Key Points:
- Alphabet’s Google is currently negotiating with Marvell Technology to develop two brand new custom computer chips.
- The companies want to build a new memory processing unit to work alongside Google’s existing hardware.
- Google is also designing a brand new Tensor Processing Unit built specifically to run complex AI models.
- The tech giants aim to finalize the design of the memory chip by next year before moving to test production.
Google wants to build better hardware for the artificial intelligence revolution. The tech giant, owned by Alphabet, is currently in deep talks with Marvell Technology to develop two brand new custom computer chips. The companies hope these new chips will run complex artificial intelligence models much faster and far more efficiently than current hardware allows. The Information first reported these secret discussions on Sunday, citing two people with direct knowledge of the ongoing corporate negotiations.
The partnership focuses on two very distinct pieces of hardware. First, the companies want to design a highly specialized memory processing unit. Engineers will build this new memory chip specifically to work alongside Google’s existing Tensor Processing Unit, commonly known as a TPU. Second, the teams plan to build an entirely new version of the TPU itself. Google wants this new processor built from the ground up, specifically for running advanced artificial intelligence models.
This massive hardware push is part of a much larger corporate strategy. Google desperately wants to make its custom TPUs a truly viable alternative to Nvidia’s famous graphics processing units. Right now, Nvidia completely dominates the artificial intelligence hardware market. Every major tech company in the world buys Nvidia chips to train its software. Google hopes to break that monopoly by offering a cheaper, faster, and more efficient in-house alternative.
Selling these custom chips is already paying off for the search engine giant. TPU sales have quickly become a major driver of growth in Google’s broader cloud computing revenue. The company rents out access to these powerful chips to other software developers. Google wants to use this growing cloud revenue to prove to nervous Wall Street investors that its massive, multi-billion dollar investments in artificial intelligence are actually generating real financial returns.
Independent reporters at Reuters could not immediately verify the details of the Sunday report. Furthermore, both Google and Marvell Technology completely declined to comment when reporters asked for official statements regarding the secret negotiations.
Despite the corporate silence, the project timeline seems aggressive. According to the report, the two companies aim to completely finalize the complex design of the new memory processing unit as soon as next year. Once the engineers approve the final blueprints, they will hand the design off to a manufacturing partner to begin initial test production. If the test chips work correctly, Google will likely mass-produce the hardware and install it inside its massive data centers around the world.
This partnership makes perfect sense for both companies. Marvell Technology specializes in designing complex silicon solutions for data centers and massive cloud computing networks. By teaming up with Marvell, Google gains access to some of the best hardware engineers in the world. Meanwhile, Marvell gets to work with one of the largest technology companies on the planet, securing a massive financial contract that could boost its own stock price and market share. As artificial intelligence continues to grow, custom hardware will become the most valuable commodity in the tech industry.
