SaaS: After the Levee Breaks – Competition in Software
SEE LAST PAGE OF THIS REPORT Paul Sagawa / Artur Pylak
FOR IMPORTANT DISCLOSURES 203.901.1633 /.1634
psagawa@ / firstname.lastname@example.org
March 2, 2014
SaaS: After the Levee Breaks – Competition in Software
Web-scale cloud platforms, leveraging massive consumer applications, have dramatic cost and performance advantages over private data centers, stemming from their scale, scope, superior design, and world-class computer science skills. These advantages, which already allow the top cloud operators to deliver 50-75% lower costs to customers vs. the all-in costs of in-house solutions, are growing wider with time and are rapidly separating AMZN, GOOG and MSFT from smaller would-be rivals. As we have often noted, these dynamics are troubling for traditional data center technology vendors, and are likely to be deflationary for IT budgets in general. Low cost, high performance cloud hosting also greatly lowers the barriers to entry for SaaS application developers – opening the door to innovation and competition in enterprise applications, pressuring SaaS pioneers with older architectures, and posing an existential threat to traditional application vendors. Success in SaaS will come from innovation, execution, and scale economies rather than customer lock-in, and product life-cycles will be shorter than in the last 30 years of the software market.
- The cloud offers dramatically lower costs than traditional data centers. Public cloud distributed data center platforms have huge cost and performance advantages vs. private data centers based on several factors: 1) Use of commodity components vs. value-added configured systems; 2) Extraordinary power efficiency; 3) Much higher utilization; 4) Higher system availability; 5) Superior backup and redundancy; 6) Flexibility, scalability, power and convenience; 7) Substantial economies of scale. All-in comparisons give the top cloud hosts a 50-75% cost advantage vs. in-house enterprise data centers, and with cloud pricing dropping at roughly 30%/yr, the gap is rapidly widening. These factors are a particular advantage for AMZN, GOOG, and MSFT, which employ cutting edge data center design, sophisticated software, and massive scale, driven by demanding consumer cloud franchises. As rivalry intensifies, service offerings will expand and prices will continue to drop, squeezing smaller competitors. Currently, IaaS prices yield 50-75% cost benefits vs. typical in house all-in costs, a gap that is widening.
- SaaS applications offer advantages beyond cost. By bundling the costs of the data center, SaaS vendors offer substantial savings vs. the all-in costs of running an application in house. Beyond these savings, most of which derive from cloud hosting, SaaS applications offer significant functional benefits. First, SaaS is scalable and predictable – costs grow proportionally to use on a transparent basis. Second, SaaS reduces complexity – expert user support is included, upgrades are automatic, costless and painless, and expert system/software maintenance is included. Third, access to SaaS applications is ubiquitous and availability is superior. Finally, inherently mobile SaaS applications tend to be far more easily implemented on mobile platforms. With switching costs falling rapidly – most SaaS vendors have sophisticated transition tools, support traditional data formats, and mimic familiar user interfaces – and incumbent stoked security/reliability fears proving unfounded, adoption of SaaS is accelerating.
- Traditional data center spending has peaked. While investment on cloud data centers is exploding, the design of the most sophisticated operators excludes the value-added systems and software infrastructure that made up the bulk of enterprise spending during the PC era. Meanwhile, enterprise IT departments are scaling back internal investments as their focus shifts to cloud hosting and SaaS applications. Spending in every major data center category has sharply decelerated over the past three years, with even the most robust categories, like storage systems, on a trajectory toward decline. While the savings from cloud solutions may yield some elasticity in overall demand, the price differential assures that the net effect will be extremely deflationary for IT budgets, with the traditional tech players – HP, IBM, EMC, Cisco, Oracle, SAP, and others – on the losing end.
- Lower barriers to entry may leave SaaS margins below traditional software. The rise of AWS and the aggressive response from MSFT and GOOG will change the competitive dynamics of the SaaS market. Historically, SaaS vendors have required substantial upfront investment in building out data centers ahead of product launch – CRM, N, et al were built this way. IaaS allows new players in with minimal capital investment, bringing a flood of new would-be entrants to the market. However, where SaaS vendors can differentiate their products, we see ample opportunity for profitability. We see four main paths to differentiation – 1) functional innovations; 2) community-based network effects; 3) cross-product ecosystem synergies; and 4) system learning. Loss of customer switching friction will pressure margins across the industry, particularly players carrying private data center infrastructure, as newer rivals leveraging increasingly low cost web-scale cloud hosts are able to undercut established prices based on fundamentally lower COGS.
- Life cycle business model is very different for SaaS. The subscription model spreads SaaS sales recognition fairly evenly through the life of a customer relationship, while expenses are front end loaded. For companies in growth mode, even at the size of Salesforce.com, this severely squeezes reported operating profit. However, as customer contracts age, they should grow continually more profitable, without the periodic sales expense of pushing upgrades, and, as the relative investment scales back in a normal growth environment, operating margins will rise. Ultimately, SaaS profitability will depend on product differentiation, and leveraging data center scale advantages. This last factor will be a strong advantage for IaaS operators competing in SaaS apps. We do not expect SaaS only players to approach the 30%+ margins earned by software leaders in the last cycle, as it will be difficult for them to make margins on the infrastructure costs embedded in their business model.
