barrons.com
Western Dig: Age of A.I. Could Sell a Lot of Disk Drives, Says Guggenheim
By Tiernan Ray
March 21, 2018 7:38 p.m. ET
Spending on cloud computing is set to heat up this year, with spending by companies such as Amazon ( AMZN), Alphabet’s ( GOOGL) Google, and Apple ( AAPL) rising by perhaps 33%, to $85 billion, according to a report out this evening from Guggenheim Securities analyst Robert Cihra.
That could be especially good for Western Digital ( WDC), among the names he follows, whose shares he rates a Buy.
That 33% spending rise would be a jump up from 18% last year, writes Cihra, and a lot of it is thanks to the rise of machine learning and artificial intelligence.
As Cihra puts it, "the two primary drivers of cloud capex to date have been #1 Consumer Cloud Service Providers (CSPs) (e.g., AAPL, GOOG, AMZN, FB) growing/expanding data center capacity and increasingly using their cash/size as a competitive weapon, and #2 Enterprise CSPs (e.g., AWS, Azure) building out public cloud capacity (IaaS, PaaS, SaaS) to take over traditional on-premises IT workloads."
But now, there is "new INCREMENTAL” spend, he writes: “the #3 driver is CSPs now building out compute infrastructure for AI/ML (both for internal use and MLaaS), which may be driving the recent reacceleration in cloud capex."
Cihra points out that this third area, collectively ML and A.I., is propelling the need for “massive” amounts of storage:
TRAINING a ML model typically involves feeding massive amounts of data through high- powered processors running in servers inside cloud datacenters, where data provide examples to an algorithm that learns by recognizing patterns. Several popular open-source deep learning frameworks include Google’s TensorFlow, Caffe, Apache MXNet (AWS’ framework of choice), Theano, Torch, Microsoft’s Cognitive Toolkit (CNTK), et al, but it takes enormous amounts of processing (GPUs, DRAM), data (HDDs, NAND) and time (often running for days or weeks) to first TRAIN a large ML model that can ultimately then be used for making predictions (inference).
And solid-state storage, meaning NAND flash, can only satisfy so much of that demand, given its relatively high cost, reasons Cihra.
A lot more may fall to the traditional spinning magnetic disk drive.
And that can be very good for Western’s profits even more than its drive shipments:
Driving STORAGE, we believe cloud builders account for the majority of high- capacity nearline HDD demand, which is important not only because those HDDs represent the market’s fastest-growing product, but also because they are the most defensible vs. SSDs (where $/GB still trumps $/IOPS), highest-priced and most profitable for HDD suppliers (e.g., high platter-count, rigorous testing). For example, we estimate nearline HDDs accounting for 13% of WDC’s HDD units in FY18E but 31% of its HDD revenue and 43% of HDD gross profits. Similarly, we estimate nearline HDDs accounting for 15% of STX’s HDD units but 32% of HDD revenue and 42% of HDD gross profits. Illustrating the positive mix impact, we estimate a nearline HDD generates 7X the gross profit/unit of a notebook HDD.
Cihra has a Neutral rating on shares of Western's competitor in drives, Seagate Technology ( STX).
Western shares today are down 83 cents, or 0.8%, at $102.84 |
