1、EQUITY RESEARCH|October 14,2018|9:06PM EDTHeather Bellini,CFA+1 212 357-Goldman Sachs&Co.LLCGoldman Sachs does and seeks to do business with companies covered in its research reports.As aresult,investors should be aware that the firm may have a conflict of interest that could affect theobjectivity o
2、f this report.Investors should consider this report as only a single factor in making theirinvestment decision.For Reg AC certification and other important disclosures,see the DisclosureAppendix,or go to employed by non-US affiliates are notregistered/qualified as research analysts with FINRA in the
3、 U.S.The Goldman Sachs Group,Inc.Ted Lin+1 212 357-Goldman Sachs&Co.LLCMark Grant+1 212 357-Goldman Sachs&Co.LLCCaroline Liu+1 212 357-Goldman Sachs&Co.LLCAs more devices generate more data from more locations,computing is facing a speed-versus-scale challenge.Thepublic cloud remains unrivaled in it
4、s compute and storage resources,but getting data there and back takes time,andultimately is limited by the speed of light and the size of internet“pipes.”In cases where big data will be used to drivereal-time decision-making,we see an opportunity for“edge computing”to become a key enabler and extens
5、ion ofthe public cloud by putting compute and storage resources closer to the device or source of data generation.Edgecomputing could unlock a$40bn incremental market($100bn in the bull scenario),including a range of newapplications that can better direct operationsfrom“when to brake”for a self-driv
6、ing truck to“when to changecourse”for an oil drill working miles underground.We see providers of a cohesive public cloud and edge solution asbest positioned and highlight Microsoft in particular,in addition to Amazon,Pivotal,and VMware.How edge computing will augment the cloud&unlock real-time,big d
7、ata applications The Cutting Edge of ComputingCloud Platforms,Volume 5For the exclusive use of GAOHANG467PINGAN.COM.CNPM summary3Shift to the cloud continues in earnest11But computing is poised to shift back to a decentralized paradigm13What is edge computing?17Edge computing demand drivers22Sizing
8、the potential market opportunity for virtualization and server operating systems28Edge computing vs.cloud computing performance33Killer apps for edge computing44Winners&losers:edge computing could sustain a renaissance in on-premise software49Valuation&risks67Disclosure Appendix6814 October 20182Gol
9、dman SachsCloud Platforms Volume 5Table of ContentsFor the exclusive use of GAOHANG467PINGAN.COM.CNPM summaryWhy is the edge important?While the overarching theme in software will continue to be the centralization ofcompute(i.e.the moving of workloads from on-premises to public cloud),we believethat
10、 computing at the edge will play an increasingly important role,augmenting thecapabilities of public cloud and bringing resources closer to the source of datageneration.In edge computing,data is processed,analyzed,and acted upon at(or closeto)the source of data generation,as opposed to raw data bein
11、g sent directly to a publicor private cloud to be acted upon.To accomplish this,edge computing adds the corebuilding blocks of public cloud including compute,networking,and storage closer tothe origin of the data,allowing insights to be generated and executed in real-time.Incontrast with centrally-l
12、ocated traditional and purpose-built on-premise data centers orprivate clouds,edge servers can be placed far from centralized computing cores in(oraround)factories,airplanes,cars,oil rigs,or in conjunction with cell phone towers.In anedge+cloud world,processing is therefore divided between the edge
13、and the cloud,and fundamentally,our view is that edge computing is complementary to(and not asubstitute for)the public cloud moving all compute to the edge would result indistributed and unmanageable clusters of chaos and forgo the scale benefits of publiccloud.Although public cloud has effectively
14、limitless resources,edge computing has severaladvantages that cannot be effectively matched by the public cloud.For instance,latency(distance to the public cloud)and bandwidth(size of the pipe connected to the publiccloud)remain issues in many instances.For use cases where reaction time is critical
15、tothe success of the overall system,the latency inherent with a round trip to the cloud viaa hub-and-spoke model may be not be acceptable.Latency can be influenced by aplethora of uncontrollable factors,including the network connectivity of the location,thenetwork provider,other network traffic,as w
16、ell as the specific region,availability zone,and data center that the user connects to.Additionally,the speed of compute and dataprocessing has far outclassed network bandwidth.Truly big data use cases will alsocreate massive data generation,orders of magnitude above what could be transmittedback to
17、 the public cloud;in fact,these big data use cases will generate sufficient datathat simply storing it,even with the resources of the public cloud(assuming that thedata can be transmitted there),will be challenging;edge computing will enable the datato be processed immediately,and only relevant data