- Invest in sustainable differentiation. Differentiation will be on the basis of functional innovation, network effects, ecosystem effects, and system learning. We are concerned for the traditional application leaders, such as Oracle, SAP and IBM. For the most part, their SaaS initiatives are compromised by the imperative to sustain the profitability of the classic franchises – a dynamicwell described by Clayton Christenson in “The Innovator’s Dilemma”. We are cautious about many of the SaaS pioneers – e.g. N, CNQR and others – concerned that their in-house data centers will struggle to remain competitive with solutions hosted on the big three IaaS platforms. We are enthusiastic about differentiated SaaS players that have embraced 3rd party hosting – i.e. WDAY. Finally, we expect the IaaS market to break into haves and have nots, with AMZN, GOOG and MSFT on track for long term dominance.
The SaaS is Always Greener on the Other Side
The idea of providing software applications as a service over the internet, provided from a centralized host data center has precedents all the way back to the original IBM mainframes, but really gained momentum during the dot.com era. Data center outsourcing – where specific customer-owned applications were moved to shared, off-premises 3rd party hosts – inspired entrepreneurs to offer shared applications as well. Companies like CRM and N rose and flourished, selling enterprises on the efficiency of the cloud, the universal availability of the web, and the convenience of turning software maintenance, support and upgrades over to the vendor. These SaaS pioneers relied on the data center technology of the day – clustered x86 servers and RAID storage, running LINUX and structured Oracle DBMS software – indeed, both CRM and N were founded by former Oracle executives. Early on, this was an advantage, as familiar systems and data compatibility made transitions from traditional solutions easier to sell and execute.
However, in the same late ‘90’s timeframe, another software revolution was also underway. GOOG, faced with the problem of indexing the entire Internet, was inventing truly web-scale data management tools, transcending the limits and rigidity of ‘90’s structured data bases, clustered servers and RAID storage. GOOG’s approach, contributed into the open source software community, became the underpinnings of a new paradigm for cloud data centers based on commodity hardware, massively parallel computing, and unstructured data. These operations, scaled for huge consumer cloud franchises like GOOG search, AMZN e-commerce, and MSFT Xbox Live, have dramatic cost and performance advantages vs. traditional enterprise data centers, including those run by the SaaS pioneers.
These advantages are available to the commercial customers of AMZN Web Services (AWS), MSFT Azure and GOOG Compute Engine. On an all-in basis, including the often forgotten costs of electricity, real estate, personnel, utilization, maintenance, back-up/redundancy, downtime, etc., hosted cloud services offer a 50-75% cost reduction vs. typical in-house enterprise data centers, a gap that has been widening with the relentless price competition amongst the big three cloud hosts. Indeed, we believe that the scale advantages of AMZN, GOOG and MSFT vs. all other would-be rivals are vast, and that the three leaders will capture the large majority of future hosting opportunities in the US.
The shift to the cloud-hosted paradigm has dramatic implications for the SaaS market. Barriers to entry for new applications are de minimus, with world-class cloud hosted infrastructure available in narrow increments and tools to transfer existing data bases widespread. This will pressure margins for existing SaaS players and may sound the death knell for packaged software hold-outs. Customer lock-in, based on data formats, user training, or customized application extensions, is eroding with technical approaches that minimize them and pricing strategies that amortize the switching costs. Competitive advantage will rise from real patent-protected innovation, from the network effects of combining many user organizations of a single platform, and from customer-specific learning derived from the data over time.
We believe many SaaS winners will establish themselves over the next few years, and that existing leaders, like CRM and N, must adapt their private data center model to exploit the cost and performance advantages of the public cloud, following the lead of WDAY in that regard. In either case, we look for competitive pressures to rise, even as the opportunity set greatly expands.
Exh 1: Major Events in the Cloud, 1997-2013
The Sons of Oracle
Larry Ellison saw the first part of the cloud coming. In 1998, Ellison backed Evan Goldberg, a former Oracle executive in starting the pioneering cloud-based Enterprise Resource Planning (ERP) system company NetSuite (Exhibit 1). ERP was an audacious product, championed by German software company SAP, that combined most of the functions needed to operate a business – product development, manufacturing, inventory and purchasing, marketing and sales, logistics, financials and human resources – into modular system that operated from a single underlying data base. During the ‘90’s, big, expensive ERP deployments consumed many a Fortune 500 IT department, many racing to beat the perceived year 2000 bug deadline, and Goldberg’s big idea was to extend the benefits of huge, complicated ERP systems to small business without the resources to operate them on their own.
