🦔A Cast AI analysis of roughly 23,000 Kubernetes clusters found average GPU utilization across enterprises sits at 5%, meaning 95% of provisioned GPU capacity is idle. CPU utilization averages 8% and memory 20%. Companies are overprovisioning out of fear of missing allocations rather than sustained demand. The CEO of Cast AI described it plainly: companies are overbuying GPUs out of fear of missing out. My Take Last night Meta, Amazon, Microsoft, and Alphabet all reported earnings beating expectations while simultaneously raising AI infrastructure spend. Meta alone raised its 2026 capex guidance to $125-145 billion. GPU spot prices are up 48% in two months. CoreWeave raised rates 20%. The narrative driving all of that is insatiable AI demand that companies cannot afford to miss. Against that backdrop, 95% of provisioned enterprise GPU capacity sitting idle is a figure I find genuinely difficult to square with the shortage narrative. The price increases and multi-year infrastructure commitments are being driven by demand signals from companies that purchased GPUs on the same fear-of-missing-out logic and are now leaving 95% of that capacity unused. Companies are paying for GPUs they aren't running while bidding up the price of GPUs they claim they urgently need. The hyperscalers booking record backlogs last night are counting commitments from the same enterprises sitting on idle infrastructure, which is a different thing from collected revenue, and the distinction is important when the bill for all of it eventually comes due. Hedgie🤗