Nvidia (NVDA.O) is venturing into a new domain by establishing a fresh business unit dedicated to crafting tailor-made chips for cloud computing firms and other entities, including cutting-edge artificial intelligence processors, as per insights from nine reliable sources familiar with the company's strategy.
As the leading global designer and supplier of AI chips, Nvidia aims to seize a portion of the rapidly expanding market for custom AI chips while safeguarding itself from the burgeoning interest among companies in exploring alternatives to its offerings.
Based in Santa Clara, California, the company currently commands approximately 80% of the high-end AI chip market, a position that has propelled its market value up by 40% this year to $1.73 trillion, following a more than threefold increase in 2023.
Major players like OpenAI, Microsoft (MSFT.O), Alphabet (GOOGL.O), and Meta Platforms (META.O) have been scrambling to acquire Nvidia chips to stay competitive in the swiftly evolving generative AI sector.
Nvidia's H100 and A100 chips serve as versatile, general-purpose AI processors for many of these major customers. However, these tech giants have begun developing their own in-house chips tailored to specific requirements. This approach helps reduce energy consumption and potentially cuts down on both cost and design time.
Now, Nvidia is positioning itself to assist these companies in developing custom AI chips, a market segment that has been captured by rival firms like Broadcom (AVGO.O) and Marvell Technology (MRVL.O), according to undisclosed sources familiar with the matter.
Greg Reichow, a general partner at venture capital firm Eclipse Ventures, highlighted the importance of tailoring compute power to specific needs, stating, "If you're really trying to optimize on things like power, or optimize on cost for your application, you can't afford to go drop an H100 or A100 in there."
While Nvidia doesn't disclose the prices of its H100 chips, they are reportedly priced higher than the previous-generation A100 chips, ranging from $16,000 to $100,000 per chip depending on volume and other factors. Meta has announced plans to procure 350,000 H100 chips this year.
Nvidia has engaged in discussions with representatives from major tech companies including Amazon.com (AMZN.O), Meta, Microsoft, Google, and OpenAI regarding the possibility of creating custom chips for them, according to two knowledgeable sources. In addition to data center chips, the company has also targeted customers in telecom, automotive, and video game sectors.
In 2022, Nvidia announced a program allowing third-party customers to integrate some of its proprietary networking technology with their chips. However, the company has remained silent about the program since then, and Reuters is now shedding light on its broader ambitions for the first time. A spokesperson for Nvidia declined to comment beyond the company's 2022 announcement.
Dina McKinney, a former executive at Advanced Micro Devices (AMD.O) and Marvell, leads Nvidia's custom unit. Her team's objective is to make the company's technology accessible to customers in various sectors, including cloud, 5G wireless, video games, and automotive industries, according to her LinkedIn profile. However, mentions of these details were removed and her title was altered after Reuters reached out to Nvidia for comment.
Amazon, Google, Microsoft, Meta, and OpenAI all declined to comment on the matter.
With estimates suggesting that the data center custom chip market could reach $10 billion this year and double by 2025, and the broader custom chip market valued at approximately $30 billion in 2023, representing roughly 5% of global chip sales, Nvidia is poised to capitalize on this lucrative segment of the industry.
Tags
tech