18、 needs to be sent back to the publiccloud to be stored and further reasoned upon.Dependence on public cloud for all dataprocessing and analytics may not be suitable for many use cases,particularly those thatfeature low or intermittent network connectivity,and we believe that even 5G may notbe adequa
19、te bandwidth for many use cases.Finally,processing the data on the deviceor at edge,versus uploading raw data to the public cloud,can yield superior results forsecurity and privacy,as there are inherent risks in transmission.14 October 20183Goldman SachsCloud Platforms Volume 5For the exclusive use
20、of GAOHANG467PINGAN.COM.CNHow big is this market?In this report,we evaluate the potential incremental infrastructure software spend thatcould be attributed to an increase in edge servers,driven by the need to performprocessing closer to the source of data generation.With 2.72bn IoT endpoints(i.e.the
21、connected“things”themselves)shipments in 2021,we estimate that in the mostconservative scenario,the incremental annual value(i.e.license,maintenance,andsubscription revenue)would be$14bn for virtualization and$7bn for server operatingsystems;in the most aggressive scenario,the incremental annual spe
22、nd would be$69bn for virtualization and$34bn for server operating systems.We note,however,thatthese estimates likely skew conservative,as it does not account for other infrastructuresoftware like NoSQL databases,which could potentially be a lightweight option for edgecomputing;nor does it account fo
23、r analytics and application software,which will dependheavily on the types of use cases leveraged for edge computing resources.We alsobelieve that container adoption could serve as a multiplier for spending,as Red Hat hascommented that OpenShift is“almost 20 x the price of RHEL on the same two-socke
24、tserver.”Finally,we highlight that these forecasts do not include any hardware orincremental storage capacity,just to name a few,that would also be directly impactedby the build out of edge networks.“Killer apps”enabled by the edgeBased on the unique advantages of edge servers relative to public clo
25、ud and small IoTendpoints,we believe that edge computing enables a broad spectrum of use cases thatleverages edge servers ability to perform advanced computational tasks at the sourceof data generation.We believe use cases like autonomous cars/trucks,digital oilfields,and video analytics have the ab
26、ility to revolutionize business processes;however,webelieve that until infrastructure to enable inference at the edge is in place,these marketswill fall short of their full potential.We highlight some potential edge computing usecases below;we note that these use cases are not an exhaustive list:Aut
27、onomous cars&trucks:Real-time processing via an onboard edge server is criticalto the safe operation of an autonomous vehicle,for both the passengers as well as thegeneral public;an autonomous vehicle cannot afford the latency required to access thepublic cloud,as any delays in reaction speed could
28、be potentially catastrophic.For thisuse case,analyzing the data in real-time a task that can only be accomplished by anedge server is critical to maintaining the vehicles safety,efficiency,and performance.AR/VR:Augmented and virtual reality use cases require large amounts of processingpower;however,
29、users are heavily sensitive to latency,precluding AR/VR fromleveraging public cloud given the networking capabilities available today.While we wouldexpect PCs remain the primary mode of compute for the time being,we could see usecases develop for the use of edge servers if this latency can be improv
30、ed over time(i.e.through 5G),particularly where device-level compute is too difficult to achieve in a formfactor that meets the needs of the user.Digital oilfields:Edge computing is slated to play an increasingly vital role in oil and gasexploration,given the remote locations in which the industry o
31、perates.For instance,using real-time processing can help to maximize drills output while minimizing energy14 October 20184Goldman SachsCloud Platforms Volume 5For the exclusive use of GAOHANG467PINGAN.COM.CNconsumption by analyzing drill data in real-time to make instant decisions about thedrills ne
32、xt best course of action.IoT enterprises:As increasing amounts of compute,storage,and analytics capabilitiesare integrated into ever-smaller devices,we expect IoT devices to continue toproliferate,and as noted previously,Gartner expects IoT endpoints to grow at a 33%CAGR through 2021.In cases where
33、reaction time is the raison dtre of the IoT system,the latency associated with sending data to the cloud for processing would eliminatethe value of the system,necessitating processing at the edge;public cloud could still beleveraged where processing is less time sensitive or in instances where the s
