2 days ago
11
The rapid build-out of AI systems is increasing data centre power demand, exposing a less visible constraint in cloud computing: access to power.
A recent agreement between Oracle and Bloom Energy means the two intend to deploy up to 2.8 gigawatts (GW) of fuel cell capacity to support data centre growth, according to a statement from Bloom Energy.
AI systems consume far more energy than traditional workloads. Model training and inference requires clusters of GPUs running for long periods, raising electricity use and adding to cooling demands. According to Barron, Oracle is working with Bloom Energy to deploy fuel cells as part of its data centre strategy.
The company did not outline specific deployment timelines. The International Energy Agency (IEA) has noted that data centre electricity use has been rising faster than overall power demand in recent years.
In parts of the US and Europe, utilities have warned that they may not be able to meet demand from new data centre projects without upgrades to infrastructure. Delays in grid connections can slow down expansion plans. Research from the IEA and Harvard’s Belfer Centre for Science and International Affairs indicates that new transmission lines and substation upgrades are often required before additional load can be connected, with timelines that can extend several years.
Fuel cells can provide steady power and can be deployed in modular units. Industry reports indicate that modular systems allow operators to add capacity incrementally with data centre build-outs. Solid oxide fuel cells, the type produced by Bloom Energy, generate electricity through an electrochemical process not combustion. These systems typically reach electrical efficiency of around 54% to 60%, comparable to large gas-fired power plants, while avoiding transmission losses by producing power on-site. Bloom Energy has said combined heat and power configurations can push total efficiency above 80%.
In many cases, cloud providers are now competing for power in the same way they compete for hardware. Access to electricity can influence where data centres are built and how quickly they come online. In some markets, power availability is becoming a primary factor in site selection, with land and connectivity.
On-site generation is becoming more common, as are renewable projects and long-term power supply agreements. Fuel cells offer one option. Some companies are exploring nuclear power, while others are investing in solar and wind.
Fuel choice affects both cost and emissions. Most current fuel cell deployments rely on gas, while hydrogen remains at an earlier stage of adoption due to higher costs and limited infrastructure. IEA hydrogen reports note that infrastructure and supply constraints continue to limit large-scale adoption. Costs vary depending on fuel and contract structure.
Energy use in data centres raises environmental concerns which are likely to grow if AI workloads expand. Renewable energy projects remain a important part of many cloud strategies. Long-term agreements for solar and wind power are common, though these sources can be less predictable than on-site generation.
Reliable base load power may come from fuel cells or other steady sources, while renewables can help offset emissions over time.
Compute capacity is still important, but it is not the only factor that limits growth. Energy supply is emerging as a important constraint. By investing in on-site power generation, the companies are addressing a problem that sits outside traditional cloud architecture.
(Photo by Alexandre Viard)
See also: Why Google is building more data centres in Asia
Want to learn more about Cloud Computing from industry leaders? Check out Data Centre Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
CloudTech News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.