AI
Google Expands Custom AI Chip Lineup with Two New Specialized Processors
Google unveiled the TPU 8tx and TPU 8i Max, expanding its in-house processor portfolio to tackle confidential computing and larger AI workloads.
TechCrunch — AI
Google is deepening its commitment to internal semiconductor development, rolling out two new tensor processing units (TPUs) designed to extend its capabilities beyond traditional AI inference and training. The move underscores the company's broader effort to lessen reliance on third-party chip suppliers for its cloud infrastructure.
New Chips Target Specific Use Cases#
The two processors—TPU 8tx and TPU 8i Max—arrive as part of Google's eighth generation of TPU silicon, which debuted earlier this year. Each addresses a distinct operational challenge: TPU 8tx focuses on privacy-sensitive workloads, while TPU 8i Max tackles memory-intensive applications.
The TPU 8tx, developed collaboratively with AMD, brings security considerations directly into the hardware. This chip is engineered to handle confidential training and inference, embedding protective measures that keep model weights and proprietary data encrypted and isolated throughout processing. A distinguishing feature is its integrated reasoning-training capability, enabling companies to fine-tune smaller models on sensitive data without moving information across systems.
TPU 8i Max consolidates two inference-focused processors from the earlier TPU 8i generation into a unified package, effectively doubling the memory footprint, compute throughput, and available bandwidth. This consolidation allows AI systems to process richer, more contextually dense prompts.
Industry-Wide Race to Build Custom Silicon#
Google's announcement arrives amid an intensifying industry trend: major cloud operators and AI labs are pursuing their own semiconductor designs to reduce dependence on Nvidia's market dominance. Competitors including Microsoft, Amazon, OpenAI, and Anthropic are simultaneously investing in proprietary chip development, coupled with tailored software stacks designed to optimize their respective AI workloads.

