Nvidia is set to establish a new business unit dedicated to designing custom chips for cloud computing firms and other clients, including advanced artificial intelligence (AI) processors, according to sources familiar with the matter. The move comes as the dominant global designer and supplier of AI chips aims to capitalize on the burgeoning market for custom AI chips and protect itself against the increasing number of companies exploring alternatives to its products.
With about 80% control of the high-end AI chip market, Nvidia’s stock market value has skyrocketed by 40% this year to reach a staggering $1.73 trillion, following a more than triple increase in 2023. However, Nvidia’s customers, such as OpenAI, Microsoft, Alphabet, and Meta Platforms, are now racing to secure the limited supply of its chips for the fast-growing generative AI sector.
While the company’s H100 and A100 chips currently serve as generalized, all-purpose AI processors for major clients, these tech companies have started developing their own internal chips tailored to their specific needs. This helps reduce energy consumption and potentially lowers the cost and time needed for design.
To address this shift, Nvidia aims to assist these companies in developing custom AI chips that have been diverted to competitors such as Broadcom and Marvell Technology, according to the sources. The company has reportedly already held discussions with Amazon.com, Meta, Microsoft, Google, and OpenAI about the possibility of creating custom chips for them. Additionally, Nvidia has shown interest in sectors beyond data centers, including telecom, automotive, and video game clients.
While Nvidia’s H100 prices remain undisclosed, they are higher than the prior-generation A100 chips, which can be sold for $16,000 to $100,000 depending on volume and other factors. Meta plans to increase its total stock to 350,000 H100s this year.
The creation of a custom unit within Nvidia is spearheaded by Dina McKinney, a former executive at Advanced Micro Devices and Marvell. McKinney’s team aims to provide the company’s technology to customers in various sectors, such as cloud, 5G wireless, video games, and automotives.
The custom chip market in data centers alone is set to reach up to $10 billion this year, and the broader custom chip market was valued at around $30 billion in 2023, accounting for approximately 5% of global chip sales, according to analysts. Notably, Broadcom and Marvell currently dominate the custom silicon design sector for data centers.
Nvidia’s expansion into this realm poses a potential threat to the sales of Broadcom and Marvell, which respectively generate around $10 billion and $2 billion from their custom silicon businesses. As more competition enters the market, the landscape is set to become increasingly challenging for these companies.
Beyond data centers, Nvidia has been in discussions with Ericsson, a leading telecom infrastructure builder, to collaborate on a wireless chip incorporating Nvidia’s graphics processing unit (GPU) technology, insiders claim.
While telecom’s custom chip market is predicted to remain stable at around $4 billion to $5 billion annually, the custom auto market is expected to grow at a steady rate of 20% per year, reaching $6 billion to $8 billion. Similarly, the custom chip market in video games, which could further expand with the introduction of next-generation consoles from Xbox and Sony, is estimated to be around $7 billion to $8 billion.
As Nvidia continues to make strides in various sectors, including cloud computing, telecom, automotive, and video games, its foray into custom chip design reflects its commitment to cater to the specific needs and demands of its clients while keeping pace with evolving market trends.
