AI Hyperscalers are currently spending 60% of their operating cash flow on capex (www.apolloacademy.com)

A new Apollo Global Management presentation highlights that hyperscalers are now dedicating a record-high 60% of their operating cash flow to capital expenditure. In plain terms, the biggest cloud and AI platform providers are pouring the majority of their cash into building out data centers and buying AI-specific hardware—GPUs/TPUs, high-bandwidth memory, networking, power and cooling, racks, and custom silicon—to support massive model training and inference workloads. For the AI/ML community this signals both opportunity and constraint: it underpins continued access to vast centralized compute for large-model research and production, while driving intense demand (and pricing power) across semiconductor, memory, and datacenter infrastructure supply chains. The heavy capex tilt increases barriers to entry, favors scale economies of major cloud providers, and pressures margins—potentially accelerating trends toward hardware-software co-design, efficiency-focused models, and differentiated managed AI services. It also raises risk of overbuild or supply tightness that could influence cloud pricing and chip availability. Note that the figure comes from Apollo’s presentation and forward-looking estimates are subject to change, but the underlying trend of capital-intense AI infrastructure expansion is clear and consequential for the ecosystem.