Marvell: Report on negotiations with Google to develop two AI chips

US tech giant Google is in talks with chip designer Marvell Technology about developing two new chips that are intended to run AI models more efficiently. This was reported on Sunday by the tech portal The Information, citing two people directly familiar with the discussions.

According to the report, the potential deal could include two different chips: a memory processing unit to complement Google's Tensor Processing Unit (TPU) and a new TPU for more efficient execution of AI models.

Google's Dual Strategy

Google's potential cooperation with Marvell according to the financial portal Finanzen.net suggests a dual strategy. The company's own TPUs are to be further developed technologically, and the chips are to be positioned more strongly as standalone cloud products.

In parallel, Google is trying to expand sales of its own AI hardware through partnerships with other cloud providers and large customers. It was only recently announced that Anthropic intends to use Google's TPUs on a large scale to run its Claude models on them.

Diversification of Supplier Relationships

Google currently works primarily with US semiconductor company Broadcom on chip development. According to The Information, however, the Alphabet subsidiary aims to diversify its supplier relationships beyond Broadcom in view of the sharply rising demand for its chips. This is in line with the trend. As a result of increasing competition in the field of Artificial Intelligence (AI), tech giants like Google and Meta are trying to reduce their dependence on external chip manufacturers and increasingly develop their own chips.

For Marvell, a collaboration with Google would be another prestigious order. Following the announcement of the talks with Google, the price of Marvell's stock rose sharply on the stock exchanges, according to a report by the Reuters news agency.

At the end of March, US chip giant Nvidia already invested two billion US dollars in Marvell. Both companies are also entering into a strategic partnership. This is intended to make it easier for customers building on Nvidia architectures to use Marvell-developed custom AI chips in combination with Nvidia's network components and processors. For several years now, Marvell has been specifically positioning itself as a provider of such tailor-made specialty chips for hyperscalers. The demand for such custom solutions is growing rapidly, as evidenced by the interest from Nvidia and now also Google.

(akn)