Citation Bureau
Vol. I
No. 383
XVI SEPTEMBER MMXXVI

Amazon plans to sell Trainium chips externally, and increased US power supply has given hyperscaler chips time to mature and compete with Nvidia.

The case

Four of Nvidia's major customers are now producing their own chips.

“Four of his customers now are producing their own chips. So I think OpenAI just announced the Halapino chip which is supposed to be twice as efficient in terms of what's per inference.”
Matt Barrie · 10 Sep 2026

The US bringing more power online than expected has delayed power scarcity becoming Nvidia's moat and given hyperscaler custom chips like TPU and Trainium more time to improve and compete.

“I think it's been probably the biggest problem for Nvidia over the last couple years is I think the US has actually brought a lot more power online than expected.”
Patrick O'Shaughnessy · 18 Aug 2026

Chinese AI company Z.AI is operating a 1-gigawatt data center running entirely on Chinese semiconductors with no Nvidia.

“There's there was something out just today that the Chinese Z.AI have some giant 1 gigawatt data center that's running entirely on Chinese semiconductors. No Nvidia.”
Luke Gromen · 23 Jul 2026

Topics

AI ChipsAI HardwareCloud ComputingHyperscalersSemiconductors

Citation Bureau · compiled from attributed public discussion. Last updated 2026-09-10.