34、cale andsophistication of public cloud need to be brought to bear.Public safety(Amber Alerts):Video analytics is an example where bandwidthlimitations,long latency,and privacy concerns converge to favor edge computing overleveraging public cloud.For instance,locating a lost child in a city is one po
35、tentialreal-world application of video analytics where public cloud limitations would preventsuccessful deployment.With an edge computing paradigm,the request to locate themissing child can instead be pushed out to all of the relevant devices:each camerawould perform the search independently using n
36、earby compute resources.If,and onlyif,the camera registers a positive match would it then upload data to the cloud:bydistributing the analytics to the small-but-numerous devices in the edge(where the dataresides),tasks can be quickly and efficiently processed.Reiterating our Buy on MicrosoftWe reite
37、rate our Buy on Microsoft and our$123,12-month price target based on thecompanys strong positioning in all three tiers of public cloud,as well as its emergingleadership in edge computing.We note that our EPS estimates are above FactSetconsensus for CY19 and CY20:One technical analogy often cited for
38、 public cloud is its similarity to a utility.Prior to the1880s and the advent of central power plants,electricity was typically generated on-siteand therefore limited to factories,hotels,and wealthy residences.These generatorswere typically located in the basement,or in close proximity(e.g.a nearby
39、river orwaterfall).However,due to variety of reasons,including scale benefits(i.e.volatility indemand,R&D,purchasing),the ability to shift capital expenditure to operating expenses,and the ability to offload non-core operations,electricity generation quickly moved tocentralized power plants,with con
40、sumers and businesses alike purchasing electricity asa service.Exhibit 1:Our EPS estimates are above consensus in CY19 and CY20GS vs.consensus(in$mn except per share data)CY19(E)CY20(E)CY21(E)GSConsensus%GSConsensus%GSRevenue127,878128,543(0.5%)140,660142,834(1.5%)154,210EPS$4.67$4.533.1%$5.48$5.254
41、.5%$6.42FCF40,67636,77810.6%44,65244,910(0.6%)48,174Source:Goldman Sachs Global Investment Research,FactSet14 October 20185Goldman SachsCloud Platforms Volume 5For the exclusive use of GAOHANG467PINGAN.COM.CNWe believe that cloud computing will follow a similar trajectory,with servers andcomputing p
42、latforms increasingly delivered as a service,due to the same benefits thatexisted for electricity to become delivered as a service:scale,capex-to-opex,andoffloading non-core operations.As such,as public cloud becomes increasingly central toenterprises IT stacks,we believe the key components of serve
43、rs(compute,networking,and storage)will increasingly resemble utilities like electricity and water,whereresources are generated centrally,then delivered and consumed as needed bycustomers.We would caveat,however,that there are important core differences in the comparisonof public cloud business model
44、s and utilities business models.Importantly,utilities are anatural monopoly,and as a result,it is functionally impossible for a company to churn off(as there are no competitors and going off the grid would be clearly infeasible).Forpublic cloud,we would foresee at least three major competitors movin
45、g forward(AWS,Azure,and GCP),and while we continue to believe in the increasing stickiness of theplatforms,particularly as customers adopt PaaS features,it is clearly possible to migrateworkloads from one platform to a competitor(and partners have noted that this indeedoccasionally occurs).Additiona
46、lly,utilities are guaranteed an ROE,and while they mayoverearn or underearn in certain years,they can generally apply to regulators to increaserevenue in the event of underearning.By contrast,public cloud services are determinedby market-clearing rates,and we note that in some instances,services may
47、,in fact,bepriced below cost.As a result,we would expect the ROE of public cloud to continue tobe more volatile than that of utilities.Finally,we note that while Microsoft pays aconsistent dividend(yield of 1.7%),this is approximately half that of the average of theutilities that we evaluated(3.4%).
48、For the major public cloud vendors,revenue derived from supplying these resources istherefore recurring and sticky.Enterprise applications(e.g.enterprise resource planningapplications,customer relationship management systems,human resourcesmanagement systems,specialized industry applications)and dat
49、a are typicallyfundamental to the operation of the business;without this infrastructure,the businessceases to operate effectively.As a result,even in the face of economic headwinds,thespending impact on this core infrastructure will be relatively muted to other areas thatmay be more susceptible to s
50、pending reductions.In the traditional enterprise softwareperpetual license+maintenance model,customers could choose to churn offmaintenance and still retain the usage of the software;this is not possible withsubscription-type models(e.g.public cloud,SaaS),where the churning off the platformmeans tha
