industry. Materials that work with it, using ingredients producers already understand, can scale in months rather than decades. This is not a future technology. It can be used today.

Three changes would speed the shift across the industry.

First, specifications should require performance, not recipes. Many project specifications still demand a minimum amount of cement regardless of how strong the finished concrete is. That rule locks in carbon and punishes innovation.

Second, producers should publish product-specific Environmental Product Declarations, the verified disclosures that show a mix’s true carbon footprint. Without them, the best concrete looks no different on paper from the worst.

Third, the companies building AI’s infrastructure should use their purchasing power. Most have made ambitious climate commitments. Few buyers in history have controlled this much construction spending at once. If they made low-carbon concrete a condition of their contracts, the market would follow within a few years.

The data centers being built today will stand for decades. The carbon in their foundations is permanent from the day it is poured. We cannot capture our way out of emissions we choose to create. The cheapest, most certain ton of carbon is the one that never enters the air.

About the Author:

Dishank Patel is co-founder and CEO of Pipedream Industries, a materials science company developing nanotechnology for lower-carbon concrete. His work focuses on reducing cement use while supporting the strength and construction schedules that projects require. He advocates for practical approaches to cutting embodied carbon, including performance-based specifications, transparent environmental reporting, and materials compatible with existing production facilities.