Industry

Google Launches TPU 8tx and TPU 8i Max Chips to Reduce Nvidia Reliance

Google introduces two specialized tensor processing units designed for confidential inference and large-model inference, expanding its in-house chip portfolio.

Last verified:

Google Expands Tensor Processing Unit Lineup

According to TechCrunch AI, Google unveiled two new specialized tensor processing units on November 10: the TPU 8tx and the TPU 8i Max. These additions to Google’s eighth-generation chip family, which debuted in April 2026, extend the company’s in-house silicon strategy into confidential computing and large-scale inference workloads. The releases signal Google’s ongoing effort to build a comprehensive alternative to third-party accelerators for its cloud infrastructure.

TPU 8tx: Confidential Computing Variant

Google co-designed the TPU 8tx with AMD, positioning it specifically for secure model training and inference. According to TechCrunch AI, the chip embeds security mechanisms directly into the hardware layer, protecting sensitive data and model weights during active computation. This architecture eliminates the need to move proprietary datasets between devices during model distillation, a capability Google highlighted as valuable for enterprises processing confidential information on their own infrastructure.

The TPU 8tx integrates a custom inference network that enables simultaneous reasoning and training, allowing efficient small-model creation from proprietary datasets without data egress risks.

TPU 8i Max: Doubling Inference Capacity

The TPU 8i Max consolidates two TPU 8i inference chips into a single package, doubling memory capacity, processing power, and bandwidth in the process. According to TechCrunch AI, this design permits larger language models to process more context in a single query, addressing a bottleneck for inference-heavy workloads. By binning two chips together, Google reduces physical footprint while maintaining performance scaling.

Industry Context: The Broader Chip Sovereignty Push

Google’s announcement reflects a widening industry trend. According to TechCrunch AI, hyperscalers including Microsoft, Amazon, and AI labs such as OpenAI and Anthropic are all developing proprietary semiconductor lines to reduce reliance on Nvidia’s dominance. Google’s expanded TPU family is part of this vertical-integration race, where controlling hardware design provides cost advantages and differentiation at scale.

Why This Matters

Google’s TPU 8tx and TPU 8i Max target two high-margin, regulation-sensitive segments of cloud AI: confidential computing and long-context inference. Enterprises in finance, healthcare, and defense—where data isolation requirements are strict—may now choose Google Cloud over public Nvidia-dependent platforms without performance compromise. The TPU 8i Max similarly addresses a pricing lever: reducing bandwidth bottlenecks in inference clusters can compress per-token serving costs, a critical metric for competing on large-model inference pricing. For cloud procurement teams evaluating in-house versus third-party silicon, these releases narrow the performance gap and strengthen Google’s pitch for workload consolidation.

Frequently Asked Questions

What are TPU 8tx and TPU 8i Max?

TPU 8tx is a confidential-compute variant co-designed with AMD that isolates sensitive data at the hardware layer; TPU 8i Max doubles the memory and bandwidth of the TPU 8i by combining two chips into one package.

How do these chips help Google reduce Nvidia dependence?

These specialized variants expand Google's in-house tensor processing unit lineup to cover additional use cases—confidential workloads and large-model inference—reducing the need for third-party accelerators.

When were the TPU 8tx and TPU 8i Max announced?

Google announced both chips on November 10, 2026, as part of its eighth-generation TPU family.

#hardware #chips #TPU #Google Cloud #AI infrastructure #semiconductor