NetSuite’s ERP was designed to be shared by many organizations while running on the company’s own centralized data center. By selling access to the system as “a service” delivered over the Internet, the benefits of an integrated ERP solution could be made affordable for much smaller organizations. In most respects, NetSuite was a “normal” ERP system, with modules corresponding to the connected functions of a business built on a common Oracle-supplied data base infrastructure and running on clustered x86 servers with RAID storage systems. Yet customers were spared the costs of building and maintaining their own data center capacity, ceded responsibility for user support and future upgrades to the vendor, and gained the convenience of “pay as you go” expansion. The concept was revolutionary, but for many small businesses, perhaps overwhelming. NetSuite’s early growth was attractive, but not explosive.
The second major SaaS pioneer, Salesforce.com, was also born of Oracle heritage – founder Marc Benioff was a 13 year veteran. Launched in 1999 with a tight focus on Customer Relationship Management (CRM) software, Salesforce.com hit a sweet spot in the nascent SaaS market. CRM, used by sales organizations to document the selling process and offer analytical support to company representatives, is an inherently mobile application and a perfect fit for SaaS. Cloud access was and is an important advantage, as was the ability to scale the tool to the size of the customer organization. Like NetSuite, Salesforce.com’s applications were built atop an Oracle RDBMS and ran on a traditional enterprise data center, but unlike NetSuite, Salesforce.com experienced explosive growth. A 2004 IPO valued the company at more than $1.1B, as growth spurred entry into adjacent market opportunities and the launch of marketplace for compatible 3rd party SaaS applications able to leverage the underlying data base.
Along the way, a range of new SaaS players joined the fray, while the big traditional application software houses worked to preserve their customer “lock-ins” – proprietary data formats, customized modules and user interfaces, etc. – while sowing “fear, uncertainty, and doubt” about the switching costs and security/reliability of cloud-based solutions. With time and experience, SaaS players began to crack the problems behind “lock ins”, ease transition costs and counter security/reliability concerns, while enterprise IT managers began to more broadly accept SaaS alternatives. With the growing success of Salesforce, NetSuite, WorkDay, and other SaaS players, the enterprise software leaders jumped in with SaaS acquisitions. SAP stepped up for Business Objects ($6.8B), Success Factors ($3.4B) and Ariba ($4.3B), amongst others. Noted SaaS skeptic Oracle played ball as well, spending nearly $8B on 6 publicly traded SaaS companies, including RightNow and Taleo, during the last three years (Exhibit 2).
Exh 2: SaaS Acquisitions by Major Enterprise Software Companies
By 2012, Salesforce.com had passed Oracle to become the number one CRM solution in the market, driving the company toward its current market cap of more than $38B, a better than 10x multiple of trailing sales with consensus projections of nearly 30% forward sales growth (Exhibit 2). Curiously, NetSuite, also projected at roughly 30% top line growth for 2014 after a big acceleration over the past three years, trades at an even richer 20x trailing sales multiple. Workday, a SaaS Human Resources Management (HRM) specialist, with a $17B market cap tops both, trading at 40x ttm sales with projections of 50% sales growth. In total, we found 34 public SaaS focused companies with market caps of at least $500M (Exhibit 3). The median projected growth for these companies was 24.4% and the average trailing sales multiple was 8.0x. Clearly SaaS has hit its stride.
Exh 3: Global CRM Vendor Market Share, 2012
Exh 4: Publicly Traded SaaS Companies, >$500M in Cap
Meanwhile, On the Other Side of the Cloud …
While the founders of NetSuite and Salesforce.com were having their epiphanies about delivering application software to users over the Internet as a service and charging on a subscription basis, Larry Page and Sergei Brin were finding a way to index the entire World Wide Web on a single system of servers and to deliver lightening fast search results based on natural language key word queries. The problem was bigger than the technology of the day – those same clustered x86 servers running vendor supported Linux with RAID storage systems and that tightly structured Oracle RDBMS software. Structured data bases had strict limits – only so many records in a data base, only so many entries in a record, each entry following an exact syntax with an exact number of characters, etc. – on the belief that such limits were the only way to assure the integrity of the data and to allow users to parse through the data quickly. Google, under the leadership of founders Page and Brin, found a different way.
Google collected unstructured descriptors for every page available on the Internet. When a query of the enormous data base was entered, the system would break the problem into sub-searches of component chunks of the data base, spreading the task over many, many server cores in parallel, a process known to programmers as “Map”. The results of the many mini-searches were then compiled, analyzed and summarized, a process called “Reduce”, with the results returned to the user in seconds. With Google’s approach, the search process was exhaustive, rigorous and fast without imposing limits on the size of the data base or the format of its contents. Eventually, Google contributed its revolutionary approach to the open source software community, where Google Map/Reduce became the progenitor of Hadoop and essentially all the other “big data” software infrastructure that underpins modern cloud data processing (Exhibit 5).
