Amazon expects to receive $5 billion of Nvidia GB300 racks before deployment
Amazon is accelerating AI hardware acquisition through an Early Rack Receipt process. Internal emails reviewed by Business Insider say the cloud unit expects roughly $5 billion in Nvidia GB300 server racks to arrive at U.S. facilities before final deployment, and projects about $3 billion more in GB300 racks each month for the foreseeable future.
Key points
- Amazon expects about $5 billion in Nvidia GB300 server racks to arrive at US sites before they are installed.
- The company projects roughly $3 billion in additional GB300 racks each month for the foreseeable future.
- Power shortages have left some racks idle in warehouses, creating “dark GPUs” and adding cash‑flow pressure.
Some AWS sites lack sufficient power, so finished racks sit in intermediary warehouses. This creates “dark GPUs,” where expensive chips remain idle until electricity capacity is available. The power bottleneck adds financial pressure; Amazon forecasts about $220 billion in capital expenditures this year and its free‑cash‑flow turned negative last quarter. The early‑rack program lets AWS secure scarce GPUs ahead of competitors but ties up billions before the equipment can generate cloud revenue. An Amazon spokesperson said staging hardware near deployment sites is standard practice and disputed claims of lacking a clear path to install the racks. CEO Andy Jassy reiterated confidence in AI investments in the annual shareholder letter.
Amazon is buying billions in Nvidia AI systems it can't yet turn on
businessinsider.com · 19 October 2026Loading the full article…
This text was published by businessinsider.com and written by Eugene Kim. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
2sources- Amazon stockpiles Nvidia AI hardware ahead of power availability -Business InsiderPress · finance.yahoo.com ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Intel and AMD collaborate on oneDNN optimizations for PyTorch · 1 src
- DeepSeek open-sources DeepInfer stack, potentially lowering inference costs · 5 src
- Infinity and HP double AI workload capacity on NVIDIA GB300 superchip · 1 src
- AMD adds FSR 4.1 support for RX 6000 GPUs, Xbox Series X/S and Ryzen AI chips · 1 src
- Google open-sources TorchTPU for native PyTorch on TPUs · 4 src
Comments
via GitHub Discussions