As the first company to ship HBM3E memory, Micron revealed that its advanced high-bandwidth memory (2024) has been sold out, and most of the production in 2025 has been allocated. Micron's HBM3E memory (or as Micron is also called, HBM3Gen2) is one of the first memories to be certified for Nvidia's latest H200/GH200 accelerators, so it looks like the DRAM maker will become an important supplier to the company in the spotlight.
"Our 2024 HBM has been sold out and the vast majority of our 2025 supply has been allocated," Micron CEO Sanjay Mehrotra said in prepared remarks for this week's earnings call. "We still expect that at some point in 2025, HBM bit share will be comparable to our overall DRAM bit share."
Micron's first HBM3E product is an 8-Hi24GB stack with a 1024-bit interface, 9.2GT/s data transfer rate and 1.2TB/s total bandwidth. Nvidia's H200 accelerator for artificial intelligence and high-performance computing will use six of these cubes, providing a total of 141GB of accessible high-bandwidth memory.
"We are on track to generate hundreds of millions of dollars in revenue from HBM in fiscal 2024, and HBM revenue is expected to be accretive to our DRAM and overall gross margin starting in the fiscal third quarter," Mehrotra said.
The company has also started sampling its 12-Hi 36GB stack, which offers 50% more capacity. These KGSDs will be mass-produced in 2025 and will be used in next-generation artificial intelligence products. Meanwhile, Nvidia's B100 and B200 don't look like they'll be using the 36GB HBM3E stack, at least initially.
Demand for AI servers hit records last year and looks set to remain high this year as well. Some analysts believe that Nvidia's A100 and H100 processors (and their various derivatives) will account for up to 80% of the entire artificial intelligence processor market in 2023. While Nvidia will face stiffer competition in inference this year from AMD, AWS, D-Matrix, Intel, Tenstorrent and others, Nvidia's H200 looks set to remain the processor of choice for AI training, especially for big companies like Meta and Microsoft, which already run fleets of hundreds of thousands of Nvidia accelerators. With this in mind, becoming the primary supplier of HBM3E for Nvidia's H200 is a big deal for Micron, as it allows Micron to finally capture a sizable share of the HBM market.
At the same time, since each DRAM device in the HBM stack has a wider interface, it is larger in physical size than an ordinary DDR4 or DDR5 IC. Therefore, the mass production of HBM3E memory will affect the bit supply of Micron's commodity DRAM.
"The growth in HBM production will limit the supply growth of non-HBM products," Mehrotra said. "Looking at the industry as a whole, HBM3E consumes approximately three times the wafer supply than DDR5 to produce a given number of bits on the same technology node."