BloombergNEF forecasts that AI workloads could push US electricity demand from data centres to 20% by 2035, raising urgent questions about grid resilience, infrastructure investment, and environmental impact.
The latest warning for the US power system is not coming from OPEC or Tehran, but from Silicon Valley and the cloud. A new BloombergNEF forecast says data centres could account for 20% of US electricity consumption by 2035, up from 5.9% today, a surge driven overwhelmingly by artificial intelligence workloads. The firm now sees demand reaching 194 gigawatts, sharply above its December 2025 estimate, underscoring how quickly the computing build-out is outpacing earlier expectations.
That matters because AI facilities do not behave like ordinary industrial users. They run continuously, drawing large amounts of electricity for training and inference around the clock. BloombergNEF says nearly half of the future demand growth will be tied directly to AI, and that the US is on course to host the majority of global AI chip-related power usage by 2033. In practical terms, the issue is no longer whether the grid can absorb more load, but how much new generation and transmission can be built before bottlenecks start to bite.
The strain is already visible in several regions. Utilities in places such as Virginia, Texas and Illinois are facing heavier load growth, longer equipment lead times and sharper pressure to expand local infrastructure. Some of the costs are already filtering through to households, with consumer advocates warning that ratepayers are being asked to help fund new substations, transmission lines and other upgrades linked to large data centre projects.
That has turned electricity pricing into a political problem. President Donald Trump has expanded a voluntary “ratepayer protection pledge” involving technology companies, utilities and state leaders, with the stated aim of preventing data centres from pushing costs on to ordinary customers. But critics argue the pledge has little force because electricity prices are still set largely by state regulators and market conditions. In some areas, bills are still rising quickly despite the initiative.
At the same time, the scale of the expansion is changing corporate energy strategy. Microsoft, Amazon, Google, Oracle and Meta are all reported to be backing nuclear power, geothermal schemes, battery storage and other forms of firm generation in an effort to secure reliable supply without overloading local grids. OpenAI has also unveiled a 3.2-gigawatt data centre project near Savannah, Georgia, with phased power delivery beginning in 2028, and has sought to reassure residents that it will manage consumption during peak periods.
The broader investment message is clear. The AI boom is no longer just a story about chips and software; it is becoming a story about transformers, substations, generation assets and grid resilience. For industrial decarbonisation specialists, that shift is crucial. Meeting AI demand without driving up emissions will require faster deployment of low-carbon baseload power, better transmission planning and far more disciplined load management than the US grid has historically delivered.
What once looked like a distant infrastructure challenge is now moving into the centre of the energy debate. If BloombergNEF’s forecast proves right, the next major stress test for the US economy may come not from petrol prices, but from the electricity needed to keep artificial intelligence running.

