Nvidia is aiming to launch a custom chips unit for cloud and AI, sources told Reuters. (Photo by Mohd RASFAN / AFP)

Jensen Huang (C), co-founder and chief executive officer (CEO) of Nvidia, speaks during a media roundtable in Kuala Lumpur on December 8, 2023. (Photo by Mohd RASFAN/AFP).

Is Nvidia considering a custom AI chips division?

  • Nvidia is aiming to launch a custom chips unit for cloud and AI, sources tell Reuters.
  • More tech firms want to craft custom chips to cut energy use, costs, and design time. Nvidia aims to aid this trend by helping to develop AI chips for others.
  • According to sources, Nvidia had met with Amazon, Meta, Microsoft, Google, and OpenAI for custom chip talks, with plans to expand beyond data centers to telecom, auto, and gaming.

Nvidia, leader in graphics processing units (GPUs) and AI is considering a significant strategic move: establishing a separate business unit dedicated to building custom AI chips for external clients. This strategic decision marks a substantial shift in the company’s approach, signaling its commitment to meeting customers’ evolving needs in these rapidly expanding sectors.

Nvidia’s dominance in the GPU market has long been undisputed, with its graphics cards powering everything from gaming rigs to supercomputers. Nvidia’s H100 and A100 chips have been the go-to for a range of big clients, serving as versatile AI processors. However, cloud computing and AI demands have grown increasingly specialized, needing tailored solutions to achieve optimal performance and efficiency. 

Recognizing this paradigm shift, Nvidia is planning proactive steps to establish a separate division focused on developing custom chips for these high-growth markets. Several sources have told Reuters recently that Nvidia is planning a division dedicated to custom AI chip development for those markets.

“Nvidia is building a new business unit focused on designing bespoke chips for cloud computing firms and others, including advanced AI processors, according to nine sources familiar with the company’s plans,” Reuters#’ report reads. The top AI chipmaker wants a slice of the booming custom AI chip market – and to defend itself against competitors exploring other options.

This move begs the question: does it make sense for Nvidia to venture into this domain? The company currently dominates 80% of the high-end AI chip market. This has boosted its stock market value by 40% this year to US$1.73 trillion, following a more than triple increase in 2023.

What’s in it for Nvidia, helping companies develop custom AI chips?

Greg Reichow, general partner at Eclipse Ventures, highlighted to Reuters the need for precise application optimization. “You can’t just throw in an H100 or A100 if you’re focused on power or cost efficiency. It’s crucial to have the perfect balance of computing tailored to your needs,” he emphasized.

In other words, customization is critical because one size does not fit all in cloud computing and AI. These fields demand highly specialized hardware solutions tailored to specific workloads and applications. By creating a dedicated custom chips division, Nvidia can offer bespoke solutions that address the unique requirements of its customers, enabling them to maximize performance and efficiency while minimizing costs.

Plus, the demand for AI chips is skyrocketing as organizations integrate AI into their operations across various industries. By offering custom AI chips, Nvidia can tap into a lucrative market opportunity and diversify its revenue streams beyond its traditional GPU business. “Beyond data center chips, Nvidia has pursued telecom, automotive, and video game customers,” the Reuters report reads. 

In 2022, Nvidia announced plans to let third-party customers incorporate its proprietary networking technology into their chips. That means, should Nvidia pursue a custom chips unit, it would present opportunities to collaborate with industry partners. 

According to the Reuters report, Nvidia officials have met with representatives from Amazon.com, Meta, Microsoft, Google, and OpenAI to discuss making custom chips for them.

Besides customization and rapid demand, the pace of innovation in cloud computing and AI has also been relentless.  

New algorithms, frameworks, and applications are constantly emerging, driving the need for cutting-edge hardware that can keep pace with these advancements. With a dedicated focus on custom chips, Nvidia can accelerate the development and deployment of next-generation technologies, giving its customers a competitive edge in an ever-evolving landscape. 

Nvidia’s reputation as a market leader in AI hardware also positions it well to compete in the custom AI chip market. Its deep understanding of AI algorithms and its vast ecosystem of developers give it a competitive edge in designing chips optimized for AI workloads. 

What are the cons of building a standalone custom chip unit?

Building a separate business unit for custom AI chips demands substantial investment in research, development, and manufacturing, meaning Nvidia would have to allocate its resources wisely while maintaining focus on its core GPU business. In an increasingly crowded market, with players like Intel and AMD trying to get more of a foothold, Nvidia must stand out through innovation and service excellence.

Managing the complexities and risks of designing custom AI chips, including technical challenges and market dynamics, is crucial. Additionally, Nvidia faces the task of convincing potential clients of the benefits of custom AI chips and overcoming adoption barriers like cost and compatibility concerns in the nascent market.

Yet, the potential rewards of a custom chip division may outweigh the risks. By using its expertise, its reputation, and its partnerships, Nvidia could establish itself as a critical player in the custom AI chip market and drive future growth and innovation in AI hardware. To achieve any of that though will require careful planning, execution, and adaptation to evolving market dynamics.