We all hear the buzz around the growing volumes of data and the expected impact that emerging technologies such as augmented reality, artificial intelligence, big data analytics, and machine learning will have on the storage market. When planning for product fit within emerging technologies, storage manufacturers must manage the day-to-day P&L needs against long-term market demands. The economics of the storage business is brutal and efficient, having narrowed a field of 22 companies (that at one time or another manufactured hard drives) down to 3 – Western Digital, Seagate, and Toshiba.
In a perfect world, manufacturers would like to know there is a home for each bit produced, but that is not the reality. With so many emerging technologies, it is difficult to consistently identify the winners from the losers. At the macro level, not only must they decide where to focus, they must plan their supply chain and resources to support next-generation products while maintaining P&L discipline. At the micro level, they navigate customer forecasts, competitors’ activity, and economic trends to allocate resources and plan production volumes. This is a process that inherently generates a margin of error.
When looking for what is next for the storage market, it is no secret that the data center space is where the action is. We saw just how influential the data center space is in the fourth quarter. Q4 2017 started off slowly with Seagate and HGST battling for market share. In December there was a significant shift in market dynamics with a combination of events flipping the supply and demand equation on its head. We saw large hyperscalers entering the market and snapping up volumes of large capacity drives. Simultaneously, the server market kicked into gear as the new tax plan offered incentives for end-users to upgrade their capital equipment. Both market-moving events happened while storage manufacturers were grappling with a worldwide capacitor shortage that challenged their supply chains, creating a quantifiable delta between supply and demand.
NAND (the memory component of SSD) yields continue to improve, with one manufacturer indicating that total NAND yields are exceeding demand by 1%. However, this number can be misleading. NAND yields are sorted into three yield grades; enterprise, client, and third party. When looking at the 1% figure of NAND yield exceeding supply, it is good for the client and third-party SSD manufacturers, as volumes of raw NAND to assemble SSD exceeds demand but not enough NAND is yielding as enterprise-grade NAND to satisfy enterprise demand.
ENTERPRISE
As discussed above in the introduction, after an initial slow start to the 4th quarter, enterprise demand turned around, with demand outpacing supply. The biggest factor driving demand is the hyperscaler sudden uptake of large-cap nearline drives. The DRAM shortage had slowed the hyperscaler’s Q4 capacity expansion. However, in December supply freed up, allowing them to ramp up expansion plans. Alternatively, the HDD manufacturers did expect an increase in demand, forecasting a 40% increase, but not to the magnitude of orders received.
After the hyperscalers received increased DRAM volumes, they subsequently increased HDD orders which came in above manufacturers’ forecasts. This, combined with the increased activity from the server market driven by the new tax plan’s accelerated depreciation rate on capital goods against a capacitor shortage, resulted in an estimated 25% delta between HDD supply and demand. The large data centers are first in line for receiving hard drives, leaving the brunt of the shortage to be felt by the rest of the market.
Performance drives were caught in the updraft too, as manufacturers were getting squeezed by the passive shortage and the new tax plan fueled increased server demand. Further contributing to supply woes was the manufacturers recognizing that SSD is the better solution, reducing support for performance drives. This led to shortfalls that sent end users scrambling. Upon speaking to several builders, they indicated that the situation has somewhat improved over the past month, but there is still a challenge, requiring that they manage their forecast and escalations more closely and more frequently. As HDD manufacturers continue to reduce support for performance drives, supply will only become harder to come by. Toshiba continues to play in this space, reporting strong results.
SSD manufacturers are dropping prices for TLC-based enterprise SSD. However, reports are that there still is not enough supply to support demand. Pure Storage reported (in their recently released results) that they don’t expect to see a notable decline in SSD pricing until the second half of 2018. Despite not having enough supply, fewer manufacturers are lowering pricing to position themselves with end users for when NAND yields improve. MLC-based SSD will prove to be even more problematic as Micron, which supplies Intel with raw NAND, will no longer be supporting MLC-based NAND by the end of 2018. In one conversation with a buyer for a white box server builder, I asked if he was experiencing issues with large-cap enterprise hard drives, and he replied that his issue was with Intel MLC SSD.
As always, when evaluating the storage market, we need to look forward to what is next. On a day-to-day basis, we read how new and exciting emerging technologies such as augmented reality, artificial intelligence, big data analytics, and machine learning will unlock the power of data. Without question, as market demand for these technologies increases and data volumes grow, we will be required to think differently about how we manage data and the growing workload. Much of the data used by emerging technologies will be captured through IoT. Oftentimes, this data is unstructured. Due to inefficiency and cost, it makes better sense to process the unstructured data locally, via edge computing and forwarding on the valuable data to the data center.
In a blog WD touts their air-based drives as ideally suited to edge computing. However, in speaking to many market participants (including builders and component suppliers, excluding surveillance) they are seeing little material benefit to their P&L. Looking forward to data being processed locally via AI or machine learning SSD seems better suited to the task of sorting and organizing data into action data. SSD versus HDD in storage within edge computing will continue to play itself out.
DESKTOP MARKET
The desktop/notebook market (being a mature market that is in decline) is an afterthought that is overshadowed by the potential of the growing cloud storage market. Often when we read the news, we draw sweeping conclusions while missing the subtleties. Despite the market as a whole moving to tablets and smartphones, there still is a viable market for desktop and notebook sales. Gartner is predicting PC sales to grow .08% as businesses upgrade computers and markets in China and Russia rebound. Microsoft corroborates this view, predicting an increase in the sales of Windows 10-based machines, as well as solving security and privacy issues in China that hampered sales.
With improving yields for TLC, NAND will improve the availability and pricing dynamics for notebook SSDs, allowing for increased capacities and further encroach on HDD notebook sales. The DRAM shortage may be the fly in the ointment due to the shortage that is forecasted to last well into 2018.
CLOSING THOUGHTS
Toshiba is to conclude the sales of its flash business the end of March. If it cannot come to terms with its suitors, the sale may fall apart, leaving Seagate as the odd one out.
One thought that I can’t seem to shake is the economics around the growing storage demand that we will see due to many new data-intensive emerging technologies. In a past market update, I referenced the digital detritus generated by my family due to selfies and videos of fires in our fire pit. As a family, we have in excess of 70,000,000 photos and 7,500 videos. I imagine less than 10% are of sentimental value. With the emergence of IoT, I question how much of the data collected is of value and what the cost is in dollars spent on infrastructure, maintenance and the power to house questionable data gathered by IoT? How much of the data we hold on to is rendered useless due to broken links? What is the cost to businesses and consumers for holding on to data that will never be used?