Memory specialist Micron (NASDAQ:MU) sells both DRAM, a type of computer memory that’s used in virtually every kind of computing device, and NAND flash, which is rapidly gaining traction for high-performance data storage applications as it’s quicker and more efficient than hard disk drive-based storage.  Micron’s business has continued to benefit from what seems like an insatiable amount of demand for both DRAM and NAND in applications such as mobile phones and data center servers. One of the sub-segments within data center servers is the market for servers that handle machine learning, commonly referred to as artificial intelligence, processing tasks.

That sub-segment is small today, with data center chip giant Intel estimating the market at around 7% of total data center server shipments in 2016, but it’s also, according to Intel, the fastest growing. The companies that make the processors that perform these machine learning computations are clearly very excited about the artificial intelligence opportunity as it means they’ll get to sell a lot more computing power over the years.

It’s not surprising, then, that Micron, which makes the memory chips those processors will rely on to handle those computations, is also excited by this opportunity. Let’s go over what Micron had to say about its view of the artificial intelligence opportunity at its most recent analyst day .

Micron says that a typical machine learning training workload will require six times the amount of DRAM that a “standard cloud server” needs and twice as much solid state drive capacity. Those are pretty huge memory content increases.

In a graphic that Micron included in its analyst day presentation, it conceded that “AI-capable” servers made up just a small fraction of overall server shipments in calendar year 2017. However, it expects that percentage to grow substantially by calendar year 2021 and to get close to half of all server shipments by 2025.  It’s important to keep in mind that making predictions is hard and the chances that those predictions will be wrong go up the further out they are. Read more from…

thumbnail courtesy of