NewsNTech
The utilization rate on deployed AI hardware has been the load-bearing argument for bears on the AI infrastructure trade.
The worry is that capital expenditure flowing into data centers and GPU inventory is running well ahead of workloads that can actually fill that capacity, leaving a large share of expensive silicon earning little.
CoreWeave, the GPU cloud provider, is pushing back on that thesis with a specific operational data point: demand for Nvidia chips that are six years old remains strong.
That detail is meaningful because it is the kind of utilization signal the bear case specifically predicts will not materialize.
Keep reading