Data Center Spending: We Don’t Get Fooled Again!
SEE LAST PAGE OF THIS REPORT Paul Sagawa / Artur Pylak
FOR IMPORTANT DISCLOSURES 203.901.1633 /.1634
sagawa@ / firstname.lastname@example.org
May 21, 2013
Data Center Spending: We Don’t Get Fooled Again!
- The data center virtualization cycle that drove enterprise spending over the past decade has peaked. IT managers are now looking to a future where many of their applications will be handled by web-scale distributed cloud data centers with dramatic cost and performance advantages over privately run operations. Still, the obstacles remain daunting – e.g. security concerns, conversion costs, management complexity – and enterprises are proceeding cautiously. The result is a fallow period in IT spending – investment in client-server era data center hardware and software is slowing, but cloud-related spending has yet to hit the knee of its growth curve. The circumstances are reminiscent of the late ‘80’s/early ‘90’s when sales of the mainframes and minis that had dominated the previous era began to give way to client-server architecture. Then, as now, overall IT spending was sluggish and analysts projected a long slow adoption curve for the new approach. Now, as then, projections of the shift to new architectural paradigm are overly conservative. As such, we remain long-term bearish on enterprise data center systems and software – e.g. configured servers, RAID storage, networking gear, infrastructure software and traditional applications – even though some of these areas are still delivering growth. We believe patient investors will be well rewarded by a focus on large distributed data center operators, Software as a service applications, IT consulting, and commodity components, such as disk drives.
- Enterprise data center IT spending has turned south. 1Q13 sales by companies in the enterprise IT segment were generally poor, continuing a trend of disappointments stretching back more than 2 years. The contagion that had been apparent in PC sales has spread to the data center, first in declining sales of server systems, then to sharp decelerations in infrastructure software, storage systems, enterprise networking, and IT services. While some of these categories are still seeing YoY growth, the downward trajectory is now well established. As a result, 1Q13 saw widespread disappointments amongst data-center oriented companies.
- Cloud spending continues to grow. Enterprise spending on cloud services is growing rapidly from a relatively small base. Hosting entities like Amazon Web Services, Microsoft Azure and Rackspace suggest 20%+ annual growth, while SaaS companies, like Salesforce.com, Workday and Netsuite, are delivering better than 30% growth. CAPEX by web-scale data center operators Google, Amazon, Microsoft and Facebook, was up 42% YoY in 1Q13, focused on commodity hardware, such as server chips and disk drives, for their self-designed architectures.
- The trend is consistent with shifting IT manager priorities. Over the last three years, Gartner’s CIO survey has seen a dramatic change in priorities, with longstanding key issue virtualization dropping out of the top 5, echoing earlier drops for ERP applications and security. Meanwhile, cloud computing has risen to the top, along with mobile computing and business intelligence/big data – all issues tied to the emerging new paradigm.
- Paradigm shift puts overall IT spending in temporary stasis. CIOs are betwixt and between, certain that the future will see many, if not most, enterprise applications served from the cloud, but uncertain how best to get there. In this environment, spending on “old paradigm” data center technologies is cut aggressively, but investment in moving to the cloud proceeds cautiously. This is reminiscent of the transition from the mainframe/mini era to PC/client-server architecture, which saw weak IT spending in ’89-’92 eventually give way to the bounty of the ‘90’s. Then, as now, the deadwood needed to be cleared before the growing companies driving the new architecture could take the tech sector along with their growth.
- Spending on “old paradigm” data center technology will NOT recover. While there may be some ebb and flow in spending on server systems, infrastructure software and traditional applications, these technologies appear to have begun their decline phase. Spending on storage systems is growing to cope with the extraordinary generation of data, but the profound cost advantages of cloud-based storage will win out and throw this segment into decline sooner rather than later. Networking would appear more robust, based on Cisco’s F3Q13 results, but progress on open source SDN alternatives to proprietary systems solutions and the eventual shift of workloads to the cloud could render the current surge in demand temporary. We note that the overall data center systems spending decline will be gradual – the advantages of the cloud are minimal for applications that are purely local and finite in scale, such as shop floor control systems, and inertia is a powerful force. IBM still sells more than $4B per year in mainframe technology nearly 50 years after the introduction of the first System360 systems. Still, core businesses for tech stalwarts such as HP, Dell, EMC, Cisco, Oracle, SAP, IBM, and others will face irreversible long-term decline.
