Google Expands Its AI Chip Lineup with TPU 8tx and TPU 8i Max

Highlights

Google has introduced two new variants of its eighth-generation tensor processing units: TPU 8tx and TPU 8i Max. The chips extend the company’s in-house AI hardware portfolio to confidential model training and inference, and to workloads that need more memory. Developed with AMD, TPU 8tx is designed to protect sensitive data and model weights while they are actively used. TPU 8i Max combines the capabilities of two inference-focused TPU 8i chips in one package, increasing memory capacity, processing power, and bandwidth. The launches reflect Google’s effort to build specialized infrastructure for AI and reduce dependence on external chip suppliers. They also come amid broader industry investment in custom AI semiconductors.

Sentiment Analysis

• The article’s overall tone is largely neutral and explanatory. It describes Google’s new chips, their intended uses, and the technical features the company says distinguish them, without presenting independent test results or making a direct judgment about their performance.
• There is a cautiously positive element in the account of expanded capabilities. Confidential computing and greater memory capacity are presented as potentially useful for organizations that want to work with proprietary data or run AI models that need to process more information. These benefits are attributed to Google rather than treated as independently verified outcomes.
• The broader competitive context introduces a more mixed tone. Google and other technology companies are developing custom chips to reduce reliance on Nvidia, while the article also reports a personal allegation involving Nvidia’s CEO. That allegation is separate from the technical discussion and is presented as reported context, not evidence about Nvidia’s products or business performance.
• Overall sentiment: neutral with a modest positive emphasis on product capabilities and a mixed competitive backdrop.

Article Text

Google has expanded its portfolio of in-house artificial intelligence chips with two new variants of its tensor processing units, or TPUs. The company says the new chips address two distinct needs: running and training AI models while protecting sensitive information, and supporting larger workloads with additional memory. The announcement is part of Google’s wider effort to develop more of the hardware used in its cloud computing business itself, rather than relying as heavily on chips made by other companies.

The new products are TPU 8tx and TPU 8i Max. Both belong to Google’s eighth generation of TPUs, which the company introduced in April this year. That generation was designed to provide more capacity for inference—the process of using a trained model to produce outputs—and to support substantial model-training requirements. The two new variants extend that foundation toward specialized workloads rather than replacing the broader aims of the eighth-generation platform.

Google developed TPU 8tx with AMD, basing it on the TPU 8t chip. Its intended focus is confidential computing for AI training and inference. According to Google, the chip incorporates security at the hardware layer so that sensitive data and model weights remain protected and isolated while actively in use. This design is relevant to organizations that want to use private information with AI systems but need safeguards around the information and models involved.

The company also describes a custom inference network in TPU 8tx. Google says the network allows inference to run alongside training, creating the possibility of distilling and training smaller models during operation. One example given is a business training a model on proprietary data without moving that data between devices. This is presented as a potential use case, not as a guarantee that all data-handling requirements will be satisfied; organizations would still need to assess how a deployment fits their own security and compliance needs.

TPU 8i Max is aimed at a different constraint: the amount of information an AI model can handle during queries. Google describes it as combining two inference-focused TPU 8i chips, which the company launched in April, into a single package. The combined design doubles memory capacity, processing power, and bandwidth compared with one TPU 8i, according to the announcement. The additional resources are intended to let models work with a broader range of information when responding to queries.

The two chips therefore address separate parts of the AI infrastructure challenge: TPU 8tx emphasizes protected use of data, while TPU 8i Max emphasizes capacity for inference workloads. Their different roles show how chip makers are tailoring hardware to particular stages and conditions of AI development, rather than treating all AI workloads as identical. The source announcement does not provide comparative benchmark results, prices, or deployment figures, so the practical performance of the new variants cannot be assessed from the reported details alone.

The launch takes place amid a wider push by technology companies to develop their own AI-specific semiconductors and accompanying software. Microsoft and Amazon, as well as AI labs OpenAI and Anthropic, are cited as working on their own chip efforts. A key motivation described in the article is reducing dependence on Nvidia, whose hardware has a prominent position in the AI chip market. Google’s expanded TPU lineup forms part of this broader competitive movement toward purpose-built computing systems.

The announcement also coincides with a separate news development involving Nvidia CEO Jensen Huang. The source reports that Huang admitted over the weekend to having an affair with someone alleged to be a Chinese spy. The allegation is presented as contextual news surrounding the launch, rather than as a technical or business factor connected to Google’s chips. Taken together, the report describes both the product-level competition in AI hardware and the wider circumstances attracting attention to companies and executives in the sector.

Key Insights Table