MediaTek and Google have announced a collaboration to develop next-generation tensor processing units (TPUs), aiming to enhance Google's AI hardware strategy.

Under this partnership, Google will oversee the overall design of the TPUs, while MediaTek will manage input/output modules to improve communication between the processor and peripherals, marking a shift from Broadcom's previous role.

Google's decision to partner with MediaTek was influenced by their close relationship with Taiwan Semiconductor Manufacturing Company (TSMC), which is expected to manufacture the new chips.

As part of its strategy to reduce reliance on external companies like Nvidia, Google has been developing its own AI server chips for internal use and cloud services.

Unlike competitors such as Microsoft and Meta, Google is focused on designing its own AI server chips, further decreasing its dependence on Nvidia's dominant processors.

If successful, the collaboration with MediaTek may lead to more efficient AI processing, potentially influencing the development of advanced large language models (LLMs).

This partnership also signifies Google's intent to lessen its dependence on Broadcom for AI tools hardware, although Broadcom remains involved in the project.

Despite the new collaboration, Google's relationship with Broadcom is confirmed to be intact, as Broadcom has been a long-time partner in developing AI chips.

Google's TPUs are critical for demanding AI computations, especially in training large language models, and the new collaboration aims to enhance their efficiency and performance.

Cost advantages are a driving factor in this partnership, as MediaTek offers a lower price per chip compared to Broadcom.

The new TPUs will prioritize affordability and energy efficiency, with production handled by TSMC as part of Google's 7th-generation TPU series.

Rumors suggest that Google's 6th-generation TPU will also be manufactured using TSMC's advanced 3nm technology.