‘You’ve taken us free cash flow negative as a company, so congratulations’: Amazon has plenty of Nvidia GPUs, but not enough energy to power its AI boom
Amazon Web Services is struggling with energy constraints despite having ample Nvidia GPU hardware to support its AI ambitions. AWS received approximately $5 billion worth of Nvidia GB300 racks in August alone, yet faces logistical challenges in powering the infrastructure. According to internal comments, the massive GPU procurement has pushed Amazon’s free cash flow into negative territory.
The supply chain strain extends beyond AWS: Nvidia’s consumer-facing RTX 5090 GPU has become difficult to find through legitimate retailers, with resale prices approaching $9,000 on secondary markets. Nvidia is deliberately shifting production resources toward enterprise AI accelerators like the GB300 to meet massive demand from cloud providers. This reallocation has left PC gamers unable to purchase high-end consumer GPUs.
The core issue is infrastructure capacity: deploying cutting-edge AI hardware requires substantial electrical power, cooling systems, and facility upgrades that take time to implement. AWS’s acquisition of billions in GPU hardware has outpaced its ability to physically deploy them in operational data centers. The energy constraints highlight a critical bottleneck in the AI infrastructure buildout—companies can acquire the chips, but powering and cooling them remains a significant challenge that may slow the pace of AI service expansion.