AI for Edge Computing: Lots of Chipsets

The edge computing market has many segments, including onboard device computing as well as premises and off-premises computing. 

Looking only at consumer and enterprise devices that compute onboard, Deloitte suggests 

2020 sales of more than 750 million edge artificial intelligence chips that support machine learning tasks directly on the device, rather than in a remote data center.

Smartphones will be a huge part of the use case for AI chips at the edge, with other consumer devices such as wearables, smart speakers and tablets consuming 20 percent of those chipsets. 

source: Deloitte

Not every segment of the edge computing ecosystem is directly concerned with AI chipset propagation, though. Mobile operators care about the relationship to mobile connectivity services, especially 5G, as well as opportunities to participate in premises or off-premises edge computing. Low power wide area networks also have opportunities for local access. 

And many suppliers of local connectivity infrastructure also see key roles for premises wireless, including Wi-Fi and other short-range networks. 

So virtually nobody believes that every internet of things use case “must” use 5G. That does not obviate the use case for 5G and ultra-low-latency connections, but there will be other connectivity options on the premises or for local access. 

AI edge-related chipset sales might represent $2.6 billion in revenue, with a compound annual growth rate of 36 percent, says Deloitte. 

“By 2024, we expect sales of edge AI chips to exceed 1.5 billion,” Deloitte sasy. That represents annual unit sales growth of at least 20 percent, more than double the longer-term forecast of nine percent CAGR for the overall semiconductor industry, according to Deloitte. 

Analysts at ABI Research concur with at least parts of the Deloitte forecast. Revenue from cloud-edge artificial intelligence chipsets are set to grow from $2.6 billion in 2020 to $12 billion in 2025, at a compound annual growth rate (CAGR) of 36 percent, ABI Research predicts. 

Many other forecasts might suggest similar growth rates for other parts of the ecosystem.. Grand View Research suggests 2027 edge computing revenues of perhaps $21 billion, going to suppliers in different parts of the ecosystem, ranging from chipsets to network infrastructure and software to services of many types. 

If the Deloitte and ABI Research assumptions are correct, they suggest strong sales of smartphones and wearables. 

The consumer edge AI chip market is much larger than the enterprise market, but it is likely to grow more slowly, with a CAGR of 18 percent expected between 2020 and 2024, say Deloitte analysts. 

The enterprise edge AI chip market is growing much faster, with a predicted CAGR of 50 percent over the same time frame, Deloitte maintains. 

source: Grand View Research

AWS Outposts Now Available in Many New Countries

AWS Outposts--the AWS managed service for premises computing running AWS services--can now be shipped and installed at customer data centers and on-premises locations in Brazil, India, Israel, Malaysia, Mexico, New Zealand, South Africa, Taiwan, and Thailand, and supported in three additional Regions: AWS Africa (Cape Town) Region, Asia Pacific (Mumbai) Region, and South America (São Paulo) Region.  

AWS Outposts enables native AWS services, infrastructure, and operating models at “virtually any data center, co-location space, or on-premises facility,” AWS says. “You can use the same APIs, the same tools, the same hardware, and the same functionality across on-premises and the cloud to deliver a truly consistent hybrid experience.”

Vapor IO, Digital Realty Launch Edge Computing Solution

 Vapor IO and Digital Realty have launched a new core-to-edge solution in Atlanta, Chicago and Dallas. 

The solution connects Digital Realty’s PlatformDIGITAL with Vapor IO’s Kinetic Edge services, making it easier for joint customers to deliver low-latency applications.  

The companies are now working on initial proof of concept workloads that will operate seamlessly together, whether deployed near the access edge on the Kinetic Edge platform or at the regional edge in Digital Realty’s centers of data exchange.

AWS Adds New Snowcone Edge Device

The edge computing “market” has several key and different segments, including edge computing infrastructure, enterprise software, “as a service” computing services, devices and some amount of real estate and communications activity. Hyperscale computing as a service suppliers are emerging in key and leading roles. 

Amazon Web Services has developed AWS Snowcone, a small, ultra-portable, rugged, and secure edge computing and data transfer device. 

Source: AWS

At under 5 lbs.  and able to fit in a standard mailbox or a small backpack, AWS Snowcone is the smallest member of the AWS Snow Family of devices, enabling customers to collect data, process it locally, and move it to AWS either offline (by shipping the device to AWS) or online (by using AWS DataSync to send the data to AWS over the network). 

“AWS Snowcone is built to withstand harsh conditions and is designed for a variety of use cases in environments outside of the traditional data center that lack consistent network connectivity and/or require portability, including healthcare, industrial IoT, drones, tactical edge computing, content distribution, data migration, video content creation, and transportation,” AWS says

Retail Co-location a Bigger Opportunity for Mobile Operators Than Wholesale?

