Apple: AI Winter is Coming

Print Friendly, PDF & Email

SEE LAST PAGE OF THIS REPORT Paul Sagawa / Artur Pylak


psagawa@ /

twitter.jpg @PaulSagawaSSR

November 27, 2016

Apple: AI Winter is Coming

AAPL is poorly positioned for opportunities created by deep learning technology, including often cited areas of future company growth, like autonomous cars, virtual assistants, and augmented reality. Ironically, the Jobsian precepts that drove the company’s domination of the smartphone era – e.g. device primacy, the integrated functional organization, data privacy, company secrecy, and DIY self-sufficiency – will be big obstacles to making up lost ground on AI. This is of critical importance, as market maturity threatens to pressure iPhone sales and margins, the engine of AAPL’s extraordinary success. AAPL’s AI problem is apparent in all three of the main ingredients for leading edge deep learning. Talent – AAPL’s roster of AI scientists is extremely weak in both size and experience relative to its chief rivals, GOOGL and MSFT. Secrecy policies have been a substantial impediment, but these may have been dropped recently to hire a noted AI guru with a mandate to build a world-class team. We are concerned that this may be too little, too late. Data – With tight control over a billion iOS and OSX users, AAPL could be amassing powerful dataset assets, but its longstanding commitment to user privacy hamstrings availability and usefulness of this information. Platform – AAPL’s devotion to device primacy leaves its data center strategy in disarray and its cloud applications (i.e. iCloud, Maps, Music, Photos, etc.) disappointing. AAPL may need to partner with an AI-optimized cloud host (e.g. GOOGL, MSFT, or AMZN), a potentially bitter pill. A transformative acquisition or a major partnership would be obvious solutions to the AI deficits, but either would be anathema to AAPL’s deeply embedded culture and philosophy. Our model rates AAPL as “cheap” but given unrealistic near term expectations and a broken narrative, we suggest waiting on the sidelines.

  • Smartphone maturity. The premium segment for smartphones is nearly saturated. The gap in performance between the iPhone and its rivals has narrowed, while the marginal benefits of spec improvements are minimal. Devices with “good enough” specs at prices less than half of the iPhone are taking swaths of smartphone share in emerging markets. Meanwhile, cloud delivered AI virtual assistants could begin to crowd out the app model as the primary device user interface, potentially giving products powered by GOOGL’s Android a critical competitive advantage. In this emerging environment, we believe AAPL’s iPhone market share and margins could be seriously pressured.
  • AI opportunities. Most of the rumored initiatives speculated as AAPL’s next big thing – e.g. self-driving cars, augmented reality, web-based consumer services, etc. – have deep learning AI technology at their core. AI allows these products to adapt to real world conditions, to anticipate situations, to assess ambiguous information, to communicate in natural human language and to personalize their responses to individual users. We believe that AAPL is well behind its rivals on all three of the key ingredients of leading edge deep learning – scientific talent, high quality datasets, and AI-tuned data center platforms.
  • AAPL lacks experienced AI talent. Based on the number of academic citations of their scholarly work, AAPL had just five scientists who would rank amongst the top 888 in the world as of July. By comparison, GOOGL had 94 and MSFT 60, and the percentage gap widens slightly when the criterion is loosened to just 1,000 citations. This is very important – deep learning tools are rudimentary and models must be carefully designed and adapted over time by experienced hands to get useful results in a reasonable timeframe. AAPL’s talent deficit is, in part, the result of secrecy requirements that have banned its employees from actively participating in the AI academic community. However, the company recently dropped this prohibition to lure rock star AI prof Russ Salakhutdinov from CMU to head AI research. This change is necessary, but probably not sufficient, to begin closing the talent gap.
  • Privacy policy encumbers AAPL data. In contrast to its chief rivals, AAPL refuses to collect data about its customers’ individual usage. All of this data – locations, messages, schedules, photos, queries, transactions, and other activity – is kept securely on each user’s device, communicated and stored only in fully encrypted form. Its usefulness for AI is limited to small learning programs that might be executed on the device itself, or through a technique called “differential privacy” which wipes away any possibility that data could be tied back to a specific individual before dumping it into a communal database. This seriously handicaps AAPL in personalizing its AI systems for its users interests and circumstances – something that we believe will be vital to the functionality of virtual assistants. It also blocks AAPL from extending a consistent AI presence across multiple use environments.
  • AAPL missed the cloud revolution. In a memo sent a year before his death, Steve Jobs urged AAPL senior management to shift its focus to the cloud, which he saw replacing the device as the hub of its users’ experience. This shift never occurred. Most of AAPL’s cloud applications (e.g. iCloud, Maps, Music, Siri, etc.) are poorly implemented me-too solutions that are widely seen as inferior to their alternatives. AAPL’s data center infrastructure is sub-scale and inadequate to its own needs – it has offloaded much of its cloud traffic to archrival GOOGL’s cloud platform. AAPL’s functional org structure has device-focused executives at the top of every category, relegating the cloud to an afterthought. This is not an ideal development environment for deep learning systems.
  • AAPL culture is an impediment to improving its position in AI. AAPL has a deep-seated culture, inspired by its founder Steve Jobs, which was critical to its dominance in the smartphone era. However, many of the central tenets of that culture are serious obstacles to making up lost ground in the emerging AI era. The primacy of simple, beautiful devices, the closed-garden platform approach, the unitary functional org structure, the commitment to data privacy, the insistence on company secrecy, the focus on internal development and reluctance to make acquisitions, the skepticism toward partnerships – these precepts work to AAPL’s disadvantage as it struggles to adapt to a cloud-based AI dominated future.
  • Expectations are too aggressive. Consensus expects 6.5% topline growth for FY17 and 6% in FY18, with leverage to earnings in both years. Given expectations of a deteriorating smartphone market, new competition on both price and functionality, and AI weakness damping the potential for success in new markets, we believe AAPL is unlikely to achieve sales or EPS expectations, despite the prospects for lower corporate taxes and the repatriation on foreign cash holdings. Moreover, AAPL could be particularly vulnerable to a trade conflict with China and other Asian nations. While the stock rates as “cheap” on a DCF basis, the prospects for missed estimates and downward revisions against a broken growth narrative leave us pessimistic.

