Storage: Disrupt This!
January 24, 2011
Storage: Disrupt This!
- The history of the data storage market has been marked by wrenching disruptions catalyzed by technological innovation. We are in the midst of such a disruption at the device level, as the rise of portable devices and cloud applications is inducing the rapid ascent of NAND flash storage at the expense of disk drives due to its superior speed, power draw and portability and despite its higher cost. We believe that this transition will be faster and more complete than most observers project. At the same time, we believe that the incorporation of flash and other solid state technologies into the enterprise data center storage market will be incremental, preserving the role of system solution integrators and allowing for continued strong growth of disk technology in that market. However, in the long term, we believe enterprises will begin to embrace cloud computing much more fully, transitioning storage-intensive applications to web-based hosts with the sophistication to design and integrate their own complex storage solutions. Such a change would commoditize storage technologies, a serious threat to current enterprise storage market leaders
- Innovation guru Clay Christensen well documented the accelerated evolution of the disk storage market in his now classic treatise “The Innovator’s Dilemma”. On multiple occasions, new entrants to the storage market were able to apply existing technology in an innovative way for an emerging customer set not well served by the incumbent vendors, establishing new architectures that would grow to overtake the existing leaders. Similar mechanics are driving the current change. The rise of the iPod and smartphones, with compelling need for portability and low latency, opened an opportunity for flash memory to take the place of disk drives despite an overwhelming cost disadvantage. This demand drove flash to a steeper cost reduction curve, has opened the door for a generation of lap-tops without hard disks, a concept unthinkable just 5 years ago
- The expansion of portable devices, epitomized by the explosion of disk-less tablets, has been buttressed by simultaneous upgrades in wireless networks, the near ubiquity of latency minimizing content delivery networks, and the overwhelming popularity of Internet-based consumer cloud applications like Facebook and YouTube. Users do not need big hard disks on board because they can quickly, easily and cheaply access seemingly unlimited storage resources on line. This is bad for the future of PCs, doubly bad for the future demand for device-class disk storage, and good for flash memory. The relative performance of drive makers like Seagate and Western Digital vs. flash vendors like SanDisk and STEC suggests that this dynamic is appreciated by investors
- Meanwhile, enterprise demand for storage systems remains robust, as companies cope with the nearly overwhelming flow of data and the perceived value to be gained from capturing and understanding it. Inside the enterprise, the clear trend toward server virtualization is concentrating storage resources to the data center and creating need for turn-key solutions and expert guidance from integrated storage vendors. While we are concerned that government market weakness will be an albatross about the neck of enterprise spending going forward, the storage market appears robust enough to buck the trend
- We do not believe that solid state memory technologies like flash are likely to reach the relative price points necessary to be an attractive replacement for enterprise mass storage within the investable future. We do expect these technologies to be established as an important adjunct to disk storage for very high value applications where access latency is a critical need. Existing vendors have already introduced hybrid storage products to add this capability, with backward compatibility a key obstacle to new entrants looking to use solid state as way into enterprise accounts. Although no real threat to demand for disks, this phenomenon should be a significant opportunity for flash chip makers
- The longer term is less favorable for value-added storage vendors like EMC, NetApp, HP or IBM. We believe that enterprises will eventually embrace the cloud more fully, shifting the preponderance data intensive applications to web-based commercial hosts. While this does not necessary lessen the industry wide demand for capacity, it changes the nature of the dominant customer. Concentrating demand into the hands of a few highly sophisticated buyers, each with in-house proprietary technology for managing storage will eventually commoditize the market to the benefit of more generic disk drive producers
- In the very long run, there are other options for solid state storage – phase change memory, racetrack memory, memsistors, and solid electrolyte memory at the forefront – but we believe these are too far away and too uncertain to allow meaningful conclusions. As magnetic tape remains in wide use for archiving purposes more than 50 years after its introduction, we suspect that the death of the hard disk has been overstated in many quarters
- We see NAND flash leaders such as SanDisk, STEC, Intel, Micron, Samsung, and others as obvious beneficiaries from the current trends, although we note that flash is a small business for the larger players on the list. We see integrated storage leaders EMC, NetApp, HP, IBM and Oracle as enjoying robust demand for their products, with the same caveat that weaknesses in other businesses may outweigh strength in storage. However, we do note that a longer term shift toward the cloud would be a negative for these companies. On the flip side, near term demand for device hard disks has already softened, yielding weak results from vendors like Seagate and Western Digital. Similarly, an architectural shift toward attaching storage directly to Ethernet networks rather than specialized storage specific equipment works against vendors like Emulex, Qlogic and Brocade
Disruptive Disk Drives
The historical industry disrupting impact of new technology on the data storage market has been exhaustively explored by innovation guru Clayton Christensen (Exhibit 1). Likening the industry to the fruit flies studied by biologists because of the furious pace of their evolution, Christensen described five waves of innovation between 1980 and the mid ‘90’s which laid waste to traditional market leaders in favor of successive rounds of market entrants. This was the basis for his influential framework of disruptive innovation and its accompanying thesis of incumbent vulnerability, all of which is detailed in his book “The Innovator’s Dilemma”. Christensen describes his conception thusly:
“Generally, disruptive innovations were technologically straightforward, consisting of off-the-shelf components put together in a product architecture that was often simpler than prior approaches. They offered less of what customers in established markets wanted so could rarely be initially employed there. They offered a different package of attributes valued only in emerging markets remote from, and unimportant to, the mainstream.”
