Photo source: Flickr
On Thursday, Advanced Micro Devices (AMD) announced the launch of its new artificial intelligence chip, the Instinct MI325X. This chip is designed to directly compete with Nvidia’s data centre graphics processors, commonly referred to as GPUs. Production for the MI325X is expected to commence before the end of 2024.
AMD’s entry into this market could significantly impact Nvidia, which has maintained approximately 75% gross margins due to high demand for its GPUs over the past year. If developers and major cloud service providers view AMD’s offering as a viable alternative, it may lead to pricing pressures on Nvidia.
Advanced generative AI technologies, such as OpenAI’s ChatGPT, have created a substantial need for data centres equipped with GPUs for processing tasks. Historically, Nvidia has held a dominant position in the data centre GPU market, while AMD has typically occupied second place. However, AMD is now targeting a larger share of this market, which is projected to be worth $500 billion by 2028.
“AI demand has actually continued to take off and actually exceed expectations. It’s clear that the rate of investment is continuing to grow everywhere,” said AMD CEO Lisa Su during the product announcement event.
While AMD did not disclose any new major clients for its Instinct GPUs during the announcement, it previously revealed that companies like Meta and Microsoft utilise its AI GPUs. Additionally, OpenAI employs these chips for certain applications. Pricing details for the MI325X remain undisclosed, as it is typically sold as part of a complete server solution.
With the introduction of the MI325X, AMD is shifting to an annual release schedule for its chips in order to keep pace with Nvidia amid the growing demand for AI technologies. The MI325X will succeed the MI300X, which began shipping late last year. Future releases are planned under the designations MI350 for 2025 and MI400 for 2026.
Competitive Scene
The MI325X will compete against Nvidia’s upcoming Blackwell chips, which are set to begin shipping in significant quantities early next year. A successful launch could attract investors looking for opportunities in companies positioned to benefit from the AI boom. So far in 2024, AMD’s stock has increased by just 20%, while Nvidia’s stock has soared over 175%. Industry estimates suggest that Nvidia commands more than 90% of the data centre AI chip market.
On Thursday, AMD’s stock experienced a decline of approximately 4%, while Nvidia shares saw a modest increase of about 1%.
One of AMD’s primary challenges in gaining market share lies in Nvidia’s proprietary programming language, CUDA, which has become a standard among AI developers. This creates a barrier that effectively locks developers into Nvidia’s ecosystem.
In response to this challenge, AMD announced improvements to its competing software platform, ROCm, to facilitate an easier transition for AI developers who wish to switch their models to AMD’s chips—referred to as accelerators. AMD asserts that its AI accelerators are particularly suited for applications where AI models generate content or make predictions rather than simply processing large datasets.
Broader Business Context
While AI accelerators and GPUs have garnered significant attention within the semiconductor industry, AMD’s core business remains focused on central processors (CPUs), which are integral components in nearly every server worldwide.
According to AMD’s July report, data centre sales during the June quarter exceeded $2.8 billion, with AI chips contributing about $1 billion.
AMD currently holds around 34% of total spending on data centre CPUs. However, it still trails behind Intel’s dominance with its Xeon line of chips. To address this, AMD introduced a new line of CPUs called EPYC 5th Gen during the same event.
These CPUs are available in various configurations—from an economical low-power 8-core chip priced at $527 to high-performance 192-core processors designed for supercomputers costing $14,813 each. These new CPUs are particularly optimised for supporting AI workloads since nearly all GPUs require a CPU within the same system for operation.
“Today’s AI is really about CPU capability, and you see that in data analytics and a lot of those types of applications.” Su added.