With the caveat that cloud computing and edge computing are different segments of the computing market, retail co-location might be a bigger opportunity than wholesale co-location. Retail co-location, where a tenant leases rack space or a cage, might seem the obvious value connectivity providers can offer in the edge computing space.

source: Synergy Research Group

It seems relatively unlikely that wholesale co-location, where a third party leases the entire facility, is the most-logical opportunity for connectivity providers. 

Still, hyperscale computing firms are the fastest growing customer category for co-location providers, Synergy Research Group says. For both wholesale and retail co-location, first quarter 2020 revenue from hyperscale customers grew more rapidly than revenues from other service provider customers or enterprises. 

While the overall co-location market grew by seven percent, compared with 2019, revenues from hyperscale operators grew by 22 percent in the wholesale segment of the market and by nine percent in the retail segment. 

While a wholesale approach might be part of the mobile operator edge computing opportunity, retail seems likely the more prevalent use case.

Public Cloud Spending Grows, Enterprise Data Infrastructure Dips

It will likely not come as a surprise to you that the Covid-19 pandemic has slightly dampened cloud infrastructure sales but not seemingly dented cloud computing spending. 

“Cloud service revenues continue to grow by almost 40 percent per year, enterprise SaaS revenues are growing by almost 25 percent, search/social networking revenues are growing by over 15 percent, and e-commerce revenues are growing by over 20 percent, all of which are helping to drive growth and increased spending on public cloud infrastructure,” said John Dinsdale, Synergy Research Group chief analyst. 

source: Synergy Research Group

Global spending on data center hardware and software declined by two percent from the first quarter of 2019, Synergy Research Group says. 

                 

Public cloud infrastructure spending grew three percent, while spending by enterprises and service providers declined by four percent.

source: Synergy Research Group

Survey Finds 34% of Enterprises Already Using 5G

Fully 34 percent of 1,000 surveyed information technology professionals in industry verticals including energy, retail, manufacturing, government and public safety, automotive and transportation, media and advertising, education in the United States and the United Kingdom report that they already are using 5G. 

The study was conducted by Parks Associates and sponsored by Nokia. 

What is not clear is how extensive 5G might be in each industry vertical. The video surveillance use case, for example, is a “point” application relatively easy to deploy, but also limited in impact. 

Apparently none of the respondents taking the structured survey was asked about edge computing or private networks. 

Methodologically, the survey used a guided method. As often is the case, researchers had to pre-define the use cases, and then essentially asked respondents which of the use cases had the most appeal. Respondents did not have the choice of naming the use cases they deemed most important, but had to choose from among the offered cases. 

The findings are helpful and suggestive, to be sure. But if you have ever participated in these sorts of surveys, you know they force fit you into a set of potential answers that might not be the choices you would have made if the question had been asked a different way, or if the question were open ended. 

The reported level of 5G adoption is three times the percentage of 5G users in South Korea, an early mover having reached adoption levels of 10 percent of mobile accounts, which now use 5G. That might seem incongruous, but the survey only assesses whether 5G is used someplace, somehow, by an enterprise. It does not try to assess the ubiquity of 5G use.

The consumer use case is binary: subscribers are either “all in” or “not using.” The enterprise use cases are much more nuanced: enterprises can prototype or pilot, without adopting broadly. They also can use 5G in specific ways, without needing to adopt organization wide. 

The survey finds enterprise information technology professionals widely believe video applications will be an early lead application for 5G, with 83 percent of 1,000 IT professionals naming video detection and alerts as a 5G use case of interest. An equal percentage named video surveillance as a promising use case. 

source: Nokia

Indoor 5G Will Drive Edge Computing, Public 5G Interconnection

A new report by the Small Cell Forum predicts that by 2025 almost 75 percent of new indoor small cell deployments will be colocated with mobile edge computing facilities or private enterprise cell networks that interconnect with public mobile networks. 

The private networks will include a mix of neutral host facilities and enterprise private networks. 

Edge computing solutions are seen as network-based, where the service provider deploys edge computing assets, or premises-based, where an enterprise or venue owner deploys edge computing assets as part of its IT infrastructure. 

SCF logically argues that small and medium-sized organizations might not be able to financially justify a private edge computing capability, and would be “better served by a network-based edge solution.” 

“Edge compute is one of the most important developments for mobile operators (MNOs) and other service providers to consider in their next generation network strategies,” the report argues. “There are significant opportunities to harness edge compute and combine localized data center (processing, storage and analytics) capabilities with connectivity.”

source: Small Cell Forum

But the report also makes clear that “both enterprise customers and mobile network operators may deploy small cell networks as well as edge computing platforms together.” In other words, it is unclear how small cell edge computing will develop. In some cases mobile or other service providers might provide the edge computing. In other cases enterprises will do so themselves. 

Revenue opportunities thus develop for infrastructure providers and service providers. As has been true in other similar situations, opportunities for infrastructure providers and managed service providers are distinct and non-overlapping. 

source: Small Cell Forum

In principle, a connectivity provider can provide colocation (data center racks, power, cooling, security), network-related functions, an edge computing platform for third party tenants, hosted applications or fully managed computing services. Risk, expense and revenue upside vary by the choices made. 

