← Voltar ao feed
low

Google unveils TPU 8tx and TPU 8i Max, two specialized variants of its eighth-generation AI chips

1 fonte(s) · 1 atualização(ões) · publicado 10/11/2026 12:54 · coletado há 2 horas

Google has added two variants to its eighth-generation TPU line: the TPU 8tx, designed with AMD and based on the TPU 8t, which builds security into the hardware layer for confidential training and inference, and the TPU 8i Max, which combines two TPU 8i chips in one package with double the memory capacity, processing power and bandwidth. The company presents both as in-house AI accelerators meant to reduce its dependence on chips from other vendors for Google Cloud. The article reports no vulnerability, exploitation, or security incident; this is a product and hardware-strategy announcement. Its claims about performance and security features come from Google's own announcement as relayed by TechCrunch, and the Google blog post was not opened for this record. For defenders, the relevance is limited to future cloud infrastructure options for confidential AI workloads.

Linha do tempo

  1. techcrunch.com ↗ · pub. 10/11/2026 12:54 · coletado 10/11 13:03 primeiro registro

    Google unveils TPU 8tx and TPU 8i Max, two specialized variants of its eighth-generation AI chips

    ler a análise desta fonte

    Google has added two variants to its eighth-generation TPU line: the TPU 8tx, designed with AMD and based on the TPU 8t, which builds security into the hardware layer for confidential training and inference, and the TPU 8i Max, which combines two TPU 8i chips in one package with double the memory capacity, processing power and bandwidth. The company presents both as in-house AI accelerators meant to reduce its dependence on chips from other vendors for Google Cloud. The article reports no vulnerability, exploitation, or security incident; this is a product and hardware-strategy announcement. Its claims about performance and security features come from Google's own announcement as relayed by TechCrunch, and the Google blog post was not opened for this record. For defenders, the relevance is limited to future cloud infrastructure options for confidential AI workloads.