Challenging NVIDIA Corporation (NVDA.US)? Alphabet Inc. Class C (GOOGL.US) will develop two AI chips.

Local time, April 19th, it was reported by media citing informed sources that Google, a subsidiary of Alphabet, is in talks with US chip design company Marvell Technology to develop two new chips aimed at running artificial intelligence models more efficiently. It is reported that one of the chips is a memory processing unit designed to work in conjunction with Google's Tensor Processing Unit (TPU), and the other chip is a new TPU specifically designed for running artificial intelligence models. Marvell Technology will take on the role of design services, similar to how MediaTek participated in Google's latest Ironwood TPU project. Currently, the negotiations have not yet resulted in a formal contract. Currently, NVIDIA's GPU still dominates the AI chip market. Google has been working to position its TPU as an alternative to NVIDIA's GPU. As Google tries to prove that its huge investment in AI is translating into commercial returns, TPU sales have become a key driver of Google Cloud growth. The sources also revealed that Google and Marvell Technology aim to complete the design of the memory processing unit as early as next year, followed by moving it into the trial production phase. Just a few days ago, Broadcom finalized a TPU cooperation agreement extending until 2031. This negotiation also reflects Google's focus shifting towards inference, a major computational cost item in AI applications. The custom ASIC market is expected to grow by 45% by 2026 and is expected to reach $118 billion by 2033. However, analysts point out that Google is not looking to replace Broadcom at this time, but rather to add a third design partner to its supply chain. The core of this strategy is diversification rather than replacement. Currently, Google's custom chip supply chain includes Broadcom, which is responsible for high-performance TPU variants, MediaTek, which is responsible for cost-optimized "e" series TPUs (20%-30% lower cost), and Taiwan Semiconductor Manufacturing Company (TSMC), responsible for manufacturing. On the other hand, analysts point out that Google's partnership with both NVIDIA and Marvell Technology clearly demonstrates the chip design company's position in the AI data center custom chip market. Previously, Marvell Technology signed a $2 billion strategic cooperation agreement with NVIDIA. Meanwhile, in the fiscal year ending February 2026, Marvell Technology's data center business achieved a record revenue of $6.1 billion, with total revenue reaching $8.2 billion, a 42% year-on-year increase. Data shows that its custom chip business generates annualized revenue of about $1.5 billion, securing design orders from 18 cloud providers, building chips for clients such as Amazon (Trainium processor), Microsoft (Maia AI accelerator), Meta (new data processing unit), and also carrying out existing design work for Google's Axion ARM CPU. So far this year, Marvell Technology's stock price has risen by over 56%, with a 30% increase in April alone due to the NVIDIA partnership and Google negotiation news. Barclays analyst Tom O'Malley has upgraded Marvell to "hold" and raised the target price from $105 to $150. This article is reproduced from "Cai Lianshe", edited by GMTEight: Feng Qiuyi.