The Alphabet subsidiary Google is apparently negotiating with Marvell Technology about the joint development of new AI processors.
• Discussions about developing two new chips
• Memory processing unit and AI TPU
• TPUs as an alternative to GPUs
As “The Information” reported on Sunday, citing insiders, Alphabet subsidiary Google and Marvell Technology are talking about the development of two specialized AI accelerators. According to “Reuters”, one of them will be designed as a so-called memory processing unit and will work closely with Google’s Tensor Processing Units (TPU) in the future, while the other chip will be a TPU that is specifically designed for the efficient execution of AI models. The aim is apparently to expand the performance of the existing TPU platform and at the same time realize cost and efficiency advantages compared to classic GPU architectures.
The design of the memory processing unit will reportedly be completed as early as 2027 and then go into test production.
This is behind Google’s chip offensive
Google is one of the few technology companies that has been developing its own AI accelerators for more than a decade. With the Tensor Processing Units, the group tried early on to make itself less dependent on standard solutions from external manufacturers. But the rapid rise of NVIDIA in the AI age has shown how dependent even hyperscalers can remain on specialized GPU architectures.
The possible cooperation with Marvell therefore suggests a double strategy. On the one hand, Google wants to further develop its TPU platform technologically. On the other hand, the company is pursuing the goal of positioning its chips more strongly as independent cloud products. In fact, according to Reuters, TPU-based offerings have now become an important growth driver for Google’s cloud division, which investors increasingly see as a central pillar of the AI strategy.
At the same time, the group is trying to expand sales of its own AI hardware through partnerships with other cloud providers and major customers. For example, it has already been reported in the past that Meta is negotiating billion-dollar access to Google’s AI chips or is already using the corresponding infrastructure.
Marvell is moving to the forefront of AI infrastructure
For Marvell, a collaboration with Google would be more than just a prestigious contract. For several years now, the semiconductor manufacturer has positioned itself specifically as a provider of tailor-made ASIC solutions for hyperscalers – i.e. precisely those special chips that are intended to supplement or partially replace classic GPU clusters.
The demand for such customized accelerators is growing rapidly. Analysts now expect that a majority of future data center architectures will consist of a combination of standardized GPUs and specially developed additional chips. Marvell has already secured key partnerships with cloud providers in this segment and plans to massively expand its ASIC sales by the end of the decade.
Carolin Ludwig, editorial team at finanzen.net
By the way: Alphabet C (ex Google) and other US stocks can even be traded on finanzen.net ZERO until 11 p.m. (without order fees, plus spreads). Open a depot now for free and secure a new customer bonus!
Selected leverage products on Alphabet C (ex Google)
With knock-outs, speculative investors can participate disproportionately in price movements. Simply select the lever you want and we will show you suitable open-end products on Alphabet C (ex Google)
The leverage must be between 2 and 20
Advertising
