Alphabet’s Google is reportedly in negotiations with Marvell Technology to build new custom artificial intelligence processors. The reported discussions focus on developing chips specifically designed for AI inference, marking a significant step in Google’s push to diversify its supply chain. By teaming up with Marvell, Google is signaling a clear intent to strengthen its position in the highly competitive semiconductor market and reduce its reliance on Nvidia’s dominant hardware.
The collaboration between the two technology giants centers on the development of two distinct processors. According to industry reports, the first is a custom memory processing unit. This component is designed to alleviate data movement bottlenecks and work seamlessly alongside Google’s existing in-house hardware, known as Tensor Processing Units, or TPUs.
The second processor is a brand-new TPU engineered entirely for inference workloads. Inference is the critical phase where trained artificial intelligence models actually generate responses and serve users in real time. As applications scale, optimizing the power efficiency, processing speed, and operating costs of inference has become a top priority for major cloud service providers.
Expanding the Silicon Supply Chain
In this proposed arrangement, Marvell would serve primarily as a design services provider. This role is comparable to the position MediaTek holds in developing Google’s latest-generation hardware. If the ongoing talks result in a signed contract, Marvell will officially become the third major design partner in Google’s rapidly expanding custom silicon supply chain.
This development follows closely on the heels of Google securing a massive, long-term supply agreement with Broadcom. That existing partnership is locked in through the year 2031, showcasing the tech giant’s aggressive strategy to build a robust network of chip designers. Analysts project that the broader custom application-specific integrated circuit market will experience a massive 45 percent growth rate in 2026 alone, potentially reaching a total value of $118 billion by 2033.
The Rivalry with Nvidia
The battle for hardware supremacy is increasingly shifting toward inference capabilities rather than just model training. NVIDIA, the undisputed leader in general-purpose graphics processing units, has recently made aggressive moves into the inference space. Late last year, Nvidia acquired the inference chip developer Grok and subsequently launched its own hardware solution, the Grok 3 LPU.
According to reports citing a Google employee, the unveiling of Nvidia’s new language processing unit accelerated Google’s internal development timelines. Interestingly, Marvell previously served as a development partner for Grok to design early-generation processors. By engaging with Marvell, Google is effectively collaborating with an ally of its primary rival.
NVIDIA is taking proactive steps to maintain its market dominance and keep its clients close. The company recently announced an expanded partnership with Marvell, utilizing its NVLink Fusion framework. This system is designed to integrate custom chips directly into Nvidia’s networking and server hardware. Furthermore, Nvidia reportedly made a $2 billion strategic investment in Marvell to ensure a smooth integration process for its cloud customers.
Despite the rising competition, overall demand for artificial intelligence infrastructure remains incredibly strong. Taiwan Semiconductor Manufacturing Company and ASML both lifted their financial forecasts recently, citing strong investment signals from cloud clients. NVIDIA itself continues to post staggering results, reporting a record $68.1 billion in revenue for its fourth quarter, fueled heavily by data center operations.
Market Reaction and Outlook
Financial markets reacted quickly to the news of the potential Google and Marvell partnership. NVIDIA shares slipped by $1.98, dropping to $200.08 in early Tuesday trading on April 21. The minor pullback slightly reduced the company’s massive market capitalization to approximately $4.53 trillion.
Conversely, Marvell Technology experienced a notable boost from the reported negotiations. Shares of the semiconductor company climbed 2.79 percent, reaching $151.97. The positive momentum was bolstered by financial analysts at RBC Capital, who raised their price target for Marvell from $115 to $170. The firm maintained an Outperform rating, pointing to the company’s strong growth outlook and the sustained demand for cloud infrastructure capacity. Morgan Stanley also noted that the rise of agentic AI—systems that plan and act autonomously—could drive further spending on memory and processing components.
Neither Google nor Marvell has provided official public comments regarding the ongoing negotiations. The reported talks align perfectly with the broader industry trend of hyperscale cloud providers investing heavily in custom silicon to meet the unique demands of highly complex workloads.
The timing of these hardware developments is notable as Google kicks off its Cloud Next event in Las Vegas on April 22. With the industry spotlight focused heavily on infrastructure capabilities, the reported negotiations highlight how rapidly the semiconductor landscape is evolving to support the next generation of digital tools.
