Enterprises Overprovision GPUs, Leaving 95% Idle Capacity

As enterprises overprovision GPUs in preparation for AI demand, a phenomenon has emerged where actual GPU utilization stays at around the 5% level.

This GPU Hoarding phenomenon causes the following problems:

• Worsening supply shortage: By companies securing capacity in advance, other companies' access to GPUs is reduced, leading to an overall market supply shortage.
• Wasted costs: Companies must pay high costs for reserved capacity regardless of actual usage, raising operational costs.

Within the industry, there are calls for the need to address this inefficiency through the adoption of a pay-per-use model, or through technical methods to increase the utilization efficiency of existing GPU resources rather than building new data centers.