- Spending on “new paradigm” cloud infrastructure and applications will accelerate. We believe the market for Infrastructure-as-a-Service hosting and Software-as-a-Service applications will accelerate well ahead of expectations. We believe hosting will be dominated by scale players, with Amazon, Microsoft and Google best positioned in the US. SaaS applications will be a more open market, with the best players focusing on best-in-class apps and leveraging the scale infrastructure available from IaaS hosts. The processing and storage demands on the web-scale operators will continue apace, requiring massive CAPEX focused on commodity components, such as server chips, solid state storage and disk drives. Enterprises, maintaining hybrid IT operations and balancing multiple cloud vendors, will lean on IT consultants to build the new competencies needed to navigate the new paradigm. Companies expected to prosper in this era include MSFT, AMZN, GOOG, N, WDAY, CRM, ACN, RAX, RHAT, CTRX, STX, WDC, AMCC, and FIO.
Data Centers – Déjà vu All Over Again
Cisco’s April quarter surprise broke a streak of 1Q13 disappointments by data center IT stalwarts like IBM, Oracle, HP, Dell, EMC, and VMWare. While investors may be tempted to read Cisco’s relative strength as an across the board recovery in IT spending, we are not enthusiastic. Spending on traditional data center IT has been decelerating for more than two years. Server system sales turned down in 2011, while infrastructure and applications software, storage system, and networking sales began a sharp deceleration from double digit growth in the same time frame.
This is consistent with changing CIO priorities. A few years ago, data center spending was an industry bright spot, as enterprises executed a multi-year implementation of virtualization tech to improve utilization and streamline the development and distribution of applications. Gartner CIO surveys saw virtualization break the top 5 priorities in 2005 and rise to number 1 by 2010. Since then, virtualization has dropped like a stone to 8th place and other legacy issues like security, ERP, and storage management have fallen off of the radar screen, as new paradigm concerns Analytics (big data), Mobile Technologies, and Cloud Computing have moved resolutely to the top 3 slots.
The move to the public cloud seems inevitable, and CIOs are reluctant to overinvest in a legacy architecture that will serve less of their application demands in the future, even as they are cautious about moving too quickly to the new paradigm. The resulting malaise is strongly reminiscent of a similar period in the late ‘80’s – early ‘90’s when spending on mainframe and mini computers flagged, while the shift to the PC based client-server architecture that would define the subsequent era had yet to gain momentum. However, once the dam broke, companies like Microsoft, Dell, Oracle, SAP, Cisco, Sun, and HP were there to reap the rewards, while the old guard suffered.
We believe the shift to the cloud has begun, and that the effects on companies on both sides of the technological paradigm shift will be more profound than analysts predict. Sales of legacy systems, software and related services will settle into irreversible decline, including categories like storage and networking that have seemed relatively immune to date. Companies depending on these products will suffer, and many will fail, consistent with the sad experience of once great mini-computer companies, like DEC, Wang, Unisys, and Data general, after the dawn of client-server. Investors hoping for a rebound from the likes of HP, IBM, Oracle, Dell, SAP, EMC, VMWare, or NetApp should be concerned, and those celebrating Cisco should take a short term perspective on the stock.
Meanwhile, expect the uptake of cloud services from Amazon, Microsoft, and Google to pick up steam, as the three have begun to leverage their massive scale into extremely aggressive pricing. Expect capital spending by these web scale operators, along with possible future rivals Facebook, Apple and others, to drive surging demand for commodity parts, such as server chips, disk drives, and solid-state storage – think Intel, AMCC, Seagate, Western Digital, Fusion I/O and others – even as they eschew “value added” systems and software in favor of their own proprietary solutions. SaaS applications, such as those offered by Workday, NetSuite, CRM and a rash of coming IPOs, will flourish, and IT managers will increasingly turn to consultants, such as Accenture and Rackspace, to help them make sense of the transition.
Once upon a time, the IT world was ruled by Mainframes and Mini-computers, nested comfortably in their raised floor, glass walled, climate controlled rooms. IT departments divvied up computer time and data storage like Dickensian workhouse masters meting out gruel. The first PCs arrived in the ‘80’s but didn’t change much. In those days, personal computers were strictly personal, with file sharing via “sneaker net” floppy disk exchanges. Analysts of the day, and undoubtedly, the CEOs of the firms supplying those raised floor data centers, saw the PC as complimentary to the Mainframe/Mini hegemony, offloading a few truly workgroup-related applications, but unable to address the broader application needs of an enterprise.
