Google broadens its AI hardware portfolio with two new specialized chips

Google has expanded its AI hardware lineup with the introduction of two new specialized chips aimed at reducing its reliance on chips from other manufacturers for cloud computing. The new additions to Google's tensor processing units (TPUs) are TPU 8tx and TPU 8i Max, both part of the company's eighth generation of chips rolled out in April. These chips enhance the capabilities of inference capacity and provide increased memory for large workloads.

TPU 8tx, developed in collaboration with AMD, specializes in training and running AI models while maintaining data security. The chip features a custom inferencing network that enables reasoning alongside training, allowing for the on-the-fly distillation and training of small models without the need to transfer data between devices. This feature could be particularly useful for companies looking to train models using proprietary data without compromising its security.

TPU 8i Max, on the other hand, essentially doubles the memory capacity, processing power, and bandwidth of two previously launched inference-focused TPU 8i chips, all housed in a single package. This increases the chip's capacity to handle larger datasets and information during queries. The news of Google's chip launch arrives as major players such as Microsoft, Amazon, OpenAI, and Anthropic are also developing their own AI-specific semiconductors and software to lessen their dependence on Nvidia's dominance in the market.

Additionally, the launch of Google's new chips comes at a crucial time for Nvidia, whose CEO, Jensen Huang, recently admitted to having an affair with a person suspected of being a Chinese spy.