Google advances diversification of chip suppliers, enters talks with Marvell for TPU customization and inference chip development.

```

Google is in negotiations with semiconductor company Marvell Technology regarding custom Tensor Processing Units (TPUs), and is exploring the possibility of developing dedicated chips for large model inference.

This move, coupled with Marvell’s previous $2 billion strategic cooperation agreement with Nvidia, has made Marvell’s position in the custom AI data center chip market increasingly clear.

According to FundaAI, Google is actively discussing the TPU development project with Marvell, with Marvell participating as a design service provider; the scope of discussions also extends to a dedicated chip optimized for large language model (LLM) inference workloads.

The negotiations come just a few days after Google renewed its long-term TPU and networking agreement with Broadcom through 2031, indicating that Google is proactively diversifying its custom chip suppliers and leveraging Marvell’s technical expertise in high-speed interconnects to optimize costs and performance.

Google's Negotiations: Strategic Positioning Behind Supply Diversification

FundaAI reports that Google's talks with Marvell involve two aspects: first, custom TPU development, with Marvell participating as a design service provider; second, a dedicated chip optimized for LLM inference scenarios, which outsiders refer to as the inference LPU (Language Processing Unit) architecture.

This development comes after Google just extended its collaboration with Broadcom through 2031.

For Google, bringing in multiple custom silicon suppliers amid rapidly expanding AI infrastructure investments is a standard approach to dispersing technical risk and maintaining bargaining power. Marvell’s expertise in high-speed interconnect technology perfectly fills Google’s gap in optimizing for specific workload requirements.

For Marvell, even though negotiations are still ongoing, the potential significance cannot be overlooked. The company’s custom ASIC business currently generates about $1.5 billion in annualized revenue, and if a Google contract materializes, it would add another high-value revenue stream to its existing design orders.

Nvidia's $2 Billion Investment, Marvell Enters the Core of the AI Ecosystem

The backdrop to Google's negotiations is Marvell’s recent key announcement last month—a strategic agreement in which Nvidia invested $2 billion and established deep cooperation through the NVLink Fusion framework.

Under the agreement, Marvell will design custom XPUs and vertically scalable networks compatible with NVLink, directly embedding them in Nvidia’s rack-level AI architecture to work in concert with GPUs, Vera CPUs, network cards, and switches. The two sides will also collaborate on silicon photonics technology for optical interconnects.

Nvidia characterizes this collaboration as an expansion of the AI ecosystem, claiming it maintains its core position in interconnect architecture while offering customers greater flexibility. For Marvell, the $2 billion cash injection directly strengthens its balance sheet and speeds up expansion in AI communications and AI-RAN markets.

Risk DisclaimerThe market involves risks, and investment needs to be cautious. This article does not constitute personal investment advice and does not take into account individual users' special investment goals, financial conditions, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their particular situation. Based on this, investment is at their own risk. ```