But the PC, poster child for the phenomenon that Dr. Clay Christensen would later detail in his well-read treatises on disruptive innovation, had already begun to metastasize within the enterprise. PCs grew more powerful and gained large hard disks. Twisted pair Ethernet LAN cards made it easy to connect PCs together and to share files and peripherals. By the late ‘80’s, high-end PCs began to deliver enough processing oomph to handle multiple simultaneous tasks and multiple users, supported by multi-user implementations of Microsoft’s DOS operating system, striking at the high-volume low-end of the mini-computer market. By the early ‘90’s, networked multi-processor PC architecture systems running not just Microsoft Windows, but also the data center favored UNIX operating system, had proprietary mini-computer makers under siege, and were a credible threat to entrenched mainframe applications as well (Exhibit 1).
Exh 1: Major eras in computing
By 1995, clusted x86 servers were crowding out all but the most entrenched mainframe applications, with the upstart companies that had driven the new PC client/server architecture forward – Intel, Microsoft, Dell, Cisco, EMC, and Oracle chief amongst them – taking control of the enterprise data center and delivering extraordinary returns, while the erstwhile mini-computer standard bearers languished. One time hot stocks DEC, DataGeneral, Prime, Sperry, Burroughs, Control Data, Wang, Amdahl, Cray, Stratus, Tandem, Bull, and others, were punished, most eventually taken out for pennies on the dollar or forced into bankruptcy (Exhibit 2). IBM and HP were the lone survivors, leaders of the previous era that managed to re-invent themselves and thrive in the client-server era.
Exh 2: Quarterly YoY Growth Rates of Major Mainframe/Mini Computer Makers versus PC Companies, 1985-1995
What Goes Around Comes Around
Fast forward to 2010. Client-server data centers were chock-a-block with Intel x86-based servers from HP, Dell, and IBM, RAID storage arrays from EMC, NetApp, HP and Oracle, and networking gear from Cisco. Highly customized 15-year old business applications were running atop 20 year old structured data base infrastructure software, both from the likes of Oracle, SAP and IBM. Enterprise CIOs were in the weeds on implementing data center virtualization, technology championed by EMC spin-out VMWare that allowed them to cleave off pieces of their collected data center resources and assign them to specific users as “virtual machines”, appearing to those users as dedicated computers but enabling superior capacity utilization, tighter control and easier support for the IT department (Exhibit 3).
Mark that as the inflection point for the client-server data center. CIO priorities were changing. Based on Gartner’s annual survey, Virtualization was the top priority in 2010, but had fallen to 8th place in 2013, following other formerly important topics, like ERP (and other enterprise applications), and security to the lower end of the list. Meanwhile cloud-computing, mobile technologies, and analytics/business intelligence shot to the top (Exhibit 4).
Exh 3: The Traditional Desktop versus the Virtualized Desktop
Exh 4: Shifting Global CIO Technology Priorities Over Time
This was the harbinger of the next “Next Big Thing” in data center computing. Cloud computing offers CIOs the promise of dramatically lower costs, extraordinary flexibility, superior performance, freedom from scale limits, global reach, and escape from many of the burdens of IT management. Mobile technologies have already overrun the consumer market, leaving enterprise IT workers clamoring to be untethered and to have the sort of powerful and intuitive interfaces that they already enjoy on their personal devices. Analytics/business intelligence applications have taken a dramatic turn to the cloud, as the “big data” tools coming into vogue demand the sort of scalability and power endemic to the modern cloud distributed data processing center architecture.
Okay, What IS Cloud Computing?
Cloud computing is an extremely scalable, low cost and high performance approach to data processing pioneered by Google in the late ‘90’s. Because the Internet already contained billions of web pages to index and search, with a growth trajectory that would obviously take that universe into the many trillions, client-server architecture and its reliance on limited scale structured data bases and expensive hardware systems was not going to cut it. The Google solution was a network of distributed data centers around the world, each with rack after rack of simple, cheap and replaceable processing and disk storage boards, all running a ground breaking proprietary software infrastructure that allowed the analysis of unstructured data sets of limitless size at extraordinary speed with unprecedented accuracy. Google contributed rudimentary versions of this architecture to the open source community, and the concepts were quickly adopted and implemented into new public standards like Hadoop and Hbase, and into the cloud platforms built by web-scale operators like Amazon, Facebook and Microsoft (Exhibit 5).
Exh 5: MapReduce/Hadoop parallel computing flow
These distributed data center operations run at dramatic cost savings vs. traditional enterprise data centers – Forrester recently estimated that cloud-based storage services cost 75% less than adding storage capacity to an internal data center, on a life time, all in, apples-to-apples basis (Exhibit 6). As the largest cloud data center operators – Google, Amazon, Facebook, and Microsoft – continue to gain scale by leveraging their own substantial internal needs, their intrinsic advantages, even against other, smaller operators deploying similar architectures, are increasing (Exhibit 7). Moreover, competition amongst Amazon, Microsoft and Google has grown intense, with each moving to one up the others with service expansions and price reductions (Exhibit 8-9). This will be to the benefit of enterprise customers, and should stimulate acceleration in the adoption of the services.
