Carbon Capture and the AI Power Boom: Could CCS Become a Critical Part of the Data Center Energy Strategy?

The scale of electricity required by AI means the data center industry is increasingly becoming an energy infrastructure industry. A hyperscale campus cannot operate around the availability of the sun or wind, nor can a multi-billion-dollar GPU cluster simply shut down whenever the electric grid becomes constrained. These facilities require extremely high levels of reliability, power quality and continuous availability. Renewable generation will unquestionably play an important role in meeting that demand, particularly because wind and solar can provide large quantities of fuel-free electricity at competitive costs. Battery energy storage will also become increasingly important for managing short-term fluctuations, providing grid services, smoothing large loads and shifting renewable electricity across hours. Nuclear energy, expanded transmission infrastructure, geothermal resources, fuel cells and other technologies could contribute as well. The difficulty is that none of these resources can currently solve every aspect of the problem in every market, particularly when developers are attempting to secure hundreds of megawatts within relatively short development windows.

That reality is one reason natural gas has returned to the center of the data center power conversation. Natural gas generation can provide dispatchable electricity, can be constructed relatively close to large loads, and benefits from an extensive U.S. production and pipeline network. Combined-cycle gas turbines can operate at high efficiencies and provide firm generation regardless of weather conditions, while reciprocating engines and other gas technologies can offer additional flexibility for behind-the-meter applications. The challenge, of course, is carbon emissions. The Rice study estimates that CO2 emitted from fossil-fueled power plants supplying data centers was already about 90 million metric tons in 2025 — and without new emissions-management strategies, that figure could climb to more than 404 million metric tons annually by 2030, a 4.5-fold increase. If hundreds of gigawatts of additional data center load ultimately drive significant construction of unabated natural gas generation, the AI revolution could substantially increase electricity-sector emissions at precisely the time utilities, corporations, and governments are trying to decarbonize the grid. Carbon capture and storage therefore deserves serious consideration not as an alternative to renewables or storage, but as another potential component of a much broader data center energy architecture.

Carbon capture and storage is relatively straightforward in concept, though considerably more difficult to execute. A power plant or industrial process separates carbon dioxide from other exhaust gases, compresses it into a dense state, transports it through pipelines or other infrastructure, and injects it deep underground into geological formations that can permanently contain it. Applied to natural gas combined-cycle generation, CCS makes it possible to retain many of the operational advantages of natural gas — dispatchability, scale, established fuel infrastructure, and the ability to locate generation near major electricity loads — while significantly reducing the amount of carbon dioxide released into the atmosphere.

For the data center industry, that possibility is important because the debate over powering AI is too often framed as a competition between renewable energy and fossil fuels. The actual challenge is much more complicated. A 500 MW or 1 GW AI campus needs energy, capacity, reliability, resilience, backup capability, power quality, and a development schedule that aligns with billions of dollars of computing investment. Solar generation may provide extremely economical electricity during daylight hours but cannot provide overnight capacity on its own. Wind can produce enormous amounts of inexpensive electricity but remains weather-dependent. Batteries provide extraordinary response speed and flexibility, but conventional lithium-ion systems are generally designed around hours rather than days of energy storage. Nuclear provides highly reliable, carbon-free generation, but new projects can face long development timelines and high capital requirements. Natural gas provides firm and dispatchable power but introduces fuel and emissions exposure. Rather than asking which of these technologies will "win," data center developers increasingly need to determine how to combine these resources into optimized energy systems. CCS could significantly improve the role natural gas plays within that portfolio.

One of the most significant elements of the Energy & Fuels analysis involves something that receives considerably less attention than power plants or data centers: the geology beneath the United States. Deep saline formations can potentially provide enormous capacity for permanently storing captured carbon dioxide. These formations exist thousands of feet underground and contain highly saline water unsuitable for conventional drinking or agricultural applications. When appropriate geological characteristics exist, operators can inject compressed CO2 into porous rock formations beneath impermeable caprock that prevents the gas from migrating back toward the surface.

The numbers behind that storage potential are striking. Saline aquifers account for more than 95% of total U.S. CO2 storage capacity across all subsurface reservoir types, with a combined theoretical capacity the study puts at roughly 11.37 trillion tons. Running a source-to-sink mapping exercise across the country, Lau and Tsai found that 34 states have enough in-state saline aquifer storage capacity to sequester more than 100 years of projected CO2 injection beyond 2030. In aggregate, the study estimates those aquifers could have stored 59 million metric tons — about 66% — of data-center-related CO2 emissions in 2025, with that capacity share growing to roughly 299 million metric tons, or 74%, by 2030 as more storage infrastructure comes online. Of the 16 states that don't have adequate in-state storage capacity, six could still access sequestration by sending CO2 to a bordering state.

