Advanced Micro Devices (AMD) launched a new artificial intelligence chip on Thursday, directly targeting Nvidia’s stronghold in the data centre graphics processor (GPU) market.
The Instinct MI325X chip, expected to begin production before the end of 2024, is positioned to rival Nvidia’s upcoming Blackwell chips, which are anticipated to start shipping in significant quantities early next year.
With Nvidia enjoying approximately 75% gross margins due to the high demand for its GPUs, AMD’s Instinct MI325X could create competitive pressure if it is perceived as a viable alternative by developers and cloud service providers.
The growing demand for AI-powered services, such as OpenAI’s ChatGPT, has led to an increased need for data centres filled with GPUs, sparking the race between semiconductor giants.
Although Nvidia has dominated this market, controlling more than 90% of the data centre AI chip segment, AMD hopes to capture larger portion, eyeing a market 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 at the event.
While AMD did not disclose new cloud or internet customers for its Instinct GPUs during the announcement, it had previously revealed that both Meta and Microsoft utilise its AI chips, with OpenAI employing them for certain applications. Pricing for the MI325X was not revealed, as the chip is typically sold as part of a complete server.
With the MI325X, AMD is accelerating its product release schedule to compete with Nvidia and capitalise on the ongoing AI chip boom. AMD plans to launch new chips annually, with the 2025 model to be named MI350 and the 2026 model dubbed MI400.
Despite its potential, AMD faces challenges. Nvidia’s proprietary programming language, CUDA, has become the industry standard, locking many AI developers into its ecosystem. In response, AMD is working on its own software, ROCm, to encourage developers to shift their AI models to its accelerators.
AMD’s Instinct MI325X is optimised for generating AI content and predictions, thanks in part to its advanced memory, which allows it to run Meta’s Llama AI model faster than some Nvidia chips. “What you see is that MI325 platform delivers up to 40% more inference performance than the H200 on Llama 3.1,” said Su, referring to Meta’s large language model.
In addition to challenging Nvidia, AMD is also taking on Intel in the central processing unit (CPU) market. Although AMD currently holds 34% of the data centre CPU market, Intel’s Xeon processors remain dominant. AMD aims to change this with the launch of its 5th Gen EPYC CPUs, which range from an 8-core chip priced at $527 to a 192-core supercomputer processor costing $14,813.
“Today’s AI is really about CPU capability, and you see that in data analytics and a lot of those types of applications,” Su said.