Alphabet's Google is planning to collaborate with Taiwan's MediaTek on the upcoming version of its AI chips, Tensor Processing Units, scheduled for release next year, as reported by The Information on Monday. Google has not severed its partnership with Broadcom, the chip designer with whom it has worked exclusively on AI chips for several years, according to a report referencing a San Jose-based company employee.
Responses to Reuters' inquiries from Google, MediaTek, and Broadcom are pending.
Similar to Nvidia, Google independently designs AI server chips for internal projects and offers them to cloud clients. This strategy allows Google to outperform competitors like Microsoft-supported OpenAI and Meta Platforms by reducing its dependency on Nvidia chips, despite significant demand for Nvidia chips in the industry.
Last year, Google launched its sixth-generation TPU to provide an alternative to Nvidia chips for itself and its Cloud customers. Google selected MediaTek partly due to its strong ties with TSMC and its competitive pricing compared to Broadcom, suggested The Information report.
Research firm Omdia estimated Google's expenditure on TPUs last year between $6 billion and $9 billion based on Broadcom's AI semiconductor revenue projections for the same period.