NVIDIA Establishes New Division to Focus on Custom Chips for AI Needs

NVIDIA is launching a new business unit dedicated to customizing processors for the artificial intelligence and other workload needs of large companies. According to exclusive reports from Reuters, NVIDIA has been in talks with major companies such as Amazon, Meta, Microsoft, Google, and OpenAI for the production of custom silicon solutions.

The new NVIDIA division will be led by semiconductor veteran Dina McKinney. McKinney, who previously served as Vice President responsible for AMD CPU design and Marvell infrastructure processors, will oversee the team developing custom chips for cloud computing, 5G telecommunications, gaming, automotive, and other fields.

NVIDIA's dominant position in the AI chip market has been a major driver of its skyrocketing market value, which has grown by 40% to reach $1.73 trillion this year alone. The company has attracted heavyweight companies such as OpenAI, Microsoft, Alphabet, and Meta Platforms, all of which utilize NVIDIA's H100 and A100 chips for generative AI work.

However, while NVIDIA's GPUs have a dominant position in AI training and high-performance computing in today's cloud computing field, more and more large-scale vendors are introducing their own custom designs. Domestic chipsets can offer more optimization in terms of cost, performance, and power requirements for certain workloads.

In fact, Amazon, Google, and Microsoft have already deployed custom chips based on NVIDIA processors in their data centers. Allowing these major customers to directly utilize NVIDIA's intellectual property to customize solutions can prevent further erosion of NVIDIA's data center stronghold.

Greg Reichow of Eclipse Ventures said, "If you really want to optimize power consumption or optimize costs for your application, you can't just use H100 or A100. You need the right mix of computing power."

The custom chip market, especially the market for data center custom chips, is poised for growth. According to Alan Weckel of 650 Group, the data center custom chip market will reach $10 billion this year and double by 2025. By 2023, the broader custom chip market will be worth around $30 billion, accounting for a significant share of global chip sales.

This move by NVIDIA builds upon its opening of its networking technology last year to integrate into customer custom designs. While details since then have been scarce, discussions with large-scale enterprises indicate that NVIDIA is committed to becoming a one-stop custom chip shop for data centers, AI, 5G, automotive, and other explosively growing markets.

As more cloud computing giants seek internal chip designs, the company will have to work hard to maintain its control over the rapidly evolving AI chip market.