Recently we had a large data center ask us when we thought HDD and SSD will reach pricing parity. This is not an easy question to answer and there simply isn’t a straightforward response. New technologies that offer blazing speed are changing the storage ecosphere and upending how we measure the cost of storage.
Let’s start by looking at the question from a purely economic standpoint. In 2018 there were roughly 900 exabytes of combined SSD/HDD storage shipped, comprising 100 exabytes of SSD. Factoring in a $.10 per gigabyte SSD cost, there’s still a substantial price header in comparison to HDD, with HDD coming in at $.023 per gigabyte working off of a $23.00 per terabyte cost.
With data expected to grow annually at 42% and NAND capacity bit growth at 40%, data growth is still outpacing NAND, albeit marginally. Compounding this, the 50% YOY drop in NAND pricing has left little incentive for NAND manufacturers to add capacity. We don’t see demand for enterprise SSD picking up until Q4 as hyperscale and cloud service providers continue to work through SSD inventory accumulated in 2018.
Enterprise computing and the challenge of NAND durability
Further complicating the cost equation is NAND durability. To lower the cost of NAND, manufacturers look to increase cell density. However, by doing so, NAND becomes inherently less stable.
For example, a single level cell (SLC) supports 100,000 write cycles. By contrast, MLC (2bit) planar NAND supports 10,000 write cycles, which goes up to 35,000 utilizing 3D NAND. Even with the use of 3D NAND, a QLC (4 bit per cell) disk only supports about 1,000 writes. To offset this instability, read and writes are spread across bits to reduce wear, which requires users to deploy larger drives. This introduces additional complexity to the cost equation.
Meanwhile, users are turning to increasingly sophisticated approaches to the tiering and caching of workloads, seeking the highest level of performance at the lowest price point. Take the introduction of storage class memory (SCM), which gives storage engineers a third option when speccing out storage systems. SCM is more likely to be used in place of DRAM than storage, offering a denser and more cost-effective solution that is infringing on SLC and MLC SSD demand requirements.
Overall, there’s little doubt storage is becoming more intelligent, not only integrating new technologies such as SCM but utilizing AI to optimize data and cost outcomes. New approaches and techniques in the reading of data are being developed to overcome SSD bottlenecks, such as the reduction in chip parallelism as storage density increases. At the same time, SSD manufacturers are busy tailoring SSD to support various use cases, such as write-intensive, read-intensive or mixed use applications.
Perhaps more than anything, NVMe and NVMe-0F will drive the uptake of SSD in enterprise applications. Similar to SCSI HDD, SATA SSD is falling by the wayside as the NVMe ecosphere grows. Keep an eye on companies trying to revolutionize the storage market, such as Vast Data and its use of data reduction technology to manage all storage functions in a single TLC-based drive. Could this spell the end of the HDD market as we know it?
The race to increase areal density
As surveillance and IoT lead the charge in data generation, HDD will continue to evolve to provide low-cost bulk storage. However, Kryder’s law (areal density doubles every thirteen months) is becoming increasingly difficult to achieve.
There are one of two ways to increase HDD capacity, either by adding more platters or storing more bits on a platter. Adding platters is limited by the 3.5” form factor, so increasing disk bit density is the logical method. We have seen SMR HDDs utilizing a host to manage storage and push areal density, but this comes with a trade-off as data writes are significantly slowed. While packing more bits into a fixed space makes sense, it leads to instability and bit errors.
This past fall, Seagate and WDC offered updates on their respective solutions to overcome bit instability when increasing bit concentration, with Western Digital touting its MAMR technology and Seagate pushing HAMR. In both scenarios, using energy assist when writing data allows for platter material that better enables increased bit concentration.
However, the product roadmap for each technology so far hasn’t quite followed plan. Seagate was to have a 16TB HAMR drive ready for the second half of 2019, but instead reported in a recent earnings call that it had started shipping its 16TB TDMR drive (achieved by adding a platter), to be followed by an 18TB SMR drive. It is pushing HAMR-based drives into 2020, reportedly starting with 20TB capacities.
We are hearing conflicting accounts around WD’s MAMR technology, with WD initially positioning MAMR as a path to 100TB drives. However, the word on the street is that WD’s MAMR program is challenged and its program is falling behind, with reports that MAMR is only good to support two transitions (20 TB and 24TB) before having to add platters and/or SMR to squeeze out more capacity.
At the same time, WDC is touting its SMR-based drives as a complement to its energy-assisted solutions. The company predicts that by 2023 half of data centers will be using SMR-based drives, to meet customers’ demand for low-cost storage.
For its part, Seagate’s HAMR will apparently scale up to 200TB, with the company reporting it is sampling 16TB HAMR drives to key customers and has been actively shipping 16TB TMR since March, with volume shipments scheduled for Q3.
Meanwhile, Toshiba is reportedly utilizing both MAMR and HAMR technologies for future builds and will be sampling a 18TB MAMR-based drive by the end of the year, as well as having already sampled 16TB TMR drives. It appears Toshiba is keeping its options open until a clear frontrunner emerges.
The word on the street is that Toshiba has been very aggressive with its pricing, winning business from HGST in the 14TB nearline market. And although HGST reports health in the midcap nearline market, there are rumors that the company is reducing internal component orders for nearline drives. We are also hearing rumors that Seagate has struggled with its 14TB drives.
