AMD on Thursday launched a new artificial intelligence chip that aims directly at Nvidia’s data center graphics processors, known as GPUs.
AMD said Thursday at an event announcing the new product that the Instinct MI325X, as the chip will be called, will begin production by the end of 2024. If AMD’s AI chips are seen by developers and cloud giants as a near-alternative to AI chips; Nvidia’s This could put pricing pressure on Nvidia, which has enjoyed roughly 75% gross margins over the past year while demand for GPUs has been high.
Advanced generative AI, such as OpenAI’s ChatGPT, requires large data centers full of GPUs to do the necessary processing, creating demand for more companies to provide AI chips.
For the past few years, Nvidia has dominated most of the data center GPU market, while AMD has historically been in second place. Now, AMD is aiming to take share from its Silicon Valley rivals, or at least capture a large chunk of a market that will be worth $500 billion by 2028.
“Demand for AI is actually continuing to grow, and is actually exceeding expectations. It’s clear that investment rates continue to grow everywhere,” AMD CEO Lisa Su said at the event. said.
AMD did not reveal any new major cloud or internet customers for Instinct GPUs at the event, but the company has previously revealed both. meta and microsoft It has purchased the company’s AI GPUs, and OpenAI is using them for some applications. The company also did not disclose the price of the Instinct MI325X, which is typically sold as part of a complete server.
With the launch of MI325X, AMD is accelerating its product schedule of releasing new chips on an annual schedule to strengthen competition with Nvidia and capitalize on the boom in AI chips. The new AI chip will be the successor to the MI300X, which began shipping late last year. AMD’s 2025 chip will be called MI350, and the 2026 chip will be called MI400, the company said.
The MI325X rollout will compete with NVIDIA’s next Blackwell chip, which NVIDIA has announced will begin mass shipments early next year.
A successful launch of AMD’s latest data center GPUs could attract the attention of investors looking for more companies to benefit from the AI boom. AMD is up only 20% so far in 2024, while Nvidia’s stock is up over 175%. Most industry estimates put Nvidia at over 90% of the data center AI chip market.
AMD stock fell 4% on Thursday. Nvidia stock rose about 1%.
AMD’s biggest hurdle to gaining market share is that rival chips use its own programming language, CUDA, which has become the standard among AI developers. This essentially locks developers into Nvidia’s ecosystem.
In response, AMD announced this week that it is improving a competing software called ROCm to make it easier for AI developers to switch more AI models to chips AMD calls accelerators.
AMD sees its AI accelerators as more competitive for use cases where AI models create content or make predictions than when AI models process terabytes of data to make improvements. are. Part of that is due to the advanced memory AMD is using on its chips, which the company said allows it to deliver Meta’s Llama AI models faster than some Nvidia chips.
“We see that the MI325 platform delivers up to 40% higher inference performance than H200 on Llama 3.1,” Su said. meta Large-scale language AI models.
Challenge to Intel
While AI accelerators and GPUs have become the hottest parts of the semiconductor industry, AMD’s core business has always been the central processors (CPUs) at the heart of nearly every server in the world.
AMD’s June quarter data center revenue more than doubled over the past year to $2.8 billion, of which AI chips accounted for only about $1 billion, the company announced in July.
According to the company, AMD accounts for about 34% of the total amount spent on data center CPUs. it is still less than intelthe company remains the market leader with its Xeon chip series. AMD aims to change that with a new series of CPUs called EPYC 5th Gen, also announced Thursday.
These chips come in a variety of configurations, from a low-cost, low-power 8-core chip at $527 to a 192-core, 500-watt processor for supercomputers at $14,813 per chip.
AMD says the new CPUs are especially good at feeding data to AI workloads. Almost all GPUs require a CPU on the same system to boot the computer.
“AI today is really focused on CPU power, and you see that in a lot of data analytics and those types of applications,” Su said.