That geographic relationship could eventually influence data center site selection in ways that are not widely considered today. Developers already evaluate utility capacity, natural gas pipelines, transmission infrastructure, renewable resources, fiber connectivity, water, taxes and available land. If natural gas generation with CCS becomes a significant strategy for supplying AI infrastructure, developers may eventually add CO2 transportation infrastructure and geological sequestration potential to that list. A location with abundant natural gas, available land, high-capacity fiber and suitable underground storage could offer a fundamentally different value proposition from a traditional data center market where electricity must be transported hundreds of miles through constrained transmission systems. Energy-producing regions could therefore gain another competitive advantage in the race to attract hyperscale computing investment.

The study's own data center growth projections point toward exactly these regions. Texas, Virginia, Pennsylvania, Ohio, Arizona, Colorado, Utah, and Illinois are identified as the states with the fastest-growing data center power capacity — and Texas alone is projected to need roughly 25 GW of additional power capacity by 2030 to meet that demand. The researchers specifically highlight Texas, Colorado, Louisiana, Mississippi, and Pennsylvania as states where data-center-related CO2 emission clusters sit particularly close to both natural gas reservoirs and saline storage formations, making them well positioned geographically for this strategy.

Texas is an obvious example. The state combines enormous natural gas production, significant pipeline infrastructure, world-class wind and solar resources, extensive energy industry expertise and rapidly expanding data center development. It also contains geological formations potentially suitable for carbon sequestration. An integrated Texas AI energy campus could theoretically combine wind and solar generation with battery storage, firm natural gas generation and carbon capture, using each technology for the function it performs best. Renewable resources could supply large quantities of inexpensive energy when available, BESS could manage short-duration variability and large computing-load fluctuations, while gas generation could provide firm capacity when renewable output declines. Carbon capture could then substantially reduce emissions from that dispatchable generation. Such a campus would effectively operate as both a data center and a sophisticated private energy system.

Perhaps the most important issue in today's data center market is not simply electricity price but time-to-power. A theoretically inexpensive utility tariff provides limited value if the utility cannot deliver the required capacity for five, seven, or even ten years because transmission lines, substations, or new generation must first be developed. AI companies are deploying extraordinarily expensive computing equipment whose economic value can decline rapidly as newer generations of processors become available. Delaying a multi-billion-dollar data center campus for several years can therefore represent an enormous opportunity cost. Increasingly, developers are asking not simply, "What will electricity cost?" but "How quickly can you deliver 300 MW, 500 MW or 1 GW of reliable power?"

That shift is driving renewed interest in behind-the-meter generation, microgrids, natural gas turbines, reciprocating engines, fuel cells, renewable generation and battery storage. Carbon capture could eventually become part of these integrated systems, particularly for large campuses where dedicated generation operates as long-term infrastructure rather than temporary backup power. A natural gas combined-cycle plant developed alongside a major AI campus and connected to CO2 transportation and sequestration infrastructure could potentially provide large quantities of firm electricity while substantially lowering the emissions associated with conventional gas generation. The economics would have to account for the additional capital requirements, parasitic energy consumption, transportation infrastructure and storage costs associated with CCS, but the calculation changes when electricity availability determines when billions of dollars of computing infrastructure can begin generating revenue. In that environment, speed, reliability, cost and emissions must be evaluated together rather than independently.

One of the most important conclusions from examining carbon capture in the context of AI infrastructure is that CCS does not need to displace renewable generation or battery storage to be valuable. In fact, the strongest future data center power architectures may incorporate several technologies simultaneously. Wind and solar can provide enormous quantities of electricity without fuel costs. Battery storage can respond within milliseconds, absorb excess renewable generation, support power quality, reduce peaks and provide short-duration energy shifting. Existing nuclear facilities can deliver exceptionally reliable carbon-free generation, while new nuclear technologies could eventually provide additional firm capacity. Utility grids provide geographic diversification and access to regional generation resources. Natural gas with carbon capture could then provide dispatchable generation when renewable production or grid capacity is insufficient.

This portfolio approach recognizes an important distinction between energy and capacity. A renewable project may generate enough megawatt-hours over a year to match a data center's annual consumption, but that does not necessarily mean electricity is available during every hour the data center operates. A hyperscale facility requires both sufficient annual energy and dependable capacity during periods of low renewable generation or grid stress. BESS can bridge shorter periods, while firm generation can address longer ones. The objective should therefore be to minimize emissions and cost while maintaining the extraordinary reliability demanded by digital infrastructure, rather than forcing every technology into a role it was not designed to perform.

This is also why carbon capture should be judged pragmatically. CCS has legitimate technical and economic challenges. Capture systems consume energy, reducing the generating facility's net efficiency. Compression and transportation infrastructure must be developed. Injection wells require extensive geological characterization, permitting and monitoring, while long-term liability and regulatory frameworks must be clearly established. The existence of theoretical underground storage capacity does not mean commercial sequestration projects can automatically be developed wherever a data center requires power. These challenges are substantial, but they are infrastructure challenges rather than reasons to dismiss the technology outright. Transmission lines, nuclear facilities, renewable projects, gas pipelines and utility-scale batteries all face their own permitting, financing and development constraints. The relevant question is whether CCS can provide enough emissions reduction and system value to justify those costs in locations where firm generation is required.

