Demand for high-bandwidth memory is set to explode in the coming quarters and years due to the widespread adoption of artificial intelligence in general and generative AI in particular. SK Hynix is leading shipments of this type of memory and could be the main beneficiary of the HBM rally as he holds a 50% share in 2022. trend force.
Analysts at TrendForce believe AI server shipments powered by computing GPUs such as Nvidia’s A100 and H100 will grow about 9% year-over-year in 2022. They now estimate that the rise of generative AI will drive demand for AI servers, and that this market will grow 15.4% in 2023 and continue to grow at a compound annual growth rate of 12.2% through 2027. .
The surge in AI server usage will also increase demand for all types of memory, including commodity DDR5 SDRAM, HBM2e and HBM3 for computing GPUs, and 3D NAND memory for high-performance, high-capacity storage devices.
TrendForce estimates that general-purpose servers have 500 GB to 600 GB of commodity memory, while AI servers use 1.2 TB to 1.7 TB. In addition, such machines use computing GPUs with more than 80 GB of his HBM2e/HBM3 memory. Each AI machine comes with multiple computing GPUs, so the total HBM content per box is currently between 320GB and 640GB, with accelerators like AMD’s Instinct MI300 and Nvidia H100 NVL more HBM3 It’s equipped with memory, so it’s set to increase even more.
Speaking of HBM3 adoption, it should be noted that SK Hynix is currently the only manufacturer mass-producing this type of memory, according to TrendForce. As a result, as the demand for this type of memory increases, it offers the most benefits. Last year, SK Hynix accounted for 50% of HBM shipments, followed by Samsung at 40% and Micron at 10%. The company will consolidate its position this year, controlling 53% of HBM shipments, while Samsung’s and Micron’s shares will drop to 38% and 9% respectively, TrendForce claims.
Currently, AI servers are mainly used by major US cloud service providers such as AWS, Google, Meta, and Microsoft. As more companies launch generative AI products, it is inevitable that they will have to use AI servers either on-premises or on AWS or Microsoft. For example, Baidu and ByteDance will introduce generative AI products and services in the coming quarters.
sauce: trend force