Exh 5: MapReduce/Hadoop parallel computing flow
We Can Make it Better Than it Was – Better? Faster, Stronger …
Removing the limits on the scale and formality of data bases also challenged the traditional hardware architecture for data centers. The biggest enterprise data centers, including those built by NetSuite and Salesforce.com, broke processing capacity into finite server clusters, sized to match the requirements of the specific data base software running atop it. Processor clusters were matched to storage arrays, which used a software logic called RAID to monitor and manage the integrity of the data contained within. These systems, sold by the leading IT hardware vendors, were expensive, containing proprietary software and other bells and whistles intended to tweak superior performance from each element of the system – hot back ups, environmental control systems, uninterruptable power supplies, arrays of sensors and indicators, virtualization software, system management dashboards, etc. As a result, enterprise data centers were complex and expensive, but limited.
Google invented another way. No expensive dedicated server clusters with matching RAID storage – Google bought server chips wholesale and hired Asian contract manufacturers to install them on bare bones circuit boards alongside commodity disk drives. The boards were then installed into open racks in massive warehouse-like data centers in locations chosen for temperate weather and proximity to telecommunications. The servers are brought to bear against computational problems, assigned by the core software infrastructure as needed to solve the atomized tasks in parallel. Data is stored directly onto the disk drives, relying on the software core and on multiple redundant copies of each record to assure data integrity.
These days, all of the top cloud-based hosts are employing these techniques, even as Google continues to drive the state-of-the-art ever forward. Consumer cloud applications – like search, maps, e-commerce, social networking, video streaming, multiplayer gaming, and others – are particularly demanding, asking for mind-boggling scale and speed. As such, it is not surprising that the market leaders in cloud hosting, Amazon, Microsoft, and Google, all tailored their investment in data center infrastructure to their own considerable consumer franchise needs before applying their assets toward commercial hosting.
Let Me Count the Ways
Web-scale distributed cloud data centers have inherent cost and performance advantages vs. smaller data centers based on the clustered PC/RMBDS paradigm. The first factor in those advantage is the use of commodity hardware components rather than value-added configured systems. For comparison, Google was, reputedly, Intel’s 5th largest server chip customer in 2012 – with a bullet – with white label server makers Quanta, SuperMicro, and Wistron, which make cheap and dirty servers to order for cloud-customers like Amazon, Microsoft, Facebook and others, also somewhere in the top ten or so server makers according to the head of Intel’s server group. Given Google’s likely ability to negotiate terms with Intel on par with traditional server makers like HP, Dell and the IBM server business being acquired by Levono, and given the bare bones design of the Google boards, we peg Google’s cost per server core at less than $600, while configured systems of equivalent capability list for more than $1900 (Exhibit 6). Storage hardware has similar dynamics – a networked RAID system runs about $779 per terabyte, while basic disk drives cost less than 10% as much, allowing for considerable storage redundancy to make up for the superior resilience of RAID systems (Exhibit 7). Other hardware elements of the data center – racks, power supplies, networking gear, etc. – likely follow along the same lines.
Exh 6: Server / Component Cost Comparison, Traditional Vendors vs. Google
Exh 7: External Controller – Based Storage Cost per TB, 2011-2017
Electricity is another key savings for the big cloud hosts. Power can eat up as much as 25% of the total operating costs of a data center, and the leading operators work VERY hard to conserve as much as possible. The most commonly used measure of data center power efficiency, Power Usage Effectiveness (PUE) compares the total data center power draw to that required specifically for the computing, storage and communications tasks performed (Exhibit 8). Power used to control the climate, provide light, ensure physical security, or any other extraneous purpose is measured as waste, above the perfect rating of 1.0. The most recently published figures show Google’s company-wide PUE at an astonishing 1.1. Microsoft claims 1.13-1.2 for its data centers. Amazon does not provide PUE statistics but is believed to be somewhere within Microsoft’s range. In comparison, the average US enterprise data center has a PUE greater than 1.8, meaning that its power use is roughly 75% higher than Google’s for the same amount of computing. Salesforce.com’s PUE is 1.47, considerably better than the average corporate data center, but still implying roughly 35% more energy use than Google, and 20-30% more than Microsoft (Exhibit 9). We note that PUE is itself not a perfect measure, as it assumes that the electricity required for computing tasks is a constant. The leading cloud hosts are also leading the charge to reduce the power draw from the productive elements of the data center – Google is reputedly exploring self-designed server chips based on the fundamentally lower power ARM processor standard, while all of the big players have already begun to make use of power-friendly solid-state storage for applications where its speed advantages vs. disk storage are valuable.
Exh 8: Calculating Power Usage Effectiveness (PUE)
Exh 9: Distribution of Large Data Center PUEs, 2012
Utilization is the next major cost consideration. Obviously, the less that valuable resources lie fallow, the more broadly data center costs can spread across productive use. Here, cloud hosts with major consumer franchises gain a great benefit, as the distribution of user demand spreads more readily outside of traditional business hours. Hard numbers on server utilization are hard to come by, but a 2012 Gartner study pegged average enterprise data center CPU utilization at 12%, up from a 2008 McKinsey study, which had found 6% utilization. In comparison, data provided by Google in July 2013 suggested that a typical server cluster in its data center network operated in the 10-50% utilization range almost all of the time, with an average utilization in the 30% range (Exhibit 10). Given that most of the costs of operating a data center are incurred no matter its utilization, this is a crucial factor in determining the cost of a unit of computing. The anecdotal data suggests that Google, and likely, Amazon and Microsoft as well, keeps its servers more than twice as busy as the average enterprise data center.