As manufacturers push the evolution of #HDD technology by increasing platters, shrinking componentry and adding new features such as multiple actuator arms, cost and complexity continue to increase and the drives become less stable.
Hyperscale spending and NAND oversupply
Setting the tone for the market are the hyperscalers, which typically buy in six-to-eight-month cycles. With the last cycle probably ending Q1 2019, there are no immediate indications that hyperscaler activity will be accelerating in the near term, though new construction permits are trending upward. Expect to see activity picking up near the end of the year when 16TB shifts into volume production.
It’s certainly no secret that the NAND market is in oversupply. Some think we have finally hit bottom, with Samsung considering raising NAND prices by 10%. In typical fashion, we see the love-hate relationship between supplier and customer rear its head amid reports that some volume consumers are pushing to lock in current prices through long-term agreements.
Will we see the roll out of 16TB drives jump a storage refresh cycle or will it be new compute platforms driven by next gen technology such as AMD’s 7nm Rome processor?
The PC market and the cloud
In comparison to the enterprise market, the direction of the desktop/notebook market is clear. There is a more linear relationship between cost and performance when comparing SSD to HDD in the PC market as opposed to the enterprise.
With end users managing most of their storage in the cloud, they are much more willing to offset the higher cost of SSD with performance gains and shrinking drive capacities. Increasingly the notebook/ desktop is becoming an appliance, with Chromebooks showing double digit growth YOY. This all seems eminently logical as consumers look to the cloud for storage needs.
After both Gartner (-6.3%) and IDC (-3%) reported PC sales were down for Q1, it looks like Q2 numbers are shaping up better, with preliminary results showing a 4% YOY growth. The bounce back can be attributed to improved CPU deliveries by Intel, whose struggles negatively impacted the supply chain, as it backfills orders. At the same time, PC manufacturers are pulling in inventory in anticipation of the next round of tariffs.
None of this, however, alters the narrative around declining PC sales. Nidec, the hard drive spindle manufacturer, projects hard drive sales will drop from 124 million units in 2018 to 65 million in 2019, with the most significant decline in the notebook market. Although back-to-school sales and the ramp to support holiday builds may support activity, any gains could be offset as the Windows 10 replacement cycle that has been in play for the last few years runs out of steam. In any event, the primary beneficiaries will be the SSD manufacturers.
The roll out of 5G will further push the deceleration of PC sales as faster data transfer rates will only make the cloud that much more compelling. We are already seeing evidence of this in the PC gaming market with the release of gaming streaming services such as Google’s Stadia and Microsoft’s Project xCloud.
HDD and the future of surveillance
Conversely, surveillance drives are a bright spot for the HDD market. It is estimated that the surveillance market will grow from USD 36.89 Billion in 2018 to USD 68.34 Billion by 2023, at a CAGR of 13.1%, driven by security concerns. Hard drive’s low-cost bulk storage profile is ideally suited to support the needs of the surveillance market.
Not far behind is the SMB NAS market, which is also projected to grow, propelled by innovations around the internet of things (IoT) by home appliance manufacturers driving up the adoption of smart home and home automation applications.
These are areas of continued strength for the HDD manufacturers that will be used in conjunction with SSD to optimize cost against performance.
The supply chain impact of the trade dispute
There seems to be no end in sight to the trade dispute between the United States and China. The Trump administration recently doubled down on its hardline position by initiating an outright ban on the Chinese telecom manufacturer Huawei from participating in US government contracts as well as restricting Huawei’s access to critical components. China quickly responded by threatening to restrict access to rare earth materials used in the manufacturing of components.
After all the saber rattling, the two leaders have agreed to postpone new tariffs and restrictions for Huawei’s access to needed components have been eased. However, the dispute, which has been simmering for quite some time, is far from over and may simply represent the new normal between the two superpowers.
China has certainly not been shy about making its intentions clear, with both its Belt and Road and Made in China 2025 initiatives. While it has the right to build a technology-focused economy, the dispute over the theft of IP is very public and real. With the Chinese government subsidizing the growth of its industrial sector and offering sweetheart financing deals that skirt WTO agreements, this is a fight for the keys to the castle.
The global economy has not been without response. We have seen the invisible hand of capitalism quickly reshuffle the deck, with manufacturing moving away from China and companies increasingly questioning the Faustian bargain often implicit when doing business in China, trading IP for access to its vast market. Affected by a lack of coherent strategy, the supply chain has descended into what is, to all intents and purposes, cross-border horse trading.
Final thoughts
The World Wide Web was created 30 years ago with the best of intentions, as a tool to share information and collaborate on the development of ideas. This has led to many new advances in science, art, medicine and given us access to vast amounts of information that have made our lives easier.
With it came many exciting developments and hope, but also a dark side. Despite all the good technology offers us, it remains a tool: a tool that can be used for good purposes or nefarious ones. Increasingly, we are seeing technology used as a means to justify the end, as a divider not a uniter. With that we see the world dividing itself into camps, with the supply chains needing to choose sides.
Stephen Buckler is chief operating officer at Horizon Technology. Read other storage market briefs by Stephen, or connect with him on LinkedIn.