In the storage market, disruptive innovation has taken hold as the storage needs of emerging product categories were not satisfied by the dominant design. A major example cited by Christensen is the 3.5 inch hard disk drive, which upon conception, did not offer sufficient capacity to satisfy the needs of the then dominant desktop PC paradigm, but suited the nascent laptop market, which was willing to sacrifice capacity and cost in exchange for superior portability. As laptop volume advanced in the early ‘90’s, the capabilities of the smaller drive bridged the gap with the previously favored 5.25 inch standard, leaving it behind and the incumbent producers who had remained faithful to it (Exhibit 2).
Enter the iPod
Fast forward to the late ‘90’s. Flash memory, invented in 1980 by Toshiba scientist Fujio Masuoka, had begun showing as a storage alternative, first in military applications, but was widely viewed as far too costly to be a viable replacement for disk storage. The dominant personal information device at the turn of the millennium was still the laptop, and the 2.5 inch hard disk drive was the ubiquitous choice for storage on the platform. Flash, with the major advantages of much faster access speeds, smaller form factor, lower power requirements and invulnerability to physical shock and motion, began to appear in pocket sized devices like cell phones and the first MP3 players, but the nearly prohibitive relative cost of the chips kept storage capacities on these products tight.
The first iPods used 1.8 inch hard disks and had a significant cost and capacity advantage over flash-based competitors. Add to this the now legendary user interface, and Apple had jump started a new market that quickly became a global consumer phenomenon, leaving flash storage players far behind. However, in January 2005, after years of scoffing at flash-storage, Apple released the iPod Shuffle, kitted with 512 MB of NAND Flash storage for $99. This was the death knell for hard disks at the device level, setting flash memory on a steeper cost reduction trajectory and ushering in an eventual wave of devices where portability and access speeds were valued at a premium (Exhibit 3).
The volumes from the wild success of the iPod have driven flash memory production costs down by a factor of more than 100 to roughly $1.50/GB, still nearly 10 times the cost of hard-disk storage, but manageable as a component cost (Exhibit 4). This has been an important factor in the rise of smartphones and now tablets, which universally eschew the big, slow and motion sensitive hard drive in favor of solid state flash memory. Indeed, the latest round of high-end notebook computers have also dropped hard disks in favor of flash, led again by Apple and its influential Airbook.
Let’s Go to the Cloud
The rise of fast wireless networks, low latency CDNs and Internet-based cloud applications has also eroded the benefits of hard disks in personal devices. Users can access seemingly limitless storage on line with very little performance penalty vs. carrying their personal archives on board, while gaining the benefits of document sharing, access from alternative devices, cloud-based processing resources, etc.. Indeed, almost all of the applications important to consumers today are cloud-based. Social networking sites, like Facebook, Twitter, and Foursquare? All cloud applications. E-commerce purveyors, like Amazon, eBay, Priceline or Fandango? All cloud applications. Streaming media vendors, like Apple iTunes, Pandora, Netflix or Google’s YouTube? All cloud applications. Search? Cloud. On-line gaming? Cloud. Just browsing blogs or news feeds? Cloud. Even document processing – i.e. spreadsheets, presentation graphics, word processing, etc. – is moving to the cloud.
From an overall storage perspective, the cloud enables users to incrementally add storage resources on the cheap, where device storage limitations are difficult and often expensive to surmount. Moreover, many cloud applications are storage intensive – loading images and video onto social networking profiles, maintaining file libraries on media streaming sites, archiving electronic correspondence for web e-mail. This is a key driver of storage demand, just not at the device level.