What Would Steve Do?

The basic strategy worked for a long time. Establish one simple, powerful and beautiful device as a digital hub, with satellite devices homed to it. Relentlessly iterate integrated software and hardware to deliver a superior and constantly improving experience. Lock in customers with proprietary applications and tight control of the ecosystem. Drive vertical integration under a single functional organization aligned to a fully integrated line of products centered on the digital hub device. Build the best device design and operations teams in the world. Keep innovation in-house and secret to thwart copiers and delight customers with surprise. Eschew large acquisitions that might affect the culture and partner only when AAPL can control the terms.

Seamlessly shifting from the Mac as the digital hub, to the iPhone, AAPL executed flawlessly – creating the smartphone market, dominating the premium segment, delighting users and them locking them in with beautiful, tightly integrated devices and well managed ecosystem of apps and peripherals. However, Steve Jobs saw the sea change coming. In an email to senior management written a year before his death, he declared 2011 “The Year of the Cloud” and warned that AAPL was in danger of hanging on to the old paradigm too long. Until recently, AAPL was still hanging on.

AAPL exquisite device excellence has not translated to the cloud. In its culture, the data center scientists and cloud app developers did not sit at the cool kids table with Sir Jony Ivie and his device designers. Data centers were a necessary evil, and unlike rivals GOOGL, MSFT and AMZN, AAPL built them from off-the-shelf technology from the usual enterprise IT vendors. The AAPL cloud is sub-scale, high cost and low performance, leading the company to reluctantly move some of its most critical cloud applications to commercial hosts – first to AMZN and MSFT, and then to GOOGL. The cloud apps themselves – SIRI, iCloud, Maps, Photos, Music, et al. – were foisted onto the loyal base but roundly bashed by critics, who found superior and cheaper alternatives amongst 3rd party offerings.

The next big thing for consumer electronics is AI. Virtual assistants will obviate the app model, giving users answers, executing requests and anticipating needs consistently across all of a user’s devices. Augmented reality will interpolate cloud based resources and the real world. Automated vehicles will revolutionize transportation. All of the ways the humans interact with information will be improved, in many cases dramatically, and new services will be enabled by AI. AAPL sees this – the new business initiatives most often rumored (cars, AR glasses, wearables, smart media, etc) all depend upon strong AI.

Unfortunately, AAPL will struggle to build strong AI. Its roster of scientific talent is barren in relation to its chief rivals. Its commitment to data privacy hamstrings it in using AI to better personalize its products or enter new markets. It will need to rely upon its competitors’ data centers to build and train its models. Of late, AAPL seems to be trying to catch up. It hired a big-name scientist to head its AI research, loosening its secrecy rules to help him recruit a team. It has invested in a technology called “differential privacy” which would let it gain some insight from its customer data even if it would still be unable to personalize its services or maintain their consistency from device to device. It has plans to step up investment in its datacenter infrastructure. We fear that this may be too little, too late.

