A fresh forecast, highlighted by New Scientist, has cast doubt on the ability of the AI industry to hit widely publicized net zero pledges by 2030. As companies including Google, Microsoft and Meta accelerate investment in high performance computing, researchers say the physical scale of the required data centre expansion in the United States will create significant environmental pressures that corporate commitments alone may not address.

The forecast estimates that the US AI server buildout could demand between 731 million and 1.125 billion additional cubic metres of water by 2030. It also projects annual emissions equivalent to between 24 million and 44 million tonnes of carbon dioxide from the incremental infrastructure. Those totals are sensitive to three variables, the report notes: the pace of AI demand growth, how many high end servers can actually be manufactured and deployed, and where new data centres are sited in relation to power sources and regional water availability.

Data centres consume water and energy primarily for cooling and power delivery at the scale required for modern large scale AI training and inference. If new facilities are concentrated in regions with carbon intensive electricity or limited water supplies, the forecast cautions that corporate net zero targets could be undercut even as data centre operators invest in efficiency and renewable procurement. Conversely, siting new capacity near low carbon grids and implementing water recycling could mitigate some impacts, but would require coordinated planning and timely investment.

The scale of the projected carbon emissions is significant when compared with national inventories and sectoral reductions needed to meet climate goals. Even at the lower bound of the forecast, the incremental emissions would absorb a meaningful share of remaining carbon budgets if left unmitigated. Water use on the scale projected would also matter most in arid regions where new data centres are increasingly being proposed, raising questions about competition for scarce resources between industry, agriculture and communities.

The forecast underscores a gap between headline corporate commitments and the practical challenges of decarbonising a rapidly growing, technology intensive sector. Efficiency improvements in chip design and data centre cooling can reduce per unit energy consumption, and renewable energy procurement can lower operational emissions. But the report suggests those measures may not be sufficient on their own given projected demand growth, and that supply chain constraints on high end server manufacturing could also shape outcomes.

Policymakers face choices about how to regulate siting, require transparency in emissions and water accounting, and coordinate power system upgrades. Regional planning that aligns data centre expansion with grid decarbonisation and water availability could help, as could incentives for technologies that reduce both energy and water footprints.

The forecast paints a sobering picture for an industry whose growth is redefining modern computing. Meeting 2030 net zero commitments will likely require faster adoption of low carbon power, aggressive improvements in efficiency, and a shift in planning practices to avoid concentrating environmental impacts in vulnerable regions. Without those changes, the physical realities of scaling AI infrastructure may prove a formidable obstacle to the targets companies have set for themselves.