Exh 10: Server Utilization Rates
Despite the lurid press associated with data center failures in the cloud, the big cloud providers have an excellent track relative for avoiding system downtime in comparison to traditional private data centers. A recent study by the Poneman Institute showed that 91% of enterprise data centers have had incidents that completely disrupted service within the past two years, with an average of two outages per data center, averaging nearly 2 hours of downtime per outage. According to Poneman, each 2 hour outage carries about $900,000 in associated costs (Exhibit 11). Including partial outages and scheduled downtime, industry professionals peg typical enterprise data center availability at 95-98.5%, compared to the Service Level Agreements offered by the leading cloud hosts promising 99.95% availability. A back-of-the-envelope calculation by searchcloudcomputing.com suggests that AWS was able to deliver better than 99.5% application availability in 2012, below its SLA, but still well above the generally accepted estimates of typical private data center availability (Exhibit 12).
Cloud providers also have advantage in managing data and process backup/redundancy. The distributed architecture of the modern cloud data center automatically provides geographic diversity, with data typically stored redundantly at multiple locations. Processing loads and network communications automatically shift to alternative, available data centers should any one site suffer failure. All of this is handled in the system software infrastructure, saving the cost and complexity of functionally specific back-up systems, such as RAID storage. With these techniques, cloud operators can offer lower systemic risk to customer data than within traditional data centers at a lower overall cost.
Exh 11: Causes of Data Center Outages, 2013 Survey
Exh 12: Cloud provider Service Level Guarantees
Beyond integrated backup, the web-scale approach of cloud architecture also offers valuable benefits to commercial customers – no practical limits on application scale, ability to accommodate enormous short term usage spikes, predictable all-in pay-as-you-go pricing, global scope, automatic system upgrades, world-class IT support and maintenance, and many others. To illustrate one of these advantages, Google’s Senior Vice President of Infrastructure, Urs Hotzel, demoed the capabilities of the company’s Compute Engine hosting service during the 2012 I/O developers conference, employing more than 750,000 server cores in parallel to complete an analysis of a human genome in less than a minute, vs. the 15 hours that it usually took on the customer’s own computing infrastructure. The combination of enormous power, complete usage flexibility and all inclusive pricing, available on a global basis with little previous notice is a powerful selling proposition.
Finally, all of it has enormous economies of scale. Hardware purchases are cheaper. Bandwidth is more readily available and cheaper. Real estate is better suited to the needs of a data center and cheaper to boot. The best in IT talent is attracted to the challenges at the cutting edge, and can be leveraged across a massive base of IT driven business. Custom hardware and software development makes sense when applied to lucrative consumer cloud franchises and can be utilized in the hosting business along side.
The Big Three
Amazon made its commitment to hosting from the cloud a decade ago, launching a simple version of what would become Amazon Web Services in November 2004. Amazon’s commitment to the public hosting business was prescient. Google’s App Engine – a platform for launching web sites – followed as a 2008 preview before moving to general availability in 2011, with the more general Compute Engine “Infrastructure as a Service” (IaaS) product launched a year later in 2012. Microsoft offered its “Platform as a Service” (PaaS) service, Azure, beginning in 2010 limited to offering Microsoft’s own infrastructure offerings as a service, gradually opening it to a wider range of software choices and morphing into a true IaaS platform. All together IaaS and PaaS hosting services are a $10B+ global industry, according to Gartner, growing at a torrid 45% annual pace (Exhibit 13).
Exh 13: Worldwide Cloud Infrastructure Services Forecast, 2010-20 CAGR: 42.6%
Exh 14: Estimated IaaS/PaaS Revenue for Top 5 providers, 4Q13 vs. 4Q12
AWS, Azure and Compute Engine are the three best positioned cloud hosts in the market, with Amazon well in the lead, with better than 27% global market share according to Synergy Research, based on its head start and its customer-focused, flexible approach (Exhibit 14). While AWS remains the platform of choice for 3rd party web sites and supplementary capacity for enterprise IT, both Microsoft (5.6% share) and Google (4.9%) have gained more than a foothold in the market and are growing with a bullet, with advantages of their own in competing for the rush of customers heading to the cloud. All three leaders have massive scale, born of their consumer cloud franchises, and leading edge data center design, yielding substantial cost and performance advantages (as detailed in the previous section), not only vs. private enterprise data centers, but also vs. smaller would be rivals eying the IaaS hosting opportunities and SaaS application providers operating from their own infrastructure.