The Tao of Steve

The iPhone was supposed to be a peripheral. After his return to Apple in 1997, Steve Jobs centered the company’s strategy around the idea of the Mac as a “Digital Hub” for all a user’s digital assets – at first, calendars, contacts, bookmarks and documents, but eventually, music files, phots, and videos. The Mac would be powerful (as well as beautiful and intuitive), freeing other devices, like the iPod, to be simple (as well as beautiful and intuitive). When the iPhone was introduced in 2007, it was designed to rely upon a PC (preferably a Mac) which handled app downloads and managed all the digital assets. However, in his genius, Jobs quickly figured out that the “Digital Hub” needed to shift from the desk top and into the pocket. Wireless networks were fast enough and available enough, CPU power and storage was cheap enough and small enough, that an iPhone could stand alone, without a PC big brother. This took the addressable market from everyone who regularly used a PC, to everyone who could afford an iPhone, and the rest is history (Exhibit 1).

Exhibit 1: Quarterly iPhone Sales, 1FQ11 – 4FQ16

That history of inventing and dominating the modern smartphone market rests on towering strength in device design and business chain execution. Following the idea of a single digital hub that served as the control point for a person’s digital life, Apple ran itself as a single functional organization with all products and services answering to the same senior executives. There were no conflicts between the “Mac” people and the “iPhone” people, because they were, essentially, the same people. The same is true between the engineering groups working on the hardware and software involved in a product – in other organizations, the OS might be developed separately or by a 3rd party, at Apple, the development was tightly integrated, allowing puts and takes to be examined explicitly and products engineered holistically to the benefit of the user. Important tasks, such as the design of the system CPU chip, were taken in house, if at all possible, to best serve the needs of the integrated design approach. This has been a huge advantage for the iPhone (Exhibit 2).

Exh 2: iPhone 7 Bill of Materials

Apple also vertically integrated into other aspects of its business chain. While it still relies on manufacturing partners, it designed their processes and, in many cases, bought their machines – all to assure the precision assumed in its designs. It aggressively rolled out its own innovative retail stores, which, even after several years of expansion, dominate all other chain retailers in all categories for sales per square foot. Within this approach is an implicit aversion to relying on other companies – when Apple partners, it does so with full control over the relationship. This control is particularly important regarding customers – these are APPLE customers and the company guards them jealously, taking a considerable toll from 3rd parties looking to reach end users.

The Apple strategy is embedded in the Apple culture. Device design and engineering is the highest caste, manufacturing and marketing are just below. Innovation comes from within. Small acquisitions are made to add skills and incremental technologies; large acquisitions, which could weaken the culture and distract focus within the unitary structure, are to be avoided. Secrecy, even across lower levels within the organization, is paramount, lest competitors copy or customers get blasé about new ideas before they reach the market. This is religion at Apple and a clear reflection of the philosophy of its founder.

Steve Leaves a Clue

In October 2010, about a year before his death, Steve Jobs composed a memo to his chief lieutenants outlining a strategy presentation to be made to an upcoming “Top 100” employee meeting. In it, Jobs proclaimed 2011 “The Year of the Cloud” pronouncing that the “digital hub (center of our universe) was moving from the PC to the cloud” and that “Apple was in danger of holding on to [its] old paradigm too long (innovators dilemma)”. Jobs noted that hated rival Google was ahead in cloud services and in integrating those services into Android, exhorting management to “catch up”, “leap frog” and “further lock customers into our ecosystem” (Exhibit 3).

Exh 3: Excerpts from Steve Jobs’ October 2010 email covering Apple FY 2011 plans


In another world, perhaps Jobs might have executed a mighty pivot to the cloud. In this one, Apple missed it. In the Apple system, the internet was a source of downloads and a back-up plan – applications and files were kept and executed locally. Meanwhile, Google had been forced to reinvent datacenter architecture to accommodate its internet search core business, contributing its findings to the open source community. Microsoft, with a search business of its own, had followed Google’s lead. Amazon, Facebook and even Yahoo had also rearchitected their datacenters based on the Google template. Apple hadn’t bothered.

Cloud operations – data centers and internet-based applications – were not the cool kids in Cupertino, stuffed down the functional organization as support to the device-focused business plans of the executives at the top of the pyramid. The infrastructure was FAR from world class – subscale, high cost and unsophisticated compared to rivals. The applications – iTunes, SIRI, iCloud, Apple Maps, Apple Photos, et al. – were only good enough to keep users from fleeing the ecosystem. 3rd party solutions, provided through the App Store with a tithe paid to Apple, were widely seen as superior. Apple couldn’t attract and retain the best people to build a cloud business, because it was clear to all that these areas were not valued within the culture. Of late, CEO Tim Cook has elevated internet software and services under trusted lieutenant Eddie Cue, but the leadership of the data center infrastructure are nowhere to be found on the org chart (Exhibit 4).

