People familiar with the matter said today that Google plans to abandon Broadcom as early as 2027 and instead develop its own AI server chips. Earlier this year, Google and Broadcom failed to reach an agreement on chip pricing, the person said. For this reason, Google decided to abandon Broadcom as the supplier of its AI chip tensor processing units (TPU).
Google executives have discussed this extensively. This means that Google will design its TPU chip independently. The move is expected to help Google save billions of dollars in costs each year. Currently, Google is investing heavily in artificial intelligence research and development. Research and development for artificial intelligence is particularly expensive compared to other types of research and development.
Broadcom CEO Hock Tan recently said that generative artificial intelligence will account for more than 25% of the company's semiconductor revenue next year. In the second fiscal quarter of fiscal 2023 that ended in April, artificial intelligence revenue accounted for approximately 15% of Broadcom's semiconductor business revenue, while this proportion will be only 10% in fiscal 2022.
Last year, Microsoft-backed artificial intelligence startup OpenAI triggered an AI technology arms race in the technology industry after it released the generative chatbot ChatGPT. This year, Google has also increased its investment in generative artificial intelligence in an effort to catch up.
Data from research firm Similarweb shows that in August this year, Google’s artificial intelligence chatbot Bard had 183 million visits, which was only 13% of ChatGPT’s visits.
At the same time, people familiar with the matter also said that Google also plans to cooperate with chip manufacturer Marvell Technology to launch a new network chip next year, with the development codename "Granite Redux".
Google, Broadcom and Marvell have yet to comment.