Google is expanding its artificial intelligence hardware strategy through a major partnership with semiconductor company Marvell Technology. The agreement will see Marvell work on a range of custom chips connected to Google’s Tensor Processing Unit, or TPU, ecosystem, as the tech giant looks for more control over the hardware powering its AI services.
An important part of the agreement gives Google the right to purchase nearly 59 million shares of Marvell at an exercise price of $206.58 per share. If fully exercised, the stock purchase could be worth about $12.2 billion. The warrants are expected to vest over time through 2033 as certain purchasing conditions are met. This means the $12.2 billion figure is not simply a cash payment from Google to Marvell for chips, but the potential value of Google’s investment rights under the wider commercial relationship.
The partnership covers several types of custom silicon designed for Google’s growing AI infrastructure. These include AI inference accelerators, storage controllers, network interface controllers, memory interface technology and near-memory computing products. Such specialized chips can help data centers process AI workloads more efficiently and reduce their dependence on general-purpose processors.
Google has spent years developing its own TPUs to handle demanding machine-learning workloads. These processors are used across Google’s AI services and Google Cloud infrastructure. As generative AI becomes more widely used, companies need huge amounts of computing power to train models and, increasingly, to run those models for millions of users.
AI inference has therefore become an important part of the chip market. Inference happens when a trained AI model receives a request and produces an answer, image, prediction or another output. Companies running large-scale AI products need chips that can perform these tasks quickly while keeping electricity and infrastructure costs under control.
Working more closely with Marvell could give Google another major partner for designing specialized processors. Google has already worked extensively with Broadcom on its TPU technology. Expanding its supplier network could provide Google with more flexibility as demand for AI computing continues to increase. Broadcom shares fell around 2% in premarket trading following the Marvell announcement, while Marvell shares jumped by more than 10%.
The agreement also shows how competition in the AI semiconductor industry is expanding beyond Nvidia. Nvidia remains a major supplier of GPUs used for artificial intelligence, but large technology companies are increasingly investing in custom chips designed specifically for their own workloads.
For Google, custom processors can help improve performance, reduce costs and provide greater control over its AI infrastructure. For Marvell, the Google partnership gives the company a stronger position in the fast-growing market for custom AI silicon and data-center technology.
The deal could become more significant over the coming years as Google continues expanding Gemini, Google Cloud and other AI-powered services. With computing demand rising rapidly, partnerships around custom chips are becoming an increasingly important part of the global AI race.