Exh 4: Apple Senior Executive Org Chart – November 2016

In this context, Apple has been forced to host its most critical cloud applications on its rivals commercial hosting operations. It must be a bitter pill to pay Google to carry iCloud on its platform. Apple vows to catch up, but even if it had been willing to pursue a transformative acquisition, it would seem to be far too late. We believe that only Amazon, Google, Microsoft, Facebook, Baidu, Tencent, Alibaba, and, maybe, IBM are big enough and sophisticated enough to compete in global hyperscale data processing. Apple is far behind any of these.

Peak iPhone

So far, Apple’s cloud weakness hasn’t really hurt them, because its devices were so dominant. Apple invented the smartphone as we know it, and it continues to claim a 12.5% share of total industry unit sales, 40.5% share of total industry value, and more than 100% of the profit earned by smartphone makers (Exhibit 5). This accounts for the vast majority of the company’s $65.8B in annual cash flow from operations, which in turn fuels its market leading $600B market capitalization. Casual observers note Apple’s well below market cash-adjusted 13.4x trailing P/E ratio against what has been prodigious growth over the past 5 years and scratch their heads. Of course, the controversy – can Apple grow its iPhone sales going forward? We believe the answer is no.

Exh 5: Top 5 Smartphone Vendors and Share, 3Q16

Only a year ago, Apple had posted astonishing 36% sales growth for iPhones in its FY15, driven by huge demand for its first large screen models (Exhibit 6). At the time, we noted that the sales figures for the iPhone6 clearly had pulled forward substantial early upgrade sales from future periods. FY16 offered the proof – iPhone sales were down 12.5% YoY. Bulls point out that if the two years were rebalanced to account for the sales pulled forward in FY15, the average annual growth rate might be closer to 10%. Consensus expects 6.5% company topline growth in FY17, driven by assumptions of a robust upgrade cycle 2 years following the iPhone 6, and another 6% growth in FY18, driven by expectations of a radical redesign on the 10th year anniversary of the original iPhone – all with stable margins (Exhibit 7).

Exh 6: Quarterly iPhone Revenue and Unit Sales, 1FQ14 – 4FQ16

Exh 7: Financial Snapshot: Apple

Exh 8: Gartner Premium Smartphone Forecast Shipments, 2014-19

We see that scenario as plausible but not probable. First, the overall smartphone market is decelerating quickly – Gartner has reduced its projections for 2016 unit sales from 7.0% YoY growth to 4.5% growth, with just 3.5% now projected for 2017, and we believe that these projections are likely quite optimistic (Exhibit 8). Smartphone populations in developed markets – U.S, Western Europe, Japan, Korea, etc. – have flatlined, while growth in big developing markets has fallen faster than expected (Exhibit 9). Moreover, the remaining market growth will come entirely from low priced segments that are not addressed by Apple – the premium market is largely saturated. This is exacerbated by the shift by wireless carriers to eliminate subsidies, giving subscribers clear incentive to keep their older phones rather than upgrading them. We expect global smartphone unit growth to average less than 6% per year through the end of the decade, with all of the growth coming from sub-premium models with ASPs below $200.

Exh 9: SSR Global Smartphone Forecast, 2015-19

This raises a second key point. For years, consumers have been eager to upgrade, valuing larger, more pixel-dense displays, faster processors, better cameras, and slimmer form factors. However, we believe that the performance of flagship devices has entered the range of seriously diminished returns. For most users, the differences between the top models on the market are subtle if not indistinguishable, and the performance of lower price models may be much more that sufficient (Exhibits 10-11). With this, the smartphone market is canting toward commoditization – perhaps an echo of the PC market many years ago. This is obviously bad news for Apple, which demands a substantial price premium, even within the premium segment. We believe that the downward trend in smartphone ASPs will pick up steam. Apple will have the choice of bending to the market and accepting lower prices, and thus margins, or holding the line and likely losing share from its installed base. Neither option is attractive to shareholders.

Now consider a third pressure on Apple. Android smartphones already compete with Apple devices using integrated Google cloud services – e.g. Maps, Google Now, Gmail, calendar, etc. – offering a marginally better experience than these same services provided as apps on an iPhone. With Google’s AI virtual assistant, these services will become utilities within a unified interface on Android products, as will the apps provided by third parties. Users will not ask to open an app, but rather, will ask the AI to perform a task and the AI will determine the service with which to best complete it. In many cases, the AI will anticipate likely tasks and complete them before the request is given – say, checking a flight or curating a morning news feed. Future smartphones will compete on their ability to serve as a vessel for these powerful assistants. We are not sure that Siri is up to the task.

Exh 10: Profiles of Flagship Smartphone Offerings, 2H 2016

Exh 11: Profiles of Potential Flagship Killers, 2H 2016

Talent – Is it Too Late to Catch Up?

