Forget about AI GPU scarcity: data center operators may have to wait a whopping three years to get key component — and this may be killing off competition by driving smaller operators out

Key Takeaways:

– The GenAI GPU scarcity has led to increased demand, higher costs, and reduced availability.
– Data centers are running out of space and power, especially for small companies providing high-performance computing colocation services.
– AI-driven growth is expected to continue, with data generation predicted to double in the next five years.
– Data center storage capacity is projected to grow significantly, requiring more data centers.
– The power demands of generative AI will necessitate more energy-efficient designs and locations.
– The design of AI-specialized data centers differs from conventional facilities, requiring operators to plan and allocate power resources accordingly.
– Traditional air-based cooling methods will be surpassed, leading to a shift towards liquid cooling and rear-door heat exchangers.
– Power grids are reaching capacity, and transformers have long lead times, requiring innovation.
– The GPU squeeze is affecting small colocation deployments, making it difficult to secure data center space.
– Major metro areas are reaching capacity, making secondary areas prime locations for new data center construction.
– The global GenAI energy demand presents opportunities and challenges.
– Finding GPUs for HPC is only half the problem, as finding suitable power resources may be a bigger challenge.
– Smaller operators may be driven out of the market due to competition for resources.

TechRadar:

The GenAI GPU scarcity has already sparked a surge in demand, increased costs, and reduced availability. However, another pressing issue is looming: data centers are running out of space and power. This is particularly problematic for small companies providing high-performance computing (HPC) colocation services, who are finding current data centers maxed out. 

A recent report from JLL, a real estate investment and management firm, highlights the AI-driven growth is expected to continue, with data generation predicted to double over the next five years. 

Source link

AI Eclipse TLDR:

The GenAI GPU scarcity has caused a surge in demand, increased costs, and reduced availability. However, another issue is emerging: data centers are running out of space and power. This is especially problematic for small companies offering high-performance computing (HPC) colocation services, as current data centers are reaching their maximum capacity. A recent report from JLL, a real estate investment and management firm, predicts that AI-driven growth will continue, resulting in a doubling of data generation over the next five years. Additionally, data center storage capacity is projected to grow from 10.1 zettabytes to 21.0 zettabytes in 2027, requiring the construction of more data centers. The power demands of generative AI, estimated at 300 to 500+ megawatts per campus, will necessitate more energy-efficient designs and locations. The design of AI-specialized data centers differs significantly from conventional facilities, requiring operators to plan and allocate power resources based on the type of data processed or stage of GenAI development. The increased use of GPUs will exceed existing standards for heat removal, leading to a shift from traditional air-based cooling methods to liquid cooling and rear-door heat exchangers. Andy Cvengros, managing director of U.S. Data Center Markets for JLL, emphasizes the importance of planning due to power grids reaching capacity and transformers having lead times exceeding three years. This GPU scarcity is particularly challenging for small colocation deployments of 4-5 racks, as they struggle to secure data center space amidst the demands of hyperscalers. Major metro areas are already saturated, making secondary locations like Reno, NV, or Columbus, OH, prime choices for new data center construction. However, the demand is expected to continue, and new data centers are projected to take 3.5 years to be operational. The increasing global demand for GenAI presents both opportunities and challenges, with the availability of GPUs being only one aspect of the problem. Finding suitable locations to accommodate these GPUs may become an even bigger challenge, especially for smaller operators who may be pushed out of the market due to resource competition.