The relationship between energy availability and computing location may ultimately become a defining characteristic of the AI infrastructure boom. Traditional data center markets developed around locations such as Northern Virginia, Silicon Valley, Dallas, Phoenix and other areas offering strong fiber networks, business ecosystems and established utility infrastructure. But concentrating enormous new computing loads in the same markets creates increasingly difficult electricity challenges. Transmission congestion, substation limitations and generation shortages can make it extremely expensive and time-consuming to add hundreds or thousands of additional megawatts. AI developers may therefore increasingly consider moving computing infrastructure toward locations where energy can be developed more easily rather than continuously attempting to transport enormous quantities of electricity into already constrained markets.

Carbon capture could expand the number of locations that can support that energy-first strategy. Regions with abundant natural gas, existing industrial infrastructure, suitable geology, renewable resources and available land could potentially develop integrated energy campuses specifically designed around large computing loads. Parts of Texas, Pennsylvania, Ohio, Colorado and other energy-producing regions may therefore become increasingly attractive to AI developers. Similar opportunities could emerge around retired power plants, brownfield industrial sites and existing energy corridors where transmission, pipelines and other infrastructure are already available. Instead of asking utilities to retrofit century-old grid infrastructure around entirely new gigawatt-scale loads, developers may increasingly build those loads around existing and developable energy resources.

This transition would profoundly change the relationship between the technology and energy industries. A 1 GW AI campus is no longer simply a real estate development containing servers. It is effectively an industrial energy project that happens to contain computing infrastructure. Developers capable of integrating generation, storage, transmission, natural gas, carbon management, and computing requirements may therefore gain an increasingly important competitive advantage. AI-era data center developers may need to think more like power developers.

The significance of the research goes beyond carbon capture itself. The larger issue is the extraordinary amount of new electricity data centers could require over the remainder of this decade and beyond. Whether any individual forecast proves exactly correct matters less than the market direction: AI is creating enormous incremental demand at a time when the United States already needs to replace aging infrastructure, expand transmission, electrify additional sectors, and maintain grid reliability. Meeting all of those requirements simultaneously will require one of the most significant electricity infrastructure investment cycles in decades.

Renewable generation will be essential to that buildout because of its economics, scalability and lack of fuel costs. Battery storage will become increasingly important as renewable penetration rises and electricity loads become more dynamic. Nuclear power could become increasingly valuable because it can provide reliable, carbon-free electricity around the clock. Expanded transmission will be necessary to connect new generation with growing demand centers. Natural gas is also likely to remain part of the equation because of the scale of existing U.S. resources and the need for dispatchable generation. If that proves correct, then reducing the carbon intensity of natural gas becomes an important part of the broader AI energy strategy, and CCS becomes much more relevant than it might have appeared only a few years ago.

The opportunity should therefore be evaluated without turning carbon capture into either a universal solution or an ideological distraction. CCS will not eliminate the need for renewable energy, storage, transmission or nuclear generation, and it will not make every natural gas project economically or environmentally attractive. What it can potentially do is provide another pathway for supplying firm power while materially reducing emissions in locations where gas generation is likely to be developed regardless. In an electricity market facing unprecedented new load growth, having another commercially viable tool available could prove extremely valuable.

The AI revolution is forcing the United States to confront a basic infrastructure reality: digital growth ultimately depends on physical energy infrastructure. Every GPU, server rack and hyperscale computing campus requires electricity, and the enormous scale of projected AI development means the industry will need generation technologies capable of providing energy, capacity, reliability and resilience simultaneously. No single resource is likely to meet every requirement. The most successful data center energy strategies will therefore combine technologies based on their strengths rather than attempting to force one technology to solve the entire problem.

Carbon capture and storage could become an important component of that portfolio, particularly if natural gas generation expands alongside AI infrastructure. America's substantial geological storage resources make it possible to combine domestic natural gas, dispatchable power generation, and permanent CO2 sequestration while renewable generation, BESS, nuclear, and expanded transmission continue scaling. Significant technical, regulatory and economic hurdles remain, and CCS should not be treated as a substitute for cleaner generation when those resources can meet the same requirements economically. But neither should it be dismissed simply because it allows continued use of natural gas.

The more important question is how quickly America can build a power system that can support the AI economy without abandoning its longer-term emissions objectives. If carbon capture can become commercially scalable, if sequestration infrastructure can be developed in the right locations and if natural gas generation with CCS can complement rather than compete with renewables, storage and nuclear power, carbon management could become another important piece of the energy infrastructure behind artificial intelligence.

Chris Kalowes, founder of WattThe?! — 15+ years in utility-scale battery energy storage (BESS), renewable energy, and AI infrastructure, across utilities, IPPs, EPCs, developers, and technology providers.

Source: Hon Chung Lau and Steve C. Tsai, "The Role of Carbon Capture and Storage in Decarbonizing U.S. Data Centers," Energy & Fuels (American Chemical Society), 2026. DOI: 10.1021/acs.energyfuels.6c01309.