On June 13, AMD announced that production of its new artificial intelligence chip, the MI300X, will increase in the fourth quarter, aiming to compete with market leader Nvidia. Lisa Su, the CEO of AMD, revealed that the MI300X chip will boast 192 gigabytes of memory, surpassing the capacity of any current Nvidia chip, a significant performance metric for handling large AI systems like ChatGPT.
Su emphasized the pivotal role of AI in driving silicon consumption in the foreseeable future. She stated that sample chips will be available to customers in the third quarter, with full-scale production ramping up towards the end of the year.
Following AMD's update on its AI strategy, Nvidia experienced a 3.9% surge, reaching a market capitalization exceeding $1 trillion and becoming the first chipmaker to achieve this milestone. Conversely, AMD's stock dropped by 3.6%. However, AMD's company shares have doubled since the beginning of the year, buoyed by the optimism surrounding increased AI investments, reaching a 16-month high earlier that day.
Su also highlighted a system that integrates eight MI300X chips into a single computer, presenting a competitive alternative to Nvidia's comparable offerings.
Furthermore, AMD announced that it has commenced shipping significant quantities of its general-purpose central processor chip, "Bergamo," to companies like Meta Platforms. Alexis Black Bjorlin, overseeing computing infrastructure at Meta, confirmed the adoption of the Bergamo chip, which targets a different sector of AMD's data center business catering to cloud computing providers and large-scale chip purchasers.
Nvidia currently dominates the AI computing market, capturing an estimated 80% to 95% market share, with its shares surging 170% year-to-date. While Intel, Cerebras Systems, and SambaNova Systems pose as competitors, Nvidia's primary threats arise from internal chip development initiatives within Google (Alphabet Inc.) and Amazon.com's cloud division, both of which lease custom chips to external developers.
Nvidia's success stems not only from its chips but also from over a decade of providing software tools to AI researchers and anticipating their chip-related needs during the lengthy design process. In contrast, AMD unveiled updates to its Rocm software, a direct competitor to Nvidia's Cuda software platform.
Anshel Sag, an analyst at Moor Insights & Strategy, expressed skepticism regarding AMD's software solution, suggesting that it may not be perceived as competitive with Nvidia's, despite offering comparable hardware performance.
During the presentation, Soumith Chintala, a Meta vice president involved in open-source AI software development, revealed a close collaboration with AMD to simplify the transition for AI developers from relying solely on the dominant vendor's chips to exploring alternative options, such as those from AMD. Chintala highlighted that the shift between platforms requires minimal effort or, in many cases, no effort at all.
