tensor Processing Units (TPU, tensor processor) was originally used by Google for its AI training models (artificial intelligence). According to the Information, Google will join hands with MediaTek to produce TPUs next year.
However, Google has not yet severed its relationship with broadcom, a chip design company that it has cooperated exclusively to develop AI chips for many years.
Google, MediaTek and broadcom have yet to comment on the issue.
Like Nvidia, Google also designs its own AI server chip, used for internal research and development purposes and leased to customers using cloud services.
This approach gives Google a competitive edge in the AI race by reducing its dependence on Nvidia, even as rivals like OpenAI and Meta Platforms backed by Microsoft have shown increased demand for Nvidia chips.
At the end of 2024, Google launched the 6th generation TPU to provide itself and customers using cloud services with an alternative to Nvidia chips, which are the most sought-after processors in the AI industry.
Google is partnering with MediaTek partly because the chipmaker has a close relationship with TSMC and charges Google less per chip than broadcom.
According to data from market research firm Omdia, Google spent $6-9 billion on TPU production last year.
