AWS has teamed up with  Orbital Materials, which uses artificial intelligence to create a novel material specifically designed to extract carbon from the hot air exhaust of data centres.

Orbital Materials CEO Jonathan Godwin thinks this process will be more cost-effective than purchasing captured carbon to offset the impact of climate change.

This system being developed at an as-yet-undisclosed data centre location operates by drawing in outside air to cool the extremely hot semiconductors that power or train robust AI models, such as Anthropic’s Claude chatbot.

Direct air carbon capture using cooling mechanisms in data centres is gaining traction, with tech giants Alphabet and Meta investing in research and development in this domain.

Data centres that power generative AI models are notably energy-intensive, generating more heat and consuming more electricity than those running conventional websites or databases.

As the demand for advanced AI tools escalates, there are growing concerns that AI data centres will inflate electricity prices and increase fossil fuel consumption. This issue is compounded as companies consolidate more processing power into single data centres, or “clusters,” necessitating more sophisticated cooling systems.

Cooling mechanisms are designed to dissipate heat from the chips and expel the hot air from the data centre. “Sorbents” can absorb carbon dioxide as air passes over them. However, the exhaust from data centres is hotter than that in traditional direct air capture methods. Orbital Materials used AI to predict molecular structures better suited for absorbing hotter air, testing various possibilities in a New Jersey lab.

Once separated from the air, carbon can be stored underground as gas if the local geology permits or converted into a solid, such as cement, and transported elsewhere. Godwin stated that companies are employing Orbital’s AI models to discover new materials that could enhance everything from batteries to semiconductors.