In a move that will be one of the most watched in the future of AI and hardware, Advanced Micro Devices (AMD) today announced “Helios,” the company’s ultimate vision for an AI server architecture, scheduled for release in 2026, along with a joint commitment with OpenAI, the author of ChatGPT.
The announcement, which came at an AMD “Advancing AI” event, represents a significant challenge to Nvidia’s lead in the fast-growing AI chip business.
AMD boss Dr. Lisa Su also appeared to discuss the chipmaker’s next-generation Instinct MI350 and MI400 series AI chips, taking aim directly at Nvidia’s upcoming Blackwell line of processors. The Helios server, a rack-scale system, will be anchored by MI400 series chips, which AMD is designing for a unified compute engine that can intertwine thousands of the chips.
Dr Su went out of her way to make clear that AMD was pursuing an open ecosystem, with much of the Helios server, including networking standards, available for all comers to use including rivals such as Intel – a stark contrast with Nvidia’s historically proprietary NVLink technology.
One of the highlight moments came from OpenAI CEO Sam Altman, who confirmed his company will be utilizing AMD’s next-gen chips, name-dropping not only AMD’s codenamed MI300X but also the MI450.
Altman was excited about the partnership, which was also described the memory architecture’s relevance for inference and training AI models. The partnership is a significant departure for OpenAI, which has primarily used Nvidia hardware.
Shipping now, the MI350 series is said to offer up to 4x the AI computational power and 35x the inferencing performance compared to prior generations and boasts 288GB of HBM3E memory. When we look forward at 2026, the MI400 series is going to take things even further with 432GB of HBM4 goodness.
AMD is looking to position these offerings as competitively priced competitor for Nvidia’s high-end offerings and is targeting the high growth AI chip market, which it believes will surpass $500 billion by 2028.
This partnership with OpenAI, along with support from the likes of Meta, xAI, Oracle and Microsoft, helps to cement AMD’s legitimacy as a solid participant. With an emphasis on powerful hardware and open software, AMD isn’t just playing bricks, it’s orchestrating an entirely new future for AI development brick by brick through the power of collaborative innovation.