Comparing the big three, Google’s infrastructure is the biggest, cheapest and most advanced, and has made a splash by offering pricing in one minute increments (10 minute minimum) rather than the customary hour, but it has not been as flexible in supporting traditional software and it has been relatively quiet in marketing the service. Amazon has a well established reputation as a cloud host, with aggressive pricing and big anchor tenant clients like NetFlix driving a 40-50% share of the IaaS market, although it much stronger in the web community than with enterprise IT shops (Exhibit 15). Microsoft has strong, direct sales relationships with nearly every enterprise IT professional and is matching AWS price cut for price cut, but it was late in supporting non-Microsoft tools and has a jaundiced reputation amongst web start ups.
The rivalry amongst the big three has driven web hosting prices down sharply, while spurring improving customer service and service flexibility. Over the past two years, AWS has cut its pricing for CPU time and storage by more than 50%, with Azure and GCE largely following suit. Google offers CPU time billed in one minute increments, with a 10 minute minimum, giving customers with variable workloads a considerable benefit vs. Amazon and Microsoft, which bill in one hour increments (Exhibit 16). Microsoft offers virtualized windows desktops, a valuable service recently matched by AWS and soon to be offered by Google in partnership with VMWare. We believe that the aggressive moves of the big three, backed by their inherent cost and architectural advantages vs. rivals without consumer franchise driven scale, will separate them from the pack. We believe the would-be rivals, such as IBM, RackSpace, Verizon, Salesforce.com, HP, VMWare and others, will be squeezed, even as the overall market continues to grow at a torrid pace.
Exh 15: Downstream Netflix Traffic on North American Fixed Broadband networks, 2H10-2H13
Exh 16: Sample Basic Cloud Services Pricing from Major Providers – January 2012, May 2013, February 2014
Hey You! Get Off of My Cloud!
The big three hosts do not buy off the shelf technology. No configured servers. No RAID storage systems. Very few “value added” networking boxes. No commercial infrastructure software – data base systems, virtualization hypervisors, branded LINUX, etc. – save for that specifically requested by hosting customers. No expensive maintenance contracts for hardware or software. Most new web-based businesses – from NetFlix to SnapChat – have embraced the advantages of using IaaS hosts, focusing on differentiating their applications rather than trying to outcompete Amazon, Microsoft and Google on infrastructure, and those that have decided to build out their own distributed data centers – most notably, Facebook and Twitter – follow the same architectural principals as the hosting leaders. Thus, the ONLY real growth market in enterprise IT has its doors closed to the proud ranks of traditional IT technology vendors.
Exh 17: Basic On-Premise versus Cloud Cost Comparison
Of course enterprises are still investing in their internal data centers, but growth has decelerated into decline for most of the important product categories, as enterprises shift their emphasis to the cloud. The cost advantages of the cloud model are profound. On an all-in-basis – accounting for the often hidden costs of things like electricity, real estate, IT personnel, telecommunications, backup, downtime, utilization, and maintenance – public cloud solutions are typically 50-75% cheaper than an internal data center (Exhibit 17). Moreover, while costs for the big hosts continue to drop with improving efficiency, lower component costs, and growing scale leverage, most of the costs for internal data center are stagnant. With the likes of AWS, Microsoft and Google seemingly hell bent on passing most of their cost improvements directly on to their hosting customers, the “make vs. buy” decision for enterprise CIOs becomes an increasing “no brainer”.
Exh 18: Gartner’s Global Data Center Systems Spending Forecast Revisions, 3Q12 – 4Q13
This is unambiguously awful for traditional IT vendors – IBM, HP, Oracle, VMWare, and others talk big talk about their own plans for the cloud, but fighting for hosting market share against the likes of Amazon and Google with sub-scale data centers and dated architectures is a losing proposition. Meanwhile, fewer systems are sold and fewer infrastructure software licenses are signed (Exhibit 18). Now add in the effect of SaaS applications. If customer X decides to buy CRM or ERP or HRM or some other major enterprise application from a SaaS vendor, not only does the packaged applications supplier lose the contract, but so do the rest of the systems and infrastructure software suppliers to customer X’s data center.
The mechanics of this are almost certainly deflationary for enterprise IT budgets. Assuming a 50% all-in cost break for moving applications to the cloud, it seems inconceivable that demand would elastic enough not to drive total enterprise data center spending lower. Of course, Gartner is characteristically optimistic, forecasting 3.1% growth in overall enterprise IT spending for 2014, a sharp reacceleration after essentially flat spending in 2013. Gartner even projects a return to growth for data center systems, a category that has been decelerating for three years and declined half a percent in 2013. However, we note that the 1Q13 forecast is a downward revision from an even sunnier projection published in October (Exhibit 19). Indeed, graphing Gartner’s quarterly forecasts reveals a long legacy of over-optimism and subsequent downward revisions as hockey sticks get continually pushed to the future.
