Google broadens its AI hardware portfolio with two new specialized chips

Summary

• Google unveiled two new in-house developed tensor processing unit (TPU) variants, TPU 8tx & TPU 8i Max, to reduce dependence on external chipmakers for its cloud computing business.
• The new units are part of Google's eighth generation of chips introduced in April & broaden the portfolio to include confidential inference & increased memory capacity for large workloads.
• TPU 8tx, designed with AMD, is based on the TPU 8t chip & features hardware-layer security to protect sensitive data & model weights during active use.
• TPU 8tx includes a custom inferencing network that allows reasoning alongside training, enabling on-the-fly distillation & training of small models on proprietary data without moving data between devices.
• TPU 8i Max combines two inference-focused TPU 8i chips into a single package, doubling memory capacity, processing power, & bandwidth to handle larger information ranges during queries.
• The launch occurs as hyperscalers like Microsoft, Amazon, OpenAI, & Anthropic develop their own AI-specific semiconductors & software to reduce reliance on Nvidia.
• The announcement follows a weekend admission by Nvidia CEO Jensen Huang regarding an affair with an individual alleged to be a Chinese spy.