Data centers and artificial intelligence (AI) innovations are developing quickly, alongside growing environmental and social concerns related to their expansion. Conversations are now shifting to sustainability and responsible data center advancement, which is envisioned to include carbon dioxide removal (CDR) as an impact mitigation tool.
In a “big picture” perspective, data center plans and projects are already reshaping economies, posing challenges with their high demand for energy, water, land, and materials.
According to estimates by the International Energy Agency (IEA), in 2025, data centers used 485 terawatt-hours (TWh) of electricity worldwide. This number is slightly above the 451 TWh used by the country of France during that same year, and the IAE projects the figure could jump to about 950 TWh in 2030.
Studies from the Environmental and Energy Study Institute also warn that larger data centers can consume up to 5 million gallons of water per day, which translates to roughly 1.8 billion gallons a year, a volume that would cover the yearly water need of roughly 16,000 average American families.
Besides being resource-heavy, data centers are also under pressure due to emissions. As one of the companies at the forefront of AI development, Microsoft reported a 25% increase in its 2025 fiscal year greenhouse gas emissions, which was explained by the corporation’s expanding data center infrastructure and pausing the use of non-additional, unbundled renewable energy certificates.
With the AI ball already rolling, official emissions regulations taking shape, and a worsening global climate, the attention is now pinned on climate-conscious AI development. In that regard, CDR is increasingly becoming part of the solutions proposed, not only as procurement but also as an integration.
If carbon removal is to be integrated, which pathways are suitable for AI facilities? Could CDR make these tech megastructures climate-forward, and what are the challenges in such scenarios? When CDR is procured, are data centers the new key buyers that can unlock scale? We reached out to experts to find out.
Which carbon removal technologies could be paired with AI data centers?
Dr. Colin McCormick, Chief Innovation Officer at Relae (formerly known as Carbon Direct), echoes a widespread industry understanding when he says that the carbon removal technology best suited to pair with AI infrastructure is direct air capture (DAC).
An AI DAC integration offers a real opportunity to share airflow and waste heat, he says, but it can come with serious engineering trade-offs, such as added strain on cooling systems, the cost of heat-transfer equipment, extra power demand for CO2 compression, and complicated logistics for storing or transporting the captured CO2.
“Both DAC and data centers need to move a lot of air. Placing DAC filters in the airflow of these data centers might “double up” on this airflow, using it for both cooling and CO2 removal, reducing costs. However, putting a DAC filter in the existing airflow of a data center would create airflow resistance and would make those air handlers or air-cooled chillers have to work harder, consuming more power and shortening their life,” Dr. McCormick explains.
Additionally, DAC uses heat energy to repeatedly “regenerate” its CO2 filters, like wringing out a towel, and data centers produce a lot of waste heat, so it might be possible to use this waste heat for DAC. However, this isn’t free, as it requires new equipment to concentrate and transfer heat from the existing data center cooling system (e.g., chillers) to the DAC equipment.
Furthermore, the CO2 removed from the DAC airflow is purified and compressed in a process that takes significant electrical power. Data center facilities that are already power-constrained may not have enough extra power for it.
Dr. McCormick also points out that the logistics of managing purified CO2 aren’t straightforward. Very large data centers equipped with DAC might produce enough to justify a dedicated sequestration well, and only the largest data centers would qualify to invest in a direct connection to an existing CO2 pipeline, while smaller projects would need to store CO2 in on-site tanks and schedule trucks or railcars to send it to storage.
In his opinion, storing removed carbon by building with mass timber or with cement and concrete blends that lock in atmospheric CO2 could offer a modest data center integration opportunity.
DAC designed with AI in mind
On the developer side, DAC companies are already working on AI-tailored offerings.
Avnos is one of the companies with a DAC-oriented solution, marketed as Hybrid DAC (HDAC) technology. These systems use low-grade waste heat to reduce the electrical demand of both data centers and the HDAC units, producing clean cooling water and redistributing heat across the infrastructure.
HDAC slashes the energy use by more than half when compared to heat-based DAC technology, according to its producer, and it improves the operational resilience through opportunistic electrical load utilization.