Leading edge AI – the kind that is needed for a virtual assistant that can communicate in colloquial human language, interpret ambiguous commands, and anticipate needs – requires three main ingredients: Talent, Data and Infrastructure. All three are challenges for Apple, relative to its obvious rivals, but talent may be the biggest issue.

Deep learning is still an immature science. Until recently, the technology was considered a dead end, requiring huge datasets and powerful computing platforms that seemed out of reach. With the rise of hyperscale datacenters and tsunamis of data from the mobile internet, these prerequisites are no longer unreasonable. Still, that skepticism toward AI dampened the flow of bright minds to the field. Now that deep learning is white hot, there are not enough experienced scholars to train the students looking to study the subject. These scholars are also in high demand from commercial enterprises – at the leading edge of deep learning, experience is crucial. Development tools are still rudimentary – roughly analogous to traditional programming before the advent of compiled languages. A deep learning model may contain many thousands of bespoke algorithms with custom feedback mechanisms programmed into layers of increasing integration. The original design is then tweaked over time to coax the model toward ever more optimal performance. This process demands seasoned experts who have learned from the experience of having built many models before.

As of six months ago, Apple had just three scientists – Andrew Watson, Rudolph van der Merwe, and Gunnar Evermann – amongst the top 859 most cited scholars in AI in May. Policies that restricted Apple employees from publishing research papers, presenting at academic conferences, engaging in joint projects outside the company, or holding part time teaching positions were deal killers for much of the scientific community. AI experts who had come aboard as part of acquisitions routinely exited the company when possible. However, this may be changing – Apple bought an AI startup called Turi in August helmed by a highly regarded professor from the University of Washington, Carlos Guestrin, who, ironically, remains the “Amazon Professor of Machine Learning” at the school. Just a few weeks ago, Apple made another move, hiring rock star AI prof Russ Salakhutdinov from Carnegie Mellon University. Salakhutdinov is a protégé of legendary University of Toronto and Google deep learning guru Geoffrey Hinton, and has amassed more than 16,000 citations since earning his Ph.D. 10 years ago. He, too, will retain his professorship while leading Apple’s AI research lab.

Salakhutdinov and Guestrin have already put out the recruiting call to the AI academic community, signaling a significant change of heart around secrecy – at least regarding deep learning. Still, this may be too little too late. Google has a hall of fame roster of AI superstars – over 90 from the 5K citations category, 250 with more than 1,000, and 2,500 engineers competent in the techniques of AI programming. Microsoft has an AI effort about 2/3rds the size of Google’s. Neither company is standing still – Google just lured iconic Stanford professor Fei-Fei Li and Jia Li, the head of research for Snap, the parent company of Snapchat, to expand their AI cloud platform offerings. These scientists are every bit as impressive as Apple’s big hires, highlighting the magnitude of the challenge ahead of Tim Cook, as he strives for the threshold of “good enough” in his AI engineering organization (Exhibit 12).

Exh 12: Artificial Intelligence Scientists by Organization and Number of Citations, Major TMT Companies

Data – Self Inflicted Wounds

Apple could have as much useful consumer data as anyone. Even more so than Google, Apple tightly controls every aspect of its users’ experiences, and were it so inclined, it would know exactly where they were, what they were doing, and, in many cases, with whom they were doing it every minute of every day. However, long ago, Steve Jobs made the decision NOT to collect any of this data, both because he valued privacy, but also because he saw little value in profiling his customers for anything beyond targeting advertising. This perspective is now well ingrained in the Apple culture, and contributed to the company’s general complacency around the cloud. In the Apple ecosystem, as much of the data as possible stays on the customer’s own device, when it is sent to the cloud, it is sent as an encrypted file readable only by the author and by authorized recipients. In most cases, Apple cannot examine the actual contents of user files – e.g. messages, photos, calendar entries, search inquiries, transactions, navigation records, etc. – even if they are stored to iCloud or used in conjunction with other Apple services (Exhibit 13).

Exh 13: Consumer Data Collection and Use by Apple

This hamstrings Apple in developing AI systems to provide more intuitive and personalized services to its customers. In the past few years, Apple has invested in a data technology called Differential Privacy, which completely masks user data, such that it is impossible to tie back a data record to its source. This is accomplished by transforming the data with one way algorithms that cannot be reversed, even with a key. The resulting data can be used to train an AI, which could provide insights to an entire population, but not to an individual user. This is helpful in training general purpose systems, such as basic voice recognition or picture labeling. However, any individual insights or personalization would have to be trained on the device itself on the data available locally, data which is flushed regularly to save storage space. This also means that the AI would not follow a user from device to device – usage on a Mac would not inform the AI on an iPhone or vice versa.

