Google adds two AI chips for secure training

New TPU 8tx and TPU 8i Max target confidential AI workloads and larger memory-heavy tasks as Google expands its in-house hardware lineup.

Google has expanded its in-house AI chip lineup with two new tensor processing units, or TPUs, aimed at different demanding workloads. The move is part of the company’s effort to rely less on chips made by other firms for its cloud business. It also lands in the middle of a broader AI hardware race, as major cloud providers and AI labs push to build more of their own infrastructure instead of depending entirely on Nvidia.

The new chips are TPU 8tx and TPU 8i Max. Both are part of Google’s eighth generation of TPUs, introduced in April. These additions extend that lineup into more specialized areas, with one focused on confidential inference and training, and the other on higher memory capacity for large AI tasks.

TPU 8tx targets confidential AI workloads

TPU 8tx was developed with AMD and is based on the TPU 8t chip. Google says it is designed for confidentially training and running AI models.

The company says the chip “builds security directly into the hardware layer,” keeping sensitive data and model weights protected and isolated even while they are in use. In practical terms, that makes it a fit for cases where companies want to work with valuable or private data without exposing it more than necessary.

Google also says TPU 8tx includes a custom inferencing network. That lets it run reasoning alongside training, which can help distill and train small models on the fly. According to IT-PUB News, Google says this could be useful for businesses that want to train models on proprietary data without moving that data between devices.

TPU 8i Max packs more memory into one chip

The second chip, TPU 8i Max, is a more memory-focused version of Google’s existing inference hardware. Google says it effectively combines two TPU 8i chips launched in April into a single package.

That setup doubles memory capacity, processing power, and bandwidth. Google says the result is a chip that allows AI models to handle a wider range of information during queries. For companies running large systems, that matters because memory limits can affect how much data a model can process at once.

TPU 8i Max is part of the same broader push as TPU 8tx: giving customers more options for specific AI workloads instead of a single general-purpose setup.

Google joins the broader shift toward custom AI silicon

The launch comes as large cloud companies and AI developers increasingly design their own chips. Microsoft, Amazon, OpenAI, and Anthropic are all working on their own AI-specific semiconductors and software, according to the source text.

The goal is to reduce dependence on Nvidia, which has dominated the market for AI chips. Google’s move fits squarely into that trend and underlines how important hardware has become in the competition around artificial intelligence. Rather than relying only on outside suppliers, these companies are trying to control more of the stack themselves, from software to the chips that run it.

For customers, that can mean more specialized tools for different tasks, including secure training and larger inference workloads. It may also shape how cloud services are built and priced, since chip design can influence performance, capacity, and infrastructure planning.

Nvidia faces pressure as rivals build alternatives

The news also arrives at a sensitive time for Nvidia. The source notes that the company is facing a particularly vulnerable moment after CEO Jensen Huang over the weekend admitted to having an affair with someone alleged to be a Chinese spy.

That development is separate from Google’s chip announcement, but it adds to the scrutiny around Nvidia as rivals try to loosen its grip on the AI hardware market. Google’s latest TPUs are another sign that competition is no longer only about models and cloud services. It also extends to the hardware underneath them.

For businesses using AI, that shift could eventually mean more choice. At the same time, the move toward custom chips shows how central hardware has become to questions of data security, model size, and processing speed. Google’s new TPUs are aimed directly at those pressures.