Overview
- Analysts still peg Nvidia at roughly 80%–90% share of AI accelerators, signaling entrenched dominance.
- Google’s latest flagship model was trained on in-house TPUs that The Times reports cost about half as much as Nvidia’s chips.
- AMD secured a 6-gigawatt compute deal with OpenAI, compared with Nvidia’s previously reported 10-gigawatt agreement with the same customer.
- Yahoo Finance cites reports that Alphabet is discussing TPU sales to Meta, a potential shift from TPUs being used exclusively inside Google’s data centers.
- CEO Jensen Huang said the company is “sold out” of cloud GPUs after a strong quarter, with scarcity pushing some hyperscalers to test alternatives.