After the news that Meta was developing its own chip, reports suggest that Google is now developing a new AI chip too.
Google is said to be developing a new Tensor Processing Unit (TPU) and is planning to partner with MediaTek on its 7th-generation TPU, which is set for production in 2026. The chip is likely to be fabricated by TSMC.
TSMC has been working closely with Google for many years, co-operating on Google’s last mobile processor, the Tensor G5, using TSMC’s 3nm node. Google’s 6th-gen TPU is rumoured to be manufactured using the Taiwanese foundry giant’s 3nm.
Google is expected to lead the design of its next-gen TPU, while MediaTek will handle the I/O module for the main processor and peripherals.
Unlike rivals Microsoft and Meta, Google designs its own AI server chips for internal R&D and cloud services, making it far less reliant on NVIDIA but which strengthens its position in the AI race.
As more companies look to lessen their dependence on NVIDIA, Amazon Web Services (AWS) is said to be adopting a price-cutting strategy to persuade customers to switch to its own Trainium AI chip-powered servers instead of the more expensive NVIDIA solutions.
According to industry watchers, while AWS is looking to use pricing as a strategy the bigger challenge will be whether it will be able to provide a smooth transition for developers whose AI workloads are heavily reliant on NVIDIA’s software ecosystem.
Despite that, according to AWS, Apple, Adobe, and AnthropiC have all been testing the latest Trainium chips as they look for alternatives to NVIDIA.
