AMD launches new AI chip to challenge Nvidia's dominance in the data centre market

Advanced Micro Devices (AMD) has made a significant stride in the artificial intelligence (AI) sector by unveiling its new Instinct MI325X chip during Computex 2024 in Taipei. This launch positions AMD directly against Nvidia, a leader in the data centre graphics processing unit (GPU) market, as the demand for advanced AI technologies continues to escalate as detailed in a report by CNBC.

The CNBC report further stated that AMD’s chairperson and CEO, Lisa Su, highlighted the growing need for powerful AI solutions, stating, "Demand for AI has continued to escalate and has surpassed expectations. It is evident that investment rates are on the rise across the board." The MI325X is expected to commence production by the end of 2024, aiming to provide developers and cloud service providers with a competitive alternative to Nvidia’s GPUs.

The surge in generative AI technologies, such as OpenAI's ChatGPT, has created an urgent demand for robust data centres capable of handling extensive processing requirements. Historically, Nvidia has dominated this sector, capturing over 90 per cent of the data centre AI chip market. However, AMD is determined to increase its market share in what it estimates will become a $500 billion industry by 2028.

AMD's strategy includes expediting its product release timeline, with plans to launch new chips annually. The MI325X follows the MI300X, which began shipping late last year. Furthermore, AMD has announced upcoming chips designated as MI350 for 2025 and MI400 for 2026. This aggressive approach aims to enhance AMD's competitiveness against Nvidia’s forthcoming Blackwell chips, which are set to begin shipping in significant quantities early next year.

While no new major clients were announced at the event, AMD previously disclosed partnerships with companies like Meta and Microsoft that utilise its AI GPUs. The pricing details for the MI325X remain undisclosed but are expected to be competitive as AMD seeks to disrupt Nvidia’s pricing strategy.

One of the challenges facing AMD is Nvidia's proprietary programming language, CUDA, which has become a standard among AI developers. To counter this obstacle, AMD is enhancing its ROCm software platform to facilitate smoother transitions for developers switching from Nvidia’s ecosystem. Su noted that AMD’s AI accelerators are particularly advantageous for scenarios involving content generation or predictive modelling.

The MI325X boasts up to 40 per cent greater inference performance compared to Nvidia's H200 Llama3 model, making it an appealing option for developers focused on training and deploying sophisticated AI models. “The MI325 platform provides up to 40 per cent greater inference performance compared to H200 Llama3,” Su stated

Despite the focus on AI chips, AMD remains committed to its core business of central processors (CPUs), which are essential for server operations globally. In the June quarter of this year, AMD reported that its data centre sales more than doubled from the previous year, reaching $2 billion, with AI chips contributing approximately $1 billion.

AMD also announced a new series of CPUs known as EPYC 5th Gen during Computex 2024. These CPUs come in various configurations designed to support data processing in AI workloads effectively. With prices ranging from $527 for an economical 8-core model to $14,813 for high-performance 192-core processors aimed at supercomputers, these new offerings reflect AMD’s ambition to reshape its position in the market.

As AMD continues to innovate and expand its product line, it aims not only to capture a larger share of the lucrative data centre market but also to establish itself as a formidable competitor in the rapidly evolving landscape of artificial intelligence technology.