Google unveiled two variants of its eighth-generation TPU family on November 9, 2026: the TPU 8tx, designed for large-scale model training, and the TPU 8i Max, optimized for low-latency inference. The TPU 8tx features hardware-level encryption developed with AMD to protect data at the chip level, particularly for government agencies and regulated industries. The TPU 8i Max targets long-context applications through increased compute density, improved memory bandwidth and 384 MB of integrated SRAM. These chips build on the TPU 8t, which can scale to 9,600 processors in a superpod delivering 121 Exaflops, while the TPU 8i supports pods of 1,152 chips and 11.6 Exaflops. Following the announcement, Alphabet shares rose by more than 2% in premarket trading.
Source: Read the original article