Given the benefits of Moore’s Law, future flash memory cost reductions are predictable and considerable (Exhibit 5). Adding to this, our general thesis of the rise to dominance of non-PC portable devices like smartphones and tablets, and the consequent denouement of PCs, we believe that small form factor hard disk drive demand will see a steady and irreversible decline. This is a difficult circumstance for disk drive makers who generate over 80% of their revenues from desktop and notebook PCs (Exhibit 6). Furthermore, the growing market for disk drives in non-pc devices, such as DVR set-top-boxes would also be at risk from the cloud as consumer electronics vendors also connect their products to the internet.
Storage demand for enterprises has been on a strong run, as the volume of information captured by organizations has mushroomed and become increasingly important as a decision making tool, as government regulations force larger archives and as a growing flow of information created by computer to computer interaction, sometimes called “dark data” is captured. CIOs have consistently cited coping with this avalanche of data as a primary challenge for their organizations and a priority for their IT budgets.
At the same time, the growth of “virtualization” is concentrating more of the typical enterprise’s computer muscle back at the data center, along with storage (Exhibit 7). This raises demand for larger capacity storage systems, storage networks and for sophisticated storage management software that can parse data to various storage locations depending on cost, access speed requirements, frequency of access, etc.. Moreover, the complexity of these storage solutions is challenging for all but the most sophisticated IT departments, raising the opportunity for higher value-added turnkey products, customized system designs and expert implementation and support. All of this portends reasonably robust demand despite a generally cautious IT environment and the overhang of likely cuts in government IT spending. This favors integrated systems vendors offering a full family of storage devices, storage networking solutions, storage software and technical services (Exhibit 8).
The circumstances also dictate that the basic architectural concept of enterprise data center storage is resistant to change, as the high cost of replacing installed assets and the risks of business disruption are nearly insurmountable obstacles to a forklift upgrade of even a modest data center storage system. Rather, innovations are introduced as components to the greater system architecture, yielding incremental improvement to elements of the system or bolt-on modules to add specific functionality. As such, product life-cycles are long and backward compatibility is critical, characteristics supporting continued dominance by incumbent leaders and reducing the potential for radical new approaches to gain a foothold. We note that magnetic tape remains a widely used technology for enterprise IT archival back-up some 50 years after its introduction and long after it ceased to play a role in any other context, and that EMC, IBM, HP, etc. remain firmly entrenched as market leaders (Exhibit 9). To this end, storage leaders have begun to incorporate solid-state memory as a low-latency storage tier in their newest products. While the relative size of this tier will likely grow to reflect improving costs and an increased value for low latency, it does not create a foothold for new entrants to break the close relationship between IT buyer and storage vendor, nor is it likely to be big enough to reverse the growth trend for disk capacity.
Let’s Go Back to the Cloud
Cloud computing is the buzzword du jour, and, as we have noted, the most important consumer applications are all cloud resident. Enterprise IT moves more cautiously. Enterprise applications that are accessed by a large number of people from wide geographical area and changing physical locations, epitomized by sales force management systems, gain real functional advantage from being resident on the internet. The explosive growth of Salesforce.com is testament to the compelling value proposition for this class of cloud applications. For those applications that are not as obvious, the justification is a trade off of near term expense and disruption for long term efficiency gains. IT managers, who may see those efficiency gains come out of their own budget and their span of control, are understandably reticent.
That said, the efficiency gains from moving enterprise applications to the cloud will likely grow both in size and certainty as the leading hosts gain scale and experience, and the trickle of enterprise apps will accelerate. As this happens, and as the consumer cloud continues to grow, the customer mix for mass storage products will tilt toward the big cloud hosts – Amazon, Google, Microsoft, et al.. For the storage industry, these customers are very different.
Unlike the typical enterprise, storage system performance and efficiency are significant factors for business success in cloud hosting, and the leading players have developed substantial in-house expertise in designing, implementing and operating complex distributed storage systems. Because of this, cloud hosts do not typically buy expensive turnkey systems or engage their vendors for high-margin software or services. Rather, they build their systems themselves, generally from commoditized components. In this way, the cloud harkens back to Christensen’s observation that disruptive innovation often proceeds from an emerging application market that uses “off-the-shelf components put together in a product architecture that was often simpler than prior approaches” (Exhibit 10).
In the longer run, this is not positive for the current leaders of the enterprise storage industry. Increasingly, the incremental data center storage customer will be a cloud host with little interest in the value-added functionality, backward compatibility and engineering services that are critical to the traditional enterprise buyer and the basis for the competitive advantage and strong margins of the market leading vendors. It is likely that this phenomenon will not have major impact on market shares or profitability over the next couple of years, but its longer term impact may become apparent to investors sooner.