Compared to Google – which can use a long history of search, messaging, location, media consumption, calendar records, photo and video archives, transactions, and other datasets gathered from any device that a user might use to log in to Google services – Apple is at a massive disadvantage. We believe that this more than overwhelms any marketing advantage the company may derive from its high-horse position on privacy, and believe that Apple would be well served by greatly loosening its stance on customer data.

Platform – Depending Upon the Kindness of Frenemies

As we noted, Apple’s device focused culture has made it difficult to develop its position in the cloud. Its five main datacenters were built from commercially available servers, storage systems and networking gear, and run industry standard software, and require climate control systems. In contrast, Amazon, Microsoft, Google and Facebook run state-of-the-art hyperscale datacenters built from standard components bought in bulk and configured into proprietary system designs, running custom software built in house, and operating at ambient temperatures. This sets Apple at a huge cost and performance disadvantage. Indeed, its datacenter problems are so vexing that it made the hard choice to outsource most of its cloud offerings – including iCloud, iTunes, the AppStore, Maps and other services – to AWS, Microsoft Azure and, most recently, Google Cloud Platform. It is believed that Apple is spending more than $1B annually for hosting these services. While reports suggest that Apple will invest to build three additional new data centers, perhaps reducing their reliance on outside hosts, we are skeptical that these facilities will be anywhere near the state-of-the-art.

Exh 14: Basic On-Premise versus Cloud Cost Comparison

It may be that Apple has missed its opportunity to be truly self-sufficient in its datacenter platforms. We have noted that the big three commercial platforms – AWS, Azure and Google Cloud – have leveraged scale and expertise from their consumer cloud franchises, to gain substantial sophistication, performance and cost advantages over ALL other datacenter operators. ( Those cost differences may mean as much as a 10:1 lead vs. the average enterprise operation (Exhibit 14). Not only is Apple behind the 8-ball on infrastructure, but, essentially, everyone but Amazon, Microsoft, Google and Facebook is in a similar position. Even if Apple were inclined to acquire their way to datacenter excellence, there is no obvious candidate. We do not expect Apple to be able to “invest away” their datacenter problems.

Exh 15: Deep Learning Services Offered by the Big 4

This is a serious problem for deep learning development. At the leading edge, machine learning systems are running millions of iterations on massive datasets, a process that requires extraordinary storage and processing capacity. Moreover, the leading players have begun to invest in specialized hardware and software optimized for AI – investing in GPUs, FPGAs, and ASICs, and building programming libraries to speed system development (Exhibit 15). It is very unlikely that Apple’s new datacenters will feature such AI specialization, and, once again, it will have a make or buy decision. Furthermore, due to the privacy policy, AIs to personalize aspects of Apple’s services will need “on the job” training right on the device, while alternative products build in customer knowledge on the cloud. While iPhone 7 teardowns show an FPGA in addition to the expected GPU, both of which would accelerate on-device training, the compute power is insignificant compared to a cloud hosted solution. This could be Apple’s Achilles heel, if, as we expect, virtual assistants become the primary interface for personal devices.

One More Thing

In the decade between 2001 and 2010, Apple introduced three paradigm-changing products – the iPod (2001), the iPhone (2007) and the iPad (2010). Since then, anxious Apple watchers have perseverated over the next big thing from Cupertino. First it was the TV hinted at in Walter Isaacson’s posthumous biography of Steve Jobs and fed by reports of prototype flat screens being built in China. After this amounted to nothing more than an update to the AppleTV streaming video box, that has, if anything, lost ground to the Google Chromecast and Amazon FireTV in the time since. The next, next big thing was the Apple Watch, launched to great fanfare in April 2015, but sales are something of a yawn, unworthy of being broken out from the other category which also includes AppleTV and trending down sharply YoY. The Apple Car, or Project Titan as it was called internally, was the big rumor of 2015 with many analysts penciling in real revenues for 2019 and conjecturing an acquisition of F1 legend McLaren, but that project and its ambitions were significantly cut back a few months ago. Now word comes that Apple is looking to augmented reality, and a new take on the Google Glasses that failed to win over the world two years ago.