Exh 19: Gartner’s Aggregate IT Spending Forecast Revisions, 2Q11 – 4Q13
We believe that the shift to the cloud will be more faster and more pronounced than Gartner, or most other industry observers, expects. This will drive overall IT spending, and in particular, data center IT spending, on a downward trajectory going forward, despite the explosive growth of cloud hosting and SaaS. Companies that derive significant revenues from value added system and software sales into the enterprise data center market – IBM, HP, EMC, NetApp, Oracle, SAP, Cisco, and others amongst them – will suffer accordingly (Exhibit 20-21)
Exh 20: Gartner’s Global Server Spending Forecast Revisions, 1Q13 – 4Q13
Exh 21: Enterprise Routers and Ethernet Switches, 2012-2017
What Can You Make Doing That?
Because SaaS players must pay for data center infrastructure, gross margins tend to be lower than the 80-90% enjoyed by premises software leaders like Microsoft, Oracle and SAP during their heydays. Salesforce, which is, by far, the largest scale predominantly SaaS application player earns about 75% gross margins, with most of the rest of the top 10 SaaS companies by market cap coming in between 50 and 70% (Exhibit 22). Of the top 10, only The Ultimate Software group and MediData are robustly profitable on an operating basis, with Concur and AthenaHealth ekeing out a bare 1% profit, and the remainder losing money after expenses.
Exh 22: Publicly Traded SaaS Companies, >$500M in Cap
This pattern is misleading. The median growth rate amongst the top 10 is a robust 36%, with the slowest growing amongst them, The Ultimate Software Group at 21%, also the most profitable, with better than 10% operating margins. This is not a coincidence. Unlike packaged software, where most of the revenues of a sale are recognized right up front, SaaS sees its revenues spread across regular monthly payments. As such, investments in R&D for new products and Sales and Marketing to sign new customers show a much longer payoff. While the companies are in rapid growth, the net result is weak operating profit with investments toward much higher future revenues matched against today’s sales. The intrinsic profitability of such a business is far higher than its reported margins, a reality that will emerge as growth inevitably decelerates.
We would expect the steady state margins of a SaaS player to range as high as 25% – slightly below the historic profitability of the best packaged software vendors – as we do not expect SaaS players to make super-normal margins on the hosting component of their product offering.
Hey – You Got Chocolate in My Peanut Butter!
On one side, we have SaaS applications, led by old school pioneers like Salesforce.com and NetSuite, delivering turnkey solutions levered by the ubiquity of Internet access. On the other side, we have IaaS hosts, offering serious improvements to data center performance and efficiency at enormous cost savings to traditional IT architectures. Recently, new market entrants have looked to exploit both of these trends at the same time. Like the pioneers, new SaaS companies look to make the switch from licensed software run on premises to cloud-based applications as easy as possible – supporting standard data formats, creating turn-key migration tools, enabling easily customized user interfaces, etc. – negating the “fear, uncertainty and doubt” that traditional software players have used to “lock in” their customers, while offering predictably lower all-in costs, convenience, flexibility and automatic upgrades. As experience negates fear mongering about transition costs, security and availability, the SaaS model has picked up momentum, benefiting old and new SaaS competitors alike (Exhibit 23).
Exh 23: Worldwide SaaS Forecast, 2011-2017
However, new SaaS players are beginning to also exploit the benfits of cloud hosting. The old guard of SaaS companies built their own data centers – Salesforce.com has PPE assets of $1.2B with another $500M in capitalized software and operating lease obligations of more than $1.5B – based on the traditional architectural model of commercial blade servers, RAID storage and structured data base system software (Exhibit 24). These data centers were tremendous assets during the first decade of growth, but can be viewed as something of an albatross today. While Salesforce and NetSuite’s data centers are almost certainly far more efficient and better utilized than the typical enterprise facility, they are also almost certainly far less so than the big cloud operators with whom they must now compete.
Exh 24: Net PPE and Operating Lease Commitments of SaaS companies, >$500M in Cap
Amazon Web Services, Google Compute Engine and Microsoft Azure remove enormous entry barriers for software developers looking to launch a SaaS enterprise application. Instead of raising significant capital to build out their own data centers, developers can pay for time on AWS as needed, eliminating upfront expenditures and tying expense growth directly to revenues. Moreover, as we have asserted, the all-in costs of relying on IaaS cloud hosts is far less than an in house solution, particularly to a start-up operation which may not want to devote precious engineering time to the mundane tasks of building and maintaining infrastructure. While AWS has been an option for SaaS start-ups for years – note that many more recent market entrants, including the highly successful WorkDay, have moved in this direction – the recent price competition amongst the leading cloud hosts is an enormous boon to new competitors.
Best Application Wins
In the cloud era, the traditional software customer lock-ins are greatly weakened, even as the barriers to entry for new application providers have fallen. We believe that the result will be a substantial reordering of the overall software market, as the traditional leaders rush to embrace the principles of SaaS, as the pioneering SaaS players look to bulk up their scale, and as myriad new entrants try to topple the leaders with competitive costs and innovative functionality.
