Nvidia, the dominant global designer and supplier of AI chips, is launching a new business unit focused on developing custom chips for cloud computing firms and other industries, according to sources familiar with the matter. The move comes as Nvidia aims to capture a portion of the expanding market for bespoke AI chips and protect itself against growing competition from companies developing alternatives to its products.
With around 80% of the high-end AI chip market under its control, Nvidia’s stock market value has surged by 40% this year to $1.73 trillion, after more than tripling in 2023. Nvidia’s customers, including the likes of Microsoft, Alphabet, and Meta Platforms, have been racing to secure its chips to gain a competitive edge in the rapidly emerging generative AI sector.
Nvidia’s H100 and A100 chips have been serving as generalized, all-purpose AI processors for many of its major customers. However, these tech companies have started developing their own tailored chips to meet their specific requirements. Designing custom chips can lead to energy efficiency improvements, cost reductions, and faster design cycles.
Now, Nvidia is looking to play a role in helping these companies create their own custom AI chips, which have been increasingly turning to competitors like Broadcom and Marvell Technology for such solutions. According to Greg Reichow, general partner at venture capital firm Eclipse Ventures, companies need to have the exact right mixture of compute power for their applications instead of using generalized chips like the H100 or A100.
While Nvidia does not disclose the pricing for its H100 chips, it is known that these chips can sell for between $16,000 and $100,000 depending on factors like volume. Meta, for example, plans to increase its H100 stock to 350,000 units this year.
Nvidia has reportedly held meetings with representatives from major tech firms, including Amazon.com, Meta, Microsoft, Google, and OpenAI, to discuss the development of custom chips for their specific needs. In addition to data center chips, Nvidia has also been eyeing opportunities in the telecom, automotive, and video game sectors.
The custom chip market is estimated to be worth around $30 billion, equivalent to roughly 5% of annual global chip sales. Currently, Broadcom and Marvell dominate the market for custom silicon design for data centers. However, Nvidia’s entry into this space has the potential to eat into their market share, posing a significant threat to these companies.
Nvidia is also reportedly in talks with telecom infrastructure builder Ericsson to collaborate on a wireless chip that incorporates Nvidia’s graphics processing unit (GPU) technology. Furthermore, the company has plans to target the automotive and video game markets with its custom chips.
The telecom custom chip market is expected to remain relatively flat at around $4 billion to $5 billion annually, according to estimates. On the other hand, analysts project steady growth for the custom auto market, which currently ranges from $6 billion to $8 billion, at a rate of 20% per year. Similarly, the custom chip market for video games, currently valued at $7 billion to $8 billion, is expected to expand with the release of next-generation consoles from Xbox and Sony.
Overall, Nvidia’s strategic move into the custom chip market demonstrates its commitment to serving the evolving needs of its customers. By offering tailor-made solutions, Nvidia aims to maintain its dominance in the AI chip sector while also capitalizing on exciting growth opportunities in various industries.
