The GenAI GPU shortage has already led to a surge in demand, higher costs and reduced availability. However, another urgent problem looms: data centers are running out of space and power. This is especially problematic for small businesses that offer high-performance computing (HPC) colocation services and find that today’s data centers are overloaded.
A recent report from JLLa real estate investment and management company, highlights that AI-driven growth is expected to continue, with data generation expected to double over the next five years.
Furthermore, data center storage capacity is expected to grow from 10.1 zettabytes today to 21.0 zettabytes in 2027, requiring more data centers. Generative AI’s power needs, estimated at 300 to 500+ megawatts per campus, will also necessitate more energy-efficient designs and locations.
The electricity grids are reaching capacity
According to the report, the design of AI-dedicated data centers differs significantly from conventional facilities, requiring operators to plan, design and allocate energy resources based on the type of data processed or the stage of GenAI development. With the massive increase in GPUs, existing heat dissipation standards will be surpassed, leading to a shift from traditional air-based cooling methods to liquid cooling and backdoor heat exchangers.
Speak with HPC wireAndy Cvengros, managing director of US Data Center Markets for JLL, emphasized the importance of planning. He explained that power grids are reaching capacity and transformers have a lead time of more than three years, making innovation necessary. The GPU crisis is impacting small colocation deployments of 4 to 5 racks, which are finding it increasingly difficult to secure data center space due to the demands of hyperscalers.
Cvengros also emphasized that all major metropolitan areas are essentially maxed out, making secondary areas such as Reno, NV or Columbus, OH prime locations for new data center construction. However, it is expected that demand will continue and that new data centers will be added in 3.5 years.
The global demand for GenAI energy offers both opportunities and challenges. Finding GPUs for HPC is only half the problem; as HPCwire notes, where to connect them can become more challenging. This issue is particularly challenging for smaller operators, who may be driven out of the market by competition for raw materials.