AMD Makes a Bold Move with New AI Chip to Challenge Nvidia's Dominance

AMD Makes a Bold Move with New AI Chip to Challenge Nvidia's DominanceOn Thursday, October 11th, AMD unveiled a new artificial intelligence chip aimed at challenging Nvidia's dominance in the data center GPU market. The chip, dubbed Instinct MI325X, is expected to begin production by the end of 2024

AMD Makes a Bold Move with New AI Chip to Challenge Nvidia's Dominance

On Thursday, October 11th, AMD unveiled a new artificial intelligence chip aimed at challenging Nvidia's dominance in the data center GPU market. The chip, dubbed Instinct MI325X, is expected to begin production by the end of 2024. If developers and cloud computing giants view AMD's AI chip as a viable alternative to Nvidia's offerings, it could put pressure on Nvidia's pricing strategy. Currently, Nvidia enjoys a gross margin of 75%, and its GPU demand has remained strong over the past year.

Generative AI applications like OpenAI's ChatGPT require data centers backed by a massive number of GPUs, creating a new market opportunity for AI chip companies. Nvidia has long held the leading position in the data center GPU market, with AMD following closely behind. Now, AMD is setting its sights on capturing more market share from its rival, Nvidia. This market is projected to soar to a value of \$500 billion by 2028.

"The demand for AI is growing at an extraordinary pace, much faster than we had anticipated. Clearly, the global investment is ramping up," said AMD CEO Lisa Su at the launch event. While AMD didn't disclose a list of new customers, it has previously confirmed that Meta and Microsoft have purchased its AIGPUs, and OpenAI has adopted the chip series for some applications.

Furthermore, AMD didn't reveal the pricing information for Instinct MI325X, which is typically sold as part of a complete server solution. With the launch of MI325X, AMD is accelerating its product release pace, planning to release new chips every year to better compete with Nvidia and fully capitalize on the booming AI chip market.

The MI325X is the successor to last year's MI300X series. AMD plans to launch MI350 in 2025 and MI400 in 2026. The release of MI325X signifies AMD's direct competition with Nvidia's upcoming Blackwell chip. Nvidia has confirmed that Blackwell will ship in large quantities early next year. This launch is expected to attract investors eager to capitalize on the AI boom.

As of now, AMD's stock price has increased by 20% in 2024, while Nvidia's stock price has skyrocketed by over 175%. Nvidia is estimated to hold over 90% of the data center AI chip market share. On Thursday, AMD's stock closed down by 4%, while Nvidia's stock saw a slight gain of roughly 1%.

AMD's biggest challenge lies in Nvidia's CUDA programming language, which has become the standard for AI developers, making it difficult for them to break free from Nvidia's ecosystem. To address this challenge, AMD announced this week that it is continuously optimizing its competing software ROCm. This initiative aims to help developers seamlessly migrate AI models to AMD chips, known as its "accelerators."

AMD emphasizes that its AI accelerators are more competitive in tasks like content generation or prediction rather than processing terabytes of data to optimize models. This advantage stems from AMD's advanced memory technology incorporated into its chips, enabling it to support Meta's Llama AI model faster than Nvidia in certain instances. Su pointed out, "Data shows that the MI325 platform delivers 40% better inference performance on Meta's Llama3.1 large language model compared to Nvidia's H200."

While AI accelerators and GPUs are receiving significant attention, AMD's core business remains central processing units (CPUs), a critical component found in nearly every server across the globe. AMD disclosed in July that its data center sales doubled year-over-year in the second quarter to reach \$2.8 billion, with AI chips contributing approximately \$1 billion. AMD currently holds about a 34% share of the data center CPU market, trailing behind industry leader Intel, which maintains dominance with its Xeon chips.

To alter this landscape, AMD launched its new EPYC fifth-generation CPU series on Thursday. The series spans from a low-cost eight-core chip to a 192-core high-performance processor designed for supercomputers, priced from \$527 to \$14,813. AMD claims that this new CPU excels at handling data processing for AI workloads. It's important to note that nearly all GPUs require coordination with the system's CPU while running. Su highlighted, "Current AI technology heavily relies on CPU performance, especially in applications like data analytics."