AMD introduces the Helios rack and MI455X accelerator in an effort to bridge the gap with Nvidia.

      TL;DR: At the Advancing AI event, AMD introduced MI455X chips, Helios racks, and Venice Epyc CPUs, with pledges from OpenAI and Cerebras to implement the hardware extensively.

      On Thursday during its Advancing AI event in San Francisco, AMD presented a range of new data center products, asserting they will surpass Nvidia's competing hardware in AI training and inference tasks. The announcements included the MI455X AI accelerator, the Helios server rack capable of housing 72 of these chips in one system, and the Venice generation of Epyc server processors manufactured using TSMC's two-nanometre process. AMD estimates that the total market it is targeting could reach $2 trillion by 2030, with AI accelerators alone projected to account for over a trillion dollars of that total.

      Despite these announcements, investors reacted negatively, with AMD shares dropping approximately four percent during the presentation, even after the stock more than doubled in value this year. This skepticism highlights the disparity between AMD's aspirations and Nvidia's market leadership: Nvidia’s Vera Rubin platform is already fully operational and being supplied to clients like OpenAI, CoreWeave, and Microsoft. Although AMD CEO Lisa Su has delivered impressive growth, Wall Street's expectations were reflected in a stock that had increased about 145 percent year-to-date before the event.

      Su invited executives from key AI firms to endorse AMD's hardware. OpenAI’s Sachin Katti expressed his expectations for deploying Helios "at massive scale" as the company seeks to enhance its infrastructure capacity. This event followed AMD's announcement of a potential $5 billion investment in Anthropic, alongside plans to deploy two gigawatts of MI450 series GPUs within Helios racks to support the Claude model, with the first gigawatt expected to be delivered in the first half of 2027.

      Additionally, AMD revealed a partnership with Cerebras, known for its wafer-scale chips that excel in ultra-fast AI inference, to integrate Helios racks with Cerebras servers for clients requiring minimal latency. AMD's hardware will manage query interpretation while Cerebras chips will deliver quick responses. Cerebras CEO Andrew Feldman noted that this combined solution will start shipping from Cerebras-operated data centers in the fourth quarter, aiming to outpace a similar offer from Nvidia resulting from its recent Groq acquisition.

      On the other hand, the Venice Epyc processor is AMD's response to Nvidia's entry into server CPUs with the Vera chip, which Nvidia claims provides a performance advantage. AMD asserted that Venice will maintain a significant lead over Vera, initiating a direct competition that independent benchmarks have yet to resolve. This rivalry encompasses accelerators, server processors, and comprehensive rack-scale systems, with both companies promoting end-to-end data center solutions rather than just individual components.