The Death of Disk Drives has Been Greatly Exaggerated
The emerging dominance of flash memory storage on devices has led many to presume a similar sea change at the data center level. We believe that this is unlikely for the foreseeable future. First, a big catalyst for the success of disk-less smartphones and tablets is their ability to leverage the enormous storage resources available on the cloud. 20 GB of flash adds a bit over $30 to a bill of materials, 10-15% of the production cost of a tablet. In this context, the solid-state equivalent of an 80 or 160GB hard disk would be prohibitive, but unnecessary owing to fast, low-latency networks linking to the cloud (Exhibit 11). Second, there are durability concerns for flash memory technology – not a problem for devices that have rapid technological obsolescence cycles and usage limited to the stamina of a single user, but a major concern in a data center serving users round the clock and with the expectation of years of reliability.
Finally, the more than 10 to 1 cost differential between flash and disk storage will narrow, but it is not certain that NAND flash will scale below 22nm production technology, which would compromise its ability to completely close the gap. With the storage needs for a cloud provider beyond gigabytes and terabytes , into the petabyte territory with exabytes on the horizon, even a 2 to 1 cost differential would be prohibitive for a wholesale changeout. Enterprise data centers, with less of an imperative for premium performance and power management would be even less inclined.
Flash memory has emerged as a component to an overall data center storage architecture that will remain centered on magnetic media at its core (Exhibit 12). By adding a tier of flash, a storage system can make use of its dramatic advantage in access latency for specific applications for which that attribute is a critical advantage, while maintaining traditional disk arrays for the large majority of storage requirements. This will result in very strong growth from a very small base for solid-state storage components, but we do not believe that it will represent a threat to demand for disk based storage for several years to come.
The Bleeding Edge
While we are skeptical that scalability and endurance issues for NAND flash memory can be adequately resolved, there are other candidate technologies that could eventually challenge the hegemony of the disk. Of these, the most promising appears to be Phase Change Memory (PCM), which uses temperature change to melt and reset microscopic crystals (Exhibit 13). This technology is much further along in development than other promising alternatives, such as solid electrolyte memory, racetrack memory or memristors, and offers promising prospects for low cost, excellent scalability and high endurance.
However, cost projections for PCM memory are obviously dependent on volume and on production yields, neither of which can be forecasted with any confidence. We believe that large scale commercial production of this or any of the other memory alternatives is unlikely in the next 5 years, and that cost parity with flash, much less hard disk drives will take far longer to achieve.
Winners and Losers
We see the storage market proceeding in stages that will favor and disadvantage different sets of players along the way. In the near term, we believe that demand for device disk drives will roll over with the demand for traditional lap-top PCs, while the ascendant portable platforms drive demand for flash storage (Exhibit 14). The major disk drive makers – Seagate, Western Digital and Hitachi – have already shown showing weakness owing to the transition to solid state storage at the device level, as these companies derive the large majority of their revenues from the PC platform (Exhibit 15). Given our perspective for global PC sales below even lowered expectations in 2011, we are not hopeful for a near term boost for these stocks, although we note that Seagate has relatively less exposure to the mobile and desktop markets. At the same time, flash drive makers such as SanDisk and STEC have a wind at their backs that should sustain their business momentum.
Enterprise storage pure plays like EMC, NetApp and Compellent have enjoyed strong growth from the expanding storage needs of companies, growth that can be expected to continue into the near to intermediate term. Integrated IT vendors HP, IBM, Oracle and Hitachi also compete in this market, although storage solutions are a relatively small part of their revenue base. We note that a significant trend toward connecting storage assets directly onto Ethernet based networks plays to the detriment of companies focused on storage-specific network technologies, such as Brocade, Emulex and QLogic.
However, we expect the demand for data center software to radically shift in the longer term, as enterprises fully embrace the cloud hosting model and highly sophisticated cloud operators like Google, Amazon, and Microsoft become the primary market for storage products (Exhibit 16). Such a change would seriously erode the value added by storage system vendors, commoditizing the market despite sustaining or even accelerating the continued growth in demand for capacity. Because of this, we are concerned that most enterprise storage companies will face significant new price pressures in the latter half of the decade. The exception would be relatively undifferentiated disk drive manufacturers – such as Seagate and Western Digital – who already cope with sharp price competition and who will have already adjusted to the deterioration of the device market.
At some point, new solid state technologies such as Phase Change Memory will become commercially viable. However, we believe that it is too early to choose winners or to project timing.