Exh 6: Costs of 100TB of Storage, Internal versus Cloud
Exh 7: Known US Datacenter Locations of Apple, Amazon, Facebook, Google, and Microsoft
Exh 8: Cloud Service Provide Price Cuts Announced Last 18 Months
Exh 9: Sample Basic Cloud Services Pricing from Major Providers – January 2012 versus May 2013
Exh 10: Server / Component Cost Comparison, Traditional IT versus Cloud Vendors
Here Comes the Pain
The shift to the cloud will be a very bad thing for client-server enterprise data center vendors. Cloud operators do not buy data center systems filled with high-margin bells and whistles, they buy commodity hardware components and have them installed into boards of their own design by low margin contract manufacturers (Exhibit 10). Cloud vendors do not buy infrastructure software from traditional vendors, they write and implement their own software solutions. Cloud vendors maintain their data centers themselves, foregoing the pricey service contracts that have been so lucrative for traditional IT vendors. Software-as-a-Service (SaaS) applications are purpose built to exploit the capabilities of web-scale distributed data centers, and SaaS developers are displacing applications built on the traditional client-server model and bringing that data out of the enterprise data center and into the cloud.
Of course there are speed bumps. Moving applications and data takes time and costs money, disrupts operations and introduces different security concerns. The choices of “what to do” and “how to do it” are complicated, and based on new technologies that are not necessarily well understood by the managers that will have to make the choices. This mix of high stakes and confusion has kept the pace of cloud adoption modest, as CIOs grapple with their anxiety.
Exh 11: Quarterly YoY Growth Rates of Select Data Center Related Business Units of Major Enterprise IT Companies, 1Q2010-1Q2013
Still, no matter how fraught the journey, the eventual destination is increasingly clear to CIOs; and if they are going to be moving to the cloud, they are not seeing a lot of reason to keep investing in the old architecture. Server system sales began to decline more than two years ago – pummeling HP and Dell (Exhibit 11). Storage systems, infrastructure software, traditional applications, and services spending, began to decelerate in the same timeframe, sharply slowing from strong double digit growth rates in 2010, with new software license signings and services spending falling into decline, and storage holding on to low single digit growth. This is a brutal environment for the companies tied to the enterprise, most of which have badly underperformed over the last two years.
1Q13 saw a parade of disappointments from the group, beginning with Oracle, which blamed its weak performance on its own sales force, only to see the same malaise trip up competitor and partner alike across the landscape. IBM missed the quarter, blaming currency shifts, European economic woes and a hardware slowdown, but overall sales growth has been negligible or down for over a year, with software down in 3 of the last quarters, and global services off for 5 in a row. HP “beat” in its January quarter, delivering a sales decline of “only” 6% against expectations of worse. EMC’s 5.8% growth was a modest disappointment, but shares rose in relief that the miss was not bigger. Still, sales, and in particular, storage system sales, decelerated for the 6th time in 7 quarters, with information storage up just 3.3% against an easy compare (Exhibit 12).
Exh 12: Quarterly Top Line Revenue Growth Rates of Major Enterprise IT Companies, 1Q2010-1Q2013
Cisco, which reports its fiscal quarters one month out of synch with calendar quarters, delivered an upside surprise for its April results, heartening the true believers looking for a secular rebound in enterprise IT spending. Overall topline growth was 5.4%, better than most other data center suppliers and up a tick sequentially, but still mired in the single digits for the 9th quarter of the past 10. Cisco reported growth in its data center segment, although that segment includes growing sales to the web-scale cloud operators, who have not fully adopted open source “software defined networking” solutions to cut Cisco and its high-margin gear out of the loop. That – and the breadth of Cisco’s business portfolio – makes it a difficult read across for the rest of the data center technology market. We remain concerned that the true effects of the coming exodus to the cloud and the rise of software defined networking (SDN) solutions have yet to be felt.
Bring Out Your Dead!
Our perspective shouldn’t be misconstrued as a prediction that sales of traditional data center servers, storage systems, structured data base management software, application software, networking gear, or services will suddenly plunge to zero as enterprises simultaneously pull their triggers on a massive wholesale move to the cloud. Rather, small steps toward the cloud offload some processing and storage intensive applications from enterprise data centers, relieving pressures on capacity and allowing IT managers to avoid additional spending that would otherwise have been required. The success of small steps will embolden larger steps in the future, perhaps allowing IT departments to downsize their internal operations somewhat, but still requiring some spending to support those applications that will inevitably remain in house. Not all applications make sense in the cloud – systems that are inherently local and that demand very low latency, such as shop floor control systems, are unlikely candidates for a move (Exhibit 13).
