Hardly any demand: Amazon Cloud gets no new AI chips from AMD

Amazon Web Services will not be using any new AI chips from AMD in its cloud for now. According to Gadi Hutt, Senior Director for Customer and Product Development at Amazon's chip division Annapurna Labs, the company is focusing on customer demand. He told Business Insider that he has not yet seen this with AMD's MI300 series.

Instead of AMD: Nvidia and Trainium

At the re:Invent in-house exhibition, Amazon announced that it would be using the next generation of Nvidia's Blackwell graphics units in its EC2 P6 instances. According to the company, these offer 2.5 times higher performance and are optimized for generative AI. At the same time, EC2 announced instances and UltraServers with Trainium2 chips. These come from the Amazon subsidiary Annapurna Labs. Developers there are already working on the successor, which should be more energy-efficient and four times more powerful than the current generation.

When AMD started series production of the Instinct MI300 accelerators last year, the company was full of expectations. The Instinct MI300X, for example, is a low-cost alternative to Nvidia's H100 chip. Although AMD's sales of AI chips are increasing this year, the company is still lagging far behind its competitors. According to Aktien-Magazin, Nvidia leads the AI accelerator sector with a market share of around 80 percent.

Amazon continues to rely on AMD CPUs for servers

Hutt did not rule out the future use of AMD's AI accelerators in an interview with the business magazine. Amazon continues to work closely with AMD in other areas. For example, AWS uses the EPYC series as server processors. Compared to the Xeon CPUs from competitor Intel, AMD processors have more cores and a faster memory subsystem, which gives them advantages in compute- and memory-intensive server instances.

(sfe)