Marvell teams up with Google to develop an AI MPU chip, and the stock price jumps 6.3% on the news

Google is in talks with Marvell Technology (MRVL) about collaborating on the development of two AI chips, the MPU and the TPU. After the news broke, Marvell shares rose 6.3% in pre-market trading today; as of the time of publication, the price was $148.89 per share, indicating the market’s approval of the outlook for customized application-specific integrated circuits (ASICs).

Marvell and Google discuss developing MPU and TPU

According to industry reports, Marvell is currently in discussions with Google to develop two core components: the dedicated memory processing unit (MPU) and the tensor processing unit (TPU). The main technical goal of this collaboration is to address bottlenecks caused by the “Memory Wall.” By leveraging the synergy between the MPU and Google’s existing TPU for collaborative computing, the partnership aims to optimize data transmission efficiency. If the collaboration goes smoothly, the two companies expect to complete the design as early as 2027 and move into test production. The development of this kind of specialized architecture will enable large language models such as Gemini to perform inference tasks with lower latency and higher energy efficiency.

Google strengthens in-house chip development and wants a bigger slice of the market

In recent years, Google has actively strengthened its TPU product line, aiming to reduce reliance on NVIDIA (NVIDIA) GPUs. After restructuring in 2022 by bringing its internal chip sales team back in-house, Google has started selling TPU access rights directly to external data-center customers, rather than limiting them to use on its own cloud platform. The Torch TPU program released earlier this month further demonstrates Google’s intent to integrate the PyTorch software framework, attempting to lower the workload migration threshold for developers. By working with Marvell to develop customized accelerators, Google expects it can further optimize capital expenditures for cloud infrastructure, while also challenging the positions of leading companies such as Broadcom and NVIDIA in the fiercely competitive customized ASIC market.

Marvell’s financial results are strong; adjusted revenue grows 81%

Marvell’s operating performance in fiscal 2026 is strong. Annual revenue reached $8.195 billion, and adjusted earnings grew 81% compared with the same period last year. Although the company’s current price-to-earnings ratio is close to 44x, its steady profitability and more than 51% gross margin have supported investors’ confidence. The company’s current product mix is not limited to traditional networking equipment; it is focusing on building end-to-end AI infrastructure. In addition to the project with Google, Marvell also has 18 ASIC projects currently under development, as well as a strategic alliance agreement with NVIDIA valued at $2 billion. These orders ensure Marvell’s key position in the artificial intelligence hardware supply chain.

The focus of the artificial intelligence industry is shifting from “training” models to “inference” on the actual application end. The market demand for customized chips that deliver high performance and energy savings is growing multiple times. With the flexibility of a fabless business model, Marvell focuses on critical technologies such as high-performance computing, network interconnects, and storage controllers, enabling it to quickly respond to cloud giants’ demand for specialized hardware. Wall Street analysts believe that although Marvell’s stock price is currently at its historical high, its positioning in the AI inference market and increasingly diversified customer base show that its business is just getting started.

This article, Marvell and Google cooperate to develop MPU AI chips, shares jump 6.3% at the earliest, first appeared on ChainNews ABMedia.