The market leading traditional software vendors – specifically, Oracle, SAP, and IBM – have been facing the classic incumbent’s dilemma. Given the strong incentives to protect their extraordinarily profitable core businesses, these incumbent players act to defend the older paradigm rather than fully embrace the new one. This phenomenon has been deeply explored by Dr. Clay Christenson, amongst others, and has been demonstrated to be a powerful force in the disruption of industries. All three enterprise application leaders have looked to buy their way into SaaS, but operate their cloud properties in ways that undermine their ability to disrupt the classic product lines. The cloud products run on internal data centers using the in house infrastructure software, premium priced as a recognition of their market position and to avoid being overly alluring to existing software customers. The one big exception to this dreary prognosis is Microsoft, which has been proactive in redefining its flagship Office application suite as SaaS product, Office 365, offering backward compatibility and fresh cloud functionality at a competitive price. It helps that Microsoft, unlike other enterprise software leaders, is an old hand at cloud infrastructure, with modern architecture and consumer franchise driven scale.
Exh 25: Worldwide Enterprise Application Software Market, 2011-2017
Ultimately, we think that continuity with existing applications and longstanding customer relationships will only go so far to sustain SaaS at traditional software houses. The cost advantages of web-scale distributed cloud infrastructure, the superior performance of the modern data center architecture, and the functional innovations of truly SaaS solutions are too compelling to be held at bay. The transition is already happening. SaaS applications are already a $22.6B piece of the $130.4B enterprise applications software market, a measure that underestimates the penetration, as SaaS applications are paid on annual contracts while most software licenses are sold upfront, and are growing at a 19.2% annual pace according to Gartner. We believe SaaS growth is higher and at least 25% (Exhibit 25). Cloud IaaS and PaaS hosting services are a $13.1B market growing at 42.6% per year. The nexus of the two cloud trends will put the defensive SaaS efforts of the traditional enterprise application leaders to heel, and even, threaten the hegemony of existing SaaS leaders. The best applications will win.
Exh 26: Universe of Enterprise Application Software – Suitability for the Cloud by Application and Current Spending on Cloud
What is Best?
With traditional customer lock-ins weakened and barriers to entry falling, applications must differentiate if cost is not to be the primary buying factor (Exhibit 26). We see four basic categories of differentiation that will work in the emerging new era of SaaS applications:
First, applications can offer true functional innovations that are not readily matched by competitors. For example, while the functional requirements of most standard enterprise apps have been well established, the rise of mobile platforms has opened the door to innovations that can help facilitate mobile use. Second, SaaS applications can drive network effects, whereby the adoption of the application by one party generates value for the other users of the same application. Business to business commerce systems are an excellent example – once on-line marketplaces are established, it is difficult to move the collected parties to an alternative marketplace. Similarly, a company may tie multiple applications together, gaining synergy from the use of common data where clear standards for sharing data between applications have not been well
established. While such standards are in place for classic applications, such as ERP, they are less defined in newer areas, such as social networking, where Salesforce.com has established its Chatter communications platform as an innovative value-add to its CRM and HRM products. Finally, applications may employ learning techniques that improve performance with ongoing use. “Big Data” business analytic tools are a salient example here – unraveling patterns in past company data allow for better insights in analyzing new company data.
Where SaaS companies can establish real differentiation for customers, the business has the potential for extraordinary margins – similar to those that have been enjoyed by traditional software leaders, albeit with a higher intrinsic COGS (infrastructure amortization, data center costs, and/or hosting costs) but lower SG&A (upgrade sales, post sale support, multi-platform development, etc.). Moreover, while there is less economic “lock-in” on customers, there is also no upgrade cycle to spur period review, rewarding customer service and performance to contractual metrics in ways that they are not for traditional software vendors. However, this structure is hidden in the current results of SaaS players, as the subscription revenue model delays revenue recognition and the hypergrowth pushes the need for investment now. As the SaaS model becomes prevalent in the software market and growth levels off to more sustainable levels, we believe available margins will be strongly positive, although the 30%+ operating margins earned by Microsoft and Oracle during their salad days may be out of reach, even for well differentiated applications, given the scale economies on the hosting side.
Winners and Losers
We see significant disruption in application software over the next few years. We are very pessimistic for traditional software players – Oracle, SAP, and IBM come to mind – given the market pressures on their core business and their weak start in pursuing SaaS. We are cautious about the old line SaaS players given the long term disadvantage in operating from subscale infrastructure and in holding to the traditional software infrastructure, but those that can differentiate and maintain non-infrastructure scale advantages may prosper – In this, we are more optimistic for Salesforce, ServiceNow and Concur than we are for NetSuite. We are enthusiastic for emerging SaaS competitors that are successfully leveraging the revolution in cloud hosting – WorkDay, which has launched its new services on AWS, is the most established player moving in this direction, but there is an army of pre-public start-ups moving into place. We are very optimistic for the cloud IaaS hosting businesses at the big three, Amazon, Google and Microsoft, but concerned for the cost and performance competitiveness of smaller would-be rivals, like Rackspace, IBM, HP, Verizon, and others, which lack massive consumer franchises to drive scale and remain well behind the leaders in the sophistication of the data center architecture.