By Adrian Siegrist
The infrastructure needed to meet rising demand for artificial intelligence is being designed, financed and built today. Data centers are not temporary installations – the steel, hardware and power contracts can be decades-long commitments.
Much of the debate rightly focuses on the immediate pressures associated with data centres: electricity demand, grid constraints, water consumption and local impacts. Efficiency, clean energy and lower-carbon construction materials must remain the foundation of responsible development.
But these measures cannot eliminate every tonne of carbon associated with the AI build-out. Even with expected progress on cleaning up power supply, steel, concrete, and computing hardware all carry embodied carbon. Once facilities become operational, further emissions arise where fossil fuels remain part of the energy supply.
Carbon dioxide removal is not yet a standard practice in data centre development. The developers, hyperscalers and tenants who are early movers in embedding removal into the development journey will not only manage their emissions more credibly, but also shape best practices for responsible AI development. For many companies, carbon removal remains a sustainability purchase made after emissions are calculated. AI infrastructure requires a more integrated approach. Removals can be proactively integrated at key decision points that lock-in emissions, including during site selection and permitting, construction, hardware procurement, power supply and tenant contracts.
Shared accountability across the value chain (among developers, hyperscalers, utilities, investors, contractors, hardware manufacturers and tenants) can be a powerful procurement tool ee propose four practical models to integrating carbon removal into the development journey as a starting point.
Connect carbon removal with regional priorities
Data-centre development is not only a technical challenge. It is also a local one.
Communities are asking legitimate questions about land, water, energy, employment and environmental impact. Carbon removal cannot replace meaningful consultation or compensate for poorly managed local impacts. But it can contribute to a broader community strategy when projects are selected for measurable benefits that reflect regional priorities.
Depending on local conditions, carbon-removal projects may support soil quality, water retention, biodiversity, agriculture or economic development. Developers should engage communities early and report transparently on both carbon removal and wider outcomes.
The objective should not be to purchase support, but to connect global climate responsibility with credible local value. This should never be presented as a guarantee of planning approval or community acceptance.
Procure across the AI value chain
No single company controls the full carbon footprint of a data-centre campus. The associated emissions sit across multiple organisations, contracts and reporting boundaries.
Hyperscalers, developers, utilities, contractors, tenants and hardware providers could therefore aggregate demand for carbon removal through shared portfolios, where the cost could be incorporated into commercial agreements rather than being left to each participant to manage independently.
Pooling demand could simplify procurement, provide longer-term signals to carbon-removal projects and distribute responsibility more proportionately across the value chain.
Collaboration must not, however, weaken accountability. Participants need clear contractual rules covering the allocation of removals, the retirement of credits and the claims each organisation may make. Without them, co-procurement risks creating double counting rather than shared responsibility.
Put strict guardrails around natural gas
Pressure to bring computing capacity online quickly has led some developers to consider on-site natural gas when grids cannot provide sufficient power.
This should never be the preferred option. Low-carbon electricity, storage, geothermal energy and other cleaner solutions must be prioritised. Where natural gas is nevertheless used temporarily, the conditions should be defined before operations begin.
These should include a clear and time-bound transition away from fossil-based generation. Carbon capture and storage should be paired with deployment to address the bulk of direct emissions, while durable carbon removal can cover the residual emissions it does not prevent.
Carbon removal alone cannot turn a fossil-fuel project into responsible infrastructure, however, with the proper guardails it can support a a transparent, time-bound transition plan.
Address embodied carbon before operations begin
A substantial part of a data centre’s carbon footprint can already be determined during the design and construction stages.
Emissions from building materials and hardware can be estimated before a facility becomes operational. Developers can therefore establish a carbon baseline and corresponding carbon-removal budget during planning.
Carbon removal can then be treated as a project cost alongside construction, equipment and other capital expenditure. Initial estimates can be refined as better data becomes available.
Reducing embodied emissions must still come first. But combining lower-carbon materials and efficient equipment with durable removal provides a more complete way to manage the carbon already committed through construction and hardware decisions.
The time to start is now
The AI sector is making long-term infrastructure decisions now. Carbon removal belongs in those decisions, not years later as compensation for their consequences.
The author is Chief Commercial Officer at Climeworks
Read the whitepaper on ‘Responsible AI Infrastructure Should Include Carbon Removal’ by Climeworks here.