Exh 13: Universe of Enterprise Application Software – Suitability for the Cloud by Application and Current Spending on Cloud
So some applications will stay on premises, and IT departments will still need to upgrade and maintain the client/server systems and software necessary to run them. 50 years after the introduction of the System360 architecture, IBM is still selling some $4.3B per year in mainframe technology, a dramatic drop from its peak to be sure, but still a substantial market for a long obsolete technology (Exhibit 14). The client/server era is already in its own glide path to eventual oblivion, and though that path may be slow and bumpy, occasional upward blips should not be mistaken as signs of resurgence.
Exh 14: Worldwide Mainframe Market, 2009-16
Most of the companies dependent on enterprise data center spending have been in states of denial, pressing a “hybrid” approach that could give them a credible claim for a web future and making splashy acquisitions to help establish their cloud credibility. None of these half-solutions will be cost competitive long term with the huge distributed data center utilities and their open-source infrastructures. Trusting customers may be sweet-talked into following along for a while, but the unstoppable progress of technology will be against them and their shortcomings will become ever more apparent with each passing year.
We are concerned for the long term health of most of these companies, including the once-great leaders of the client server era – HP, Dell, EMC, NetApp, Oracle, SAP, VMWare, and even Cisco and IBM amongst them. Some of these companies will fail, like the DECs and Primes before them. Others will find new ways to thrive – we are enthusiastic for the future of Microsoft as a cloud host and SaaS application provider.
Winner, Winner Chicken Dinner
We firmly believe that cloud hosting is a scale game, and that companies that cannot leverage huge consumer web application franchises to drive infrastructure investment will be doomed to long term cost disadvantages. Google, Amazon and Microsoft have their own internal “anchor tenants” that compel them to build out massive distributed data center capacity, giving them massive advantage in the commercial hosting market. Given its own breakneck capital investment pace, it would be unsurprising to see social network king Facebook in this game at some point, with Apple a potential future player as well (Exhibit 15). Already, the big three have begun a tack of aggressive pricing and generous service definitions that will make it difficult for subscale entities, or application specific entities to keep pace on cost. To this end, we see Amazon, Microsoft and Google as big winners in the shift to the cloud. Would-be cloud hosts without high volume consumer cloud franchises, such as Google search, Amazon e-commerce or Microsoft Xbox live, will likely find themselves badly squeezed on costs in the future.
Exh 15: Capital Expenditures of Major Data Center Players
SaaS applications should also be a hot sector over the next decade (Exhibit 16). Salesforce.com is the pioneer, with an enviable head start in establishing its CRM service as an industry standard, but we remain somewhat concerned that the company’s commitment to traditional structured data base infrastructure and operating its own data centers will put it at grave long term disadvantage to competition that will lever more modern unstructured database approaches leveraging truly web-scale hosts like Microsoft, Google and Amazon. Workday and NetSuite are fast growing public SaaS companies addressing ERP apps, and there is a long line of prosperous private SaaS companies held back by the frigid tech IPO environment. Moreover, we believe Microsoft is uniquely well positioned to exploit its own infrastructure and its deep sales relationships with the global enterprise IT world to attack its traditional software competitors, Oracle and SAP, with well designed SaaS applications.
Exh 16: Winners and Losers
On the hardware side, component suppliers will continue to see dramatic volume growth driven by the investment of web-scale distributed data centers. Intel’s sales of server chips to fuel the growth of distributed cloud data centers have blunted the impact of shrinking sales to PC and commercial server makers. Longer term, we remain concerned that server chips based on the widely licensed RISC-based ARM processor designs favored by the mobile industry will eventually threaten Intel’s x86 server hegemony in the cloud market, but that threat is most likely several years away. We are bullish on future demand for disk drives, as total global storage needs continue to grow at more than 40% per year and competition is down to a well behaved trio of Seagate, Western Digital and Toshiba. Solid state storage is still far from cost competitive for the vast majority of cloud storage needs, but we believe that the growing niche market for extremely high speed storage will be more than sufficient to fuel early market leaders like Fusion I/O.
From an enterprise perspective, the move to the cloud is scary. We believe that most CIOs will choose to balance multiple cloud vendors, and will look to rely on 3rd party expertise to help them adjudicate the process. This is an opportunity for traditional IT consultants, such as Accenture, and for cloud-specific operations, like Rackspace, to pick up the gauntlet.