Google broadens its AI hardware portfolio with two new specialized chips

Two new TPUs, two new use cases

Google has added two new Tensor Processing Units (TPUs) to its AI hardware lineup. TPU is the name of Google's family of custom chips built for AI computation.

According to the company, the new chips take the portfolio into more use cases. The two highlighted areas are confidential model training and higher memory capacity.

What confidential training means

Confidential model training is a security approach that aims to keep the data processed during training and the model itself closed to outside access. That matters especially for organizations working with sensitive data.

Higher memory capacity, meanwhile, aims to ease bottlenecks in training and running large models.

Why it matters

Google continues to strengthen both its own cloud infrastructure and the compute it offers customers in AI chips. The new TPUs can be read as a move to spread its chip portfolio from a single workload to multiple scenarios.