This month, the company launched the Avnos HDAC™ Module, its first standardized, factory-built product family that moved this technology into commercial manufacturing. As initial target clients for this innovation, Avnos lists data centers, carbon storage initiatives, and E-fuels.
What should a pilot data center-CDR integration look like?
From the perspective of AJ Simon, Senior Director of Industrial Decarbonization at Relae, a well-designed AI–CDR pilot would focus on where the two industries meet, testing waste heat use, DAC’s effects on data center airflow and energy, CO2 handling, and local water impacts.
Before such a project could be scaled, he states, it would need community support, limits on added environmental or nuisance burdens, and tangible local benefits like jobs and a share of carbon credit revenue.
“There is plenty of innovation happening in data center design and DAC technology development. Therefore, an AI-CDR pilot program should be designed to address the key challenges and opportunities at the interface between these two industries,” he says.
Such work is already picking up on the ground, and Microsoft is one of the companies currently running a DAC integration at a data center site, with a pilot deployment named DACinDC.
Relevant: Microsoft Runs New DAC System Powered With Waste Heat From Its Data Centers
The system has been integrated directly into the tech corporation’s existing data center infrastructure, where it operates on waste heat generated from computing workloads, optimizing the DAC performance and efficiency via materials developed with the help of AI.
Other companies also exploring such integrations include Meta and Amazon, among others.
Is AI the new source of long-term CDR demand?
In a recently published whitepaper, Climeworks points to procurement when talking about CDR integration in AI, listing it along with resource efficiency, clean energy, and low-carbon materials as the four main pillars of responsible AI development.
CDR procurement, the whitepaper outlines, can be done via four models, including creating regional impact portfolios, ecosystem co-procurement, natural gas transition guardrails, and in the build stage emissions neutralization.
Pathways that should be procured through the proposed models include biochar, bioenergy with carbon capture and storage (BECCS), direct air capture and storage (DACCS), enhanced rock weathering (ERW), and afforestation, reforestation, and revegetation (ARR).
Data center frontrunners Microsoft, Google, Amazon, and most recently Athropic are already purchasing carbon removals, and this trend is expected to continue as data center growth accelerates.
Read more: Google, Anthropic And Other Tech Giants Commit $915M To Scale Carbon Removal
Milkywire, known for its CDR catalyzing efforts, recently announced interest in potentially procuring CO2 removals from solutions integrated within data center infrastructure, which successfully unlock a clear technical or economic advantage for durable CDR.
Another strong CDR purchasing signal was provided by A Healthier Earth with the introduction of the world’s first integrated carbon removal platform from the data center sector. This offering provides hyperscalers, global corporates, and institutional buyers across Europe and beyond with a scalable and financeable supply of high-integrity biochar and carbon credits.
Talking about how significant of a CDR demand source AI infrastructure could become in the following years, Dr. McCormick says that because today’s DAC capacity is so small, very large data centers could meaningfully expand it. But the removal any single site achieves, he stresses, would be minor compared to its own emissions if it runs on unabated natural gas, driving up future need for carbon removal.
The most realistic pairing with relevant scale is DAC, he says, and current global DAC deployments are modest, with only a handful of facilities currently operating or scheduled to commence in the near future.
Looking ahead
While CO2 removal technologies are being included in data center climate impact neutralization scenarios, it becomes evident that there are numerous considerations and uncertainties when it comes to physical CDR integrations within AI infrastructure.
Direct air capture is cited as the most viable option, and even as such, this technology comes with challenges that need further evaluation. As the interest in enhancing data centers with CDR increases, it will most likely fuel advanced research and testing in that direction.
On the other hand, data centers and AI facilities could address hard-to-abate emissions through CDR procurement while providing carbon removal projects with the demand leap needed for scaling deployment.
Carbon removal is only one of the components of a more complex approach to making data centers sustainable, and it needs to be combined with other measures such as clean energy, resource efficiency, and low-carbon materials.
Beyond ensuring the environmental safety of the data center expansion, developers also need to consider communities and act in alignment with their needs and well-being.
