Nvidia to launch a new business unit for custom chip design, report
Datacentre custom chip market could swell to $10 billion this year and potentially double by 2025
American GPU maker Nvidia is reportedly gearing up to launch a dedicated unit aimed at crafting custom chips tailored to the unique needs of cloud computing firms and other clients, with a particular focus on cutting-edge AI processors.
Sources familiar with Nvidia's plans told Reuters that the move is strategically aimed at capitalising on the burgeoning market for custom AI chips, estimated to be worth a staggering $30 billion, while safeguarding its position as a dominant force in the field.
Based in Santa Clara, California, Nvidia currently commands an impressive 80% share of the high-end AI chip market, a feat that has propelled its stock market value by 40% this year alone, reaching $1.73 trillion after its value more than tripled in 2023.
While Nvidia's H100 and A100 chips have served as versatile, all-purpose AI processors for many of its major customers, such as Microsoft, Google and Meta, there has been a noticeable trend towards developing internal chips to optimise performance and efficiency across various sectors.
Recognising this paradigm shift, Nvidia is now actively engaging with these companies to assist in the development of custom AI chips, a move that could potentially challenge rival firms such as Broadcom and Marvell Technology.
Greg Reichow, general partner at venture capital firm Eclipse Ventures, underscored the importance of fine-tuning compute resources to match specific application needs, highlighting the necessity for custom chip solutions.
"If you're really trying to optimise on things like power, or optimise on cost for your application, you can't afford to go drop an H100 or A100 in there," Reichow told Reuters. "You want to have the exact right mixture of compute and just the kind of compute that you need."
Projections from research firm 650 Group indicate that the datacentre custom chip market could swell to $10 billion this year and potentially double by 2025.
Moreover, analysts from Needham estimate that the overall custom chip market reached a valuation of approximately $30 billion in 2023, representing about 5% of the global annual chip sales.
Nvidia's ambitions extend beyond datacentre chips, with the company pursuing opportunities in the telecom, automotive and video game sectors.
Talks with telecom infrastructure builder Ericsson for a wireless chip incorporating Nvidia's GPU technology are already underway, signalling Nvidia's intent to diversify its offerings and target new growth avenues.
Nintendo's popular Switch handheld console, which already features Nvidia's Tegra X1 chip, is expected to receive a new version this year with an Nvidia custom design.
As Nvidia sets its sights on carving out a larger slice of this lucrative market, industry observers await the ripple effects of this bold move, which poised to reshape the semiconductor landscape in the years to come.
Last month, Intel CEO Pat Gelsinger revealed Intel's ambitious goal of achieving up to a 6x improvement in AI performance across successive processor generations.
In November, Microsoft unveiled new AI chips to support services like OpenAI and Copilot. They include a general purpose chip based on Arm designs, called Cobalt, and a specialised AI CPU named Maia, both of which will be used in Microsoft's Azure datacentres this year.