Of these rumored products, only the Apple Watch has come to market in the form that had been conjectured, and the success of the Watch is a matter of perspective – compared to the massive iPhone, the large iPad or even the last reported sales of the iPod in 2014, the Watch is a small product for Apple. Bulls would point out that the best may be yet to come for all these initiatives, but we also note that future success in wearables, self-driving cars and augmented reality will require leading edge AI that remains a substantial deficit for Apple. Wearables, even more so than future smartphones, will depend on virtual assistants as their primary interface – we believe that Siri is badly disadvantaged vs. Google Assistant, Microsoft Cortana, and Amazon Alexa in that competition. Autonomous driving is from the start a deep learning problem of the highest order – Apple has not even begun gathering testing data, leaving it more than eight years and nearly 3.5 million miles behind Alphabet, even if it had the AI development chops to catch up, which it doesn’t. Augmented reality – interpolating computer generated images into the context of a person’s field of view – is an enormous AI challenge (much more so than for fully immersive virtual reality systems). Microsoft may lead this field with the HoloLens prototypes they demonstrated two years ago, and Alphabet has a substantial equity stake in hot startup Magic Leap, which is also perceived as a technology leader. As usual, Apple’s plans are shrouded in secrecy, but it has not made the sort of acquisitions or hires that would be expected if it were well into plans for an augmented reality device.

What About Services?

It may be that the run of hero products that grew from Steve Jobs’ vision of a digital hub is done. A new product would have to have the potential to be a $50B+ business to move the needle at a company with annual revenues of $215B – an implausible goal for most of the ideas being discussed. Lately, Apple management has touted services as the key to its future growth, an idea that passes the sniff test on long term potential. Internet based services has spawned enormous companies – the FANGs (Facebook, Amazon, Netflix and Google) are the obvious examples. Apple, which can count more than a billion total devices in operation by likely 600M+ discrete users, has a massive base of loyal and comparatively wealthy customers under its control. In the most recent quarter, sales of services to these users grew at a healthy 24% clip. Surely it can continue to drive growth by providing valuable services to this customer base (Exhibit 16).

Exh 16: Apple Services Revenue Detail

The two largest components of Apple’s service revenues are the its Apple Care device insurance and repair program and the App Store, together likely comprising the large majority of the category sales. The growth in both of these elements has followed the expansion of Apple’s iOS device installed base. While the number of new iPhones sold has declined, the total number of active iOS users has grown as the length of time users keep their devices increases and as iPhones pass into the used device market. However, this installed base growth will slow on a lagged basis following the sharply decelerating cyclical patterns of new iOS device sales. Apple Care revenues have been juiced by a significant recent one time hike in rates. App revenues – Apple takes a 30% cut of app sales and media sales, along with a cut of subscription revenue (30% for the first year and 15% thereafter), and 30% on advertising and in-app purchases from apps downloaded to iOS devices – face further pressure. The love affair with apps may be over – by some reports, app downloads by the top publishers are down as much as 20% in the US this year, as users are concentrating their time with a small number of their favorite apps and rarely downloading new ones (Exhibit 17). International markets are better, but the trends are directionally similar.

Exh 17: Share of Time Spent on Apps by Rank

Furthermore, we believe that the increasing functionality of AI-powered virtual assistants will reduce user reliance on apps. While Apple may generate new revenue by service referrals through Siri, the monetization model has yet to be developed. Apple can also generate revenue through its own cloud-based services, of which, Apple Music is the most prominent. Here, the dominant position of iTunes (still 10-15% of Apple’s service mix and falling rapidly) built during the heyday of the iPod and downloaded digital music files was a strategic impediment that kept Apple from offering a streaming music service until last year, giving Spotify, Pandora and others a 4-5-year head start. Despite the huge advantage of those hundreds of millions of iOS devices in its tightly managed walled garden, Apple Music has been disappointing – still a fraction of the size of Spotify and plagued by less than enthusiastic reviews of the service. The same dynamic is at work in other Apple for-pay services, like iCloud, where 3rd party alternative hold significant share and are generally rated higher by users and critics.

Increasingly, we expect future mobile services to be AI-based – using data on consumers in general and from users specifically to anticipate needs and optimally respond to requests – whether from the virtual assistant UI or within individual apps. This is clearly a problem for Apple’s services strategy. Moreover, its unwillingness to use its customers’ usage patterns to target advertising cuts off a major source of monetization, although it benefits obliquely by taking its share of targeted ads sold by 3rd party apps without such restrictions.

The Hard Path Forward

With an installed base of over a billion devices in the hands of historically loyal and well-off customers used to paying a substantial premium for the Apple experience, and clear leadership in the design, manufacturing, marketing and distribution of consumer electronics, Tim Cook and company have valuable assets, even if the battle is moving away from its strengths (Exhibit 18). Even with share loss and margin compression in a mature smartphone market, Apple still will generate extraordinary cash flows, to add to its already gargantuan cash holdings. With the prospect of new tax rules to dampen the impact of repatriating cash held overseas, Apple has the resources to make a big move to reposition itself for the cloud-AI era.

