VDe Vector

Google announces two new AI chip systems

The new TPU configurations focus, according to Google, on hardware security and very large context windows.

Google announced two new configurations of its own AI chips on Monday. According to the company, the TPU 8tx and TPU 8i Max are primarily intended to improve the security of AI computations and the processing of long contexts.

Google presented the systems on 9 November via its corporate blog. The chip configurations are intended for large AI workloads in data centres and cloud environments. Google identifies hardware-based data security and support for contexts of up to one million tokens as the two main objectives.

According to Google, the TPU 8tx is aimed at training and large-scale AI applications, while the TPU 8i Max is intended for inference: running already-trained models. Google says both systems should help companies put autonomous AI agents into production, although the announcement does not say that such applications will therefore automatically be reliable or secure.

Google says the new systems provide security at hardware level. This means the company does not want to leave protection solely to software and access controls. Independent test results are needed to assess how the security compares with existing systems.

Reuters reported that Google is stepping up competition with Nvidia in the AI accelerator market with the new chips. Google already uses TPUs for its own services, such as Search, YouTube and Gemini, and also makes them available through Google Cloud. For cloud customers, this could offer an alternative to using chips from external suppliers.

Google says that one of the new configurations can achieve up to twice the speed of the previous version without any significant loss of performance. That is a company claim, not an independent benchmark. For Dutch companies, the key questions are whether the systems will become available through cloud contracts, what costs will be involved and whether European customers will retain control over their data.

The announcement is part of a broader battle over AI infrastructure. Nvidia remains a major supplier of GPUs, while Google, Microsoft, Amazon and other cloud companies are developing their own chips. The new TPUs will not immediately transform the market, but they make clear that AI capacity is increasingly becoming a combination of models, data centres and specialised hardware.

Respond to this article

← Back to the edition