WDC and Seagate both indicated in their earning calls that they are the beneficiaries of strong near-line demand being driven by hyperscalers building capacity to manage demand from Artificial Intelligence, Machine Learning, and IoT. Steve Mulligan, the CEO of WDC, indicated that connected devices are expected to grow sharply from current volumes of 9 billion in the coming years, which is a catalyst for the need for storage.
Interestingly enough, SSD seems to have been shoved to the background with the focus on near-line demand, making a strong argument that boils down to cost for storage. As discussed in March’s update, affordability at the user level plays a significant role in shaping future storage demand. Despite the performance advantage of SSD, the high cost of SSD in comparison to HDD prohibitively limits consumer adoption rates of new technologies. As AI functions become more and more a day-to-day reality it turns from a novelty to an expectation, cementing the need for low-cost storage.
The earnings calls and question-and-answer periods were revealing of current and future market conditions as well as disclosing the divergent views on how to manage future storage needs. Currently, both WDC and Seagate are feasting on a data center capacity upgrade cycle, driving near-line sales to dizzying heights. Both recognize the need for low-cost storage but that is where the similarity ends, with each choosing different paths on how to provide the next generation of low-cost storage. With WDC, it is choosing MAMR (microwave-assisted magnetic recording) and reporting that samples will be shipping end of 2018 and volume shipments in 2019. At the same time, Seagate is betting on HAMR (heat-assisted magnetic recording) technology and indicating that they will be shipping in volume in 2019.
The difference between the two widens further around SSD. Seagate seems to almost take a dismissive view, indicating that they only expect up to 8% cannibalization of HDD sales by SSD. And, Seagate extols the financial freedom of not having to tie up capital in fab capacity, instead relying on its Toshiba relationship, while HGST subscribes to the benefits of being vertically integrated.
HDD Storage
As indicated in the introduction, both Seagate and WDC are benefitting from hyperscalers bulking up on storage, with Seagate’s revenue up 5% year over year, HDD Y/Y exabyte shipments up 34%, and average selling price per unit up 6% year over year. WDC reported Y/Y revenues up 8% and a record 100TB in exabytes shipped, with ASPs up a whopping 14%, driven by the high concentration of large-cap near-line builds supporting hyperscaler demand.
Looking forward, both WDC and Seagate have bullish outlooks on the back of continued cloud investment, with Seagate opining that we are in the early stages of a broad-based global cloud storage transformation and that exabyte demand will continue to grow year over year.
WD adjusted its exabyte growth forecast for 2018 from 50% to 65%. In turn, Seagate is seeing increasing revenue contribution from its 12TB drive, while WDC is releasing its 5th generation helium 14TB drive.
Under all this rosiness, we need to ask ourselves how long this hyperscaler expansion cycle will last, as when supply catches up with demand, prices will quickly adjust downward. If demand wanes, that downward adjustment will accelerate as there certainly is some double ordering baked into the numbers.
As indicated earlier, both manufacturers are optimistic that demand will continue, with Seagate stating that demand is more linear than past demand cycles and highlighting that they also expect on-premise demand to be strong into the back half of 2018. In the question and answer period of their earning calls, both discussed component shortages impacting their ability to meet demand, indicating that they are pulling components from desktop/notebook lines and talked of optimizing by reshuffling builds and the movement of buffers stocks to support near line demand.
Despite characterizing shortfalls as a result of component shortages, in speaking to those within the manufactures, it isn’t so much a component shortage impacting near-line supply but supply mix issues. Drives configured to hyperscaler’s specifications use specific head and media assemblies, base castings and actuator arms. When taking orders, sales reps look at material availability but not all the way down to component part number level, which leads to supply gaps if there is a mismatch due to the 16-week component delivery from time of order. To bridge that gap, components are pulled from ditsy supply, desktop/NB (flat Q1 sales) and surveillance (did not meet expectations) to the more lucrative near-line drives. Which explains the gaudy near-line numbers.
News from the HGST distribution side of things is that, after having very limited sight into an available enterprise, product visibility is much improved and they expect supply to catch demand by the end of June. I asked if that means that supply will catch Hyperscale demand and was told not necessarily because of the mentioned mix issue. It was postulated that the increased distribution volumes are QA-fallout from the increased Hyperscale volume kicking out drives that did not pass Hyperscaler ongoing reliability tests (ORT) but satisfy distribution QA requirements. Predicting the end of the shortage is never easy but logic tells me that, if the distribution is seeing product freeing up, then the product must be freeing upstream.
As we are looking to the immediate future for drives to satisfy data center demand, the 14TB is front and center, with all three HDD manufacturers having a PMR drive in the mix that they are sampling and/or shipping. This gives Seagate an opportunity to put its base plate issues behind them, which has hampered its 10TB and 12TB capacities’ cost structure. If Toshiba can successfully execute the launch of its 14TB on the back end of 2018, it has the potential to be disruptive around pricing, by giving data centers a third option. Beyond the 14TB, it is a footrace between Seagate’s HAMR and WDC’s MAMR approach, which we discussed in detail in our October market brief.
Solid State Storage
As indicated in the introductory paragraph, Seagate and WDC results seemed to suck the oxygen out of the room, leaving SSD as an afterthought. This shows that HDD continues to be the storage medium of choice for large data centers, which leaves us to question if 96-layer NAND will bring down bit costs to be a cost-effective alternative to HDD in Hyperscaler storage. As per last month’s market update in comparing the costs of HDD against SSD it is not a straight cost-per-bit calculation. Several variables, such as cost per bit, performance, power consumption, and footprint need to be considered when evaluating TCO between the two technologies.
As stated earlier, Seagate seemed almost dismissive of SSD. They recognize the near-term threat of SSD by announcing their moving away from sub 1TB HDD capacities for client drives and the 15k SAS market, indicating that they constitute less than 8% of total revenue. (Note, WDC officially announced last-time buys with shipments to cease at year end.) Are they staking their claim on the cost advantage of HAMR combined with multi-actuator technology as the low-cost technology of choice for cloud storage? Or will they prove to be very prescient, relying on partnerships with fab partners, such as Toshiba and the emergence of YMTC as a NAND contender, driving NAND prices lower rather than pouring funds into fab capacity?
Seagate’s position summarizes the dynamic between SSD and HDD. As stated many times before, SSD will undoubtedly play a role as it is combined with HDD to optimize workloads against cost. It is a question of percentages as recent results demonstrate just how an important role cost plays. Remember Friedman’s book “The Word is Flat,” detailing the confluence of events that opened low-cost computing to the masses.
Pay attention to hyperscalers bringing in more and more of component design in-house with reports that they are looking to build their own SSD using lower-grade NAND. Clearly they are looking to lower the price points of NAND to bring in-house and benefit from the performance benefits of NAND.
Looking at the market as a whole, as indicated on SK Hynix’s earnings call, the SSD market was soft, with client pieces sagging 10% Q/Q, primarily driven by weak mobile demand and seasonality. Although SSD manufactures are predicting pricing to rebound the second half of the year with possible stretching lead times, as SSD content grows and seasonal demand kicks in. The enterprise market has remained tight, with MLC-based SSD being in highest demand.
Final Notes
The biggest take away for me is that, despite the sheer enormity of change as data and how we use data transforms commerce and banking, the old-school economic laws such as Alfred Marshal’s law of supply and demand and David Ricardo’s law of diminishing returns continue to be the underpinnings of our economy.
Lastly, with this bounty of convenience, it is easy to get lost in the self. Think of your fellow man and creatures. Use responsibly.