Exh 18: Number of Monthly Active Users by Property

While the entrenched Apple culture is strongly opposed to transformative acquisitions, a bold deal could jar the organization forward. Still, this late in the game, the leaders in the cloud-AI game – Alphabet, Microsoft, Amazon and Facebook – are out of the question, even with Apple’s resources. Some companies further down the list in terms of AI talent, like Netflix, Snapchat or even Twitter, could bring an influx of necessary cloud savvy to Apple’s device dominated organization, but would come with baggage and would require very careful integration. We are very skeptical that this management team would take such a bold step.

Building AI and cloud capabilities from the ground up will be harder than many suppose. Perhaps Uber is the best model for how to do this – it swooped in and poached a 40-person robotics department from Carnegie Mellon University and set them up in a spanking new research facility down the road from CMU in Pittsburgh. This team had been working together on self-driving technology for many years, winning DARPA sponsored autonomous vehicle challenges in 2004 and 2007. Apple could take a page from Uber and swoop in on a well-respected deep learning team at a university. Of course, the top AI companies have had the same idea. Google has major AI research clusters near top universities (University of Toronto, University of Montreal, Cambridge/Oxford, Stanford, etc.) with professors splitting time between academic and commercial pursuits (Exhibit 19). Likewise, Microsoft has major joint research centers in Washington, China, Switzerland, and other sites. Facebook is tightly partnered with NYU and has a major AI research center in Paris with participation from French Universities. With Apple’s recent move to lure Dr. Russ Salakhudonov from CMU, we expect to see Apple MUCH more aggressive on this front.

Exh 19: Academic Centers of Excellence in Artificial Intelligence

Time is even shorter for Apple to gain ground in the cloud. The infrastructure game may already be lost – the scale and sophistication advantages are so large in the hyperscale datacenter market, that we believe Alphabet, Microsoft, Amazon and Facebook have established insurmountable leads over the rest of the field. Apple may have to make its peace with being a long term partner with one of these players for its platform and focusing on developing its cloud application capabilities. Again, we expect the most effective way forward would be M&A but do not expect Apple to pursue it.

Microsoft Redux

A few years back, we posited that Apple’s struggles to cope with the emerging cloud era were roughly analogous to Microsoft’s well publicized failure to address the mobile revolution. Under Steve Ballmer, Microsoft delivered modest growth and huge cash flows, but failed to establish a winning tack as the wind shifted from the desktop to the smartphone. Microsoft was a “cheap” stock for nearly a decade, regarded by investors as a relic with a dim future. It took a regime change and an aggressive move to break up the waterlogged organization and realign to a cloud-AI strategy to shift to a winning narrative.

A few more voices have joined the chorus on Apple. We believe that Tim Cook’s current strategy has Apple on a trajectory that will see moribund sales growth and slow erosion of the iPhone driven margins. While on a purely mechanical basis, Apple rates as a “cheap” stock (our valuation suggests a DCF based fair value of about $123), we do not believe that investors will be satisfied until the company can re-establish a plausible growth narrative (Exhibit 19). We are not certain that Apple’s culture would support a major realignment without much greater and more obvious market pressures. As such, we believe that the company’s business and stock market fortunes will have to get worse before they can get better.

Consensus expectations project 6.5% sales growth for Apple in its FY17 and 6% in FY18 (Exhibit 20). We see this as wishful thinking as the global premium smartphone segment stagnates on market saturation and lengthening replacement cycles. We do not expect new product categories to contribute meaningfully to Apple revenues and we project services growth to decelerate as well. Consensus also expects leverage to the bottom line, more likely the result of share repurchases than margin expansion, but we also see this as optimistic. Low cost, high spec smartphones will pressure prices for premium smartphones, while Google’s AI virtual assistant will give flagship Android models a clear competitive advantage over the iPhone to boot. While Apple rates as cheap on many value models, and shareholders may look forward to the repatriation of those massive overseas cash holdings, the market has not been kind in its history to fallen growth companies with deteriorating cash flow streams. This was true of Ballmer era Microsoft, and, we fear, it will be true of Cook era Apple as well.

Exh 19: Apple DCF Model

Exh 20: Apple Consensus Expectations and Financial Snapshot

©2016, SSR LLC, 225 High Ridge Road, Stamford, CT 06905. All rights reserved. The information contained in this report has been obtained from sources believed to be reliable, and its accuracy and completeness is not guaranteed. No representation or warranty, express or implied, is made as to the fairness, accuracy, completeness or correctness of the information and opinions contained herein.  The views and other information provided are subject to change without notice.  This report is issued without regard to the specific investment objectives, financial situation or particular needs of any specific recipient and is not construed as a solicitation or an offer to buy or sell any securities or related financial instruments. Past performance is not necessarily a guide to future results.

Print Friendly, PDF & Email