Keep in mind that for a firm such Amazon Web Services, infrastructure as a service means computing cycles on demand. For a mobile operator, infrastructure might mean the data center rack space functions. 

A big strategic issue for mobile operators is how much of the actual computing function can be supplied on a sustainable basis. Will some customers primarily want rack space? Yes. Beyond that, might some customers want to buy  mobile operator computing, storage, platform or app offers? That is much less clear.

Nokia Sees Important Role for 4G Private Manufacturing Networks

Nokia believes private networks--using 4G and 5G-- will be an important driver of its sales in the 5G era, though private 4G is expected to be the platform for industrial private networks. The advantages are likely easiest to envision for various internet of things and edge computing networks with high requirements for latency performance. 

source: Nokia

Network slicing can conceivably extend private networks through the wide area network, when a public network is the platform, or can use a private WAN when available. 

source: Nokia

Who Owns the Edge, Where is the Revenue?

Who will own the edge, and who will pay? Those are the two key questions for mobile or fixed network operators looking for a role in the edge computing ecosystem. So far, despite the rhetoric, most tier-one telcos with edge computing aspirations still are focusing on connectivity revenues or real estate (colocation in telco facilities, for example). 

AT&T is working with Microsoft to create edge locations supporting customers of  Microsoft Azure cloud services and AT&T’s 5G network. AT&T’s upside seems to be colo revenue and incremental connectivity business. 

BT seems to be positioning for access revenues for retail customers, though it might also expect to use edge compute locations to support its own 5G and other services using a more  virtualized network.

Indeed, it is quite possible that the earliest value of edge computing for many tier-one service providers will be as a support for its own operations. 

BT definitely is increasing the number of its metro exchanges to reduce latency, but its description of “edge” is “managed LAN, WLAN and wireless” solutions. That seems more a case of traditional services being recast as “edge” revenue. 

BT says it offers “a fully managed service that securely connects corporate end-users to voice, data and Internet of Things (IoT) applications.” There’s nothing wrong with that, but it is not what many of us would consider “edge computing.”

CenturyLink seems to be generating connectivity revenue at the moment, though some might question the characterization of much of that revenue as “edge” related. 

Telefonica’s “open architecture” likewise seems aimed at internal uses, though external customers will be able to take advantage of the architecture as well. Telefónica says its edge computing platform will enable a “wide community of developers (including Telefónica’s own).”

Telefónica could “deploy a huge range of basebands and radios from different vendors, enabling multi vendor interoperability.”

Telefónica “will benefit from native RAN virtualization, modularity and virtualization.”

Verizon has a partnership with Amazon Web Services offering the “Wavelength” program that collocates AWS computing resources inside Verizon facilities. 

So far, aside from support for virtualized network operations on an internal basis, the two incremental revenue streams seem to be connectivity and collocation. That could change, over time. But right now, infrastructure edge arguably is more a support for core telco operations than a major driver of incremental revenue.

Can Low Earth Orbit Satellite Constellations Play a Role in Edge Computing?

It is not yet clear whether there are edge computing implications. What is clear that some amount of bandwidth will be available over the entire surface of the earth. Whether that is most beneficial for premises computing (onboard ships or on drilling platforms) or distributed computing is not yet clear. In principle, the low latency possible with low earth orbit satellites could make possible distributed computing facilities that approximate the sorts of edge computing 5G networks envision (within 10 km to 100 km). 

CenturyLink Touts Edge Computing Revenue

CenturyLink’s discussion of edge computing opportunities illustrates the complexity of evaluating connectivity provider revenue opportunities in edge computing. Talking about customer wins related to edge computing, Shaun Andrews, CenturyLink CMO says the range fo use cases includes enabling private networks using Citizens Broadband Radio Service, private LTE networks or Wi-Fi 6 networks, generally related to connecting IoT devices on premises. 

Some of us would say those are examples of traditional connectivity sales, not edge computing in a direct sense. Doubtless few would quibble about edge computing driving potentially higher access revenues, to support premises computing. At the same time, many would likely argue that Wi-Fi per se is not edge computing. Local area networks are not automatically edge computing. 

It remains to be seen how big a role in the computing value chain connectivity providers will be able to capture. 

Consumer IoT Sales Dip Because of Pandemic

Consumer internet of things sales, especially wearable devices, have seen a sharp reduction in sales during the Covid-19 pandemic, ABI Research says. Wearables have been affected because consumer interest in buying non-essential devices has dropped in the first quarter of 2020. Supply chain issues also are an issue.

But ABI Research also believes the pandemic dip only postpones demand for wearable devices. And sales of healthcare related wearables arguably have increased.  

Wearable shipments in 2020 are now expected to be 254 million, down from the previously forecasted 281 million. That still represents a  minor increase from the 241 million shipments in 2019. 

IoT Attacks Increase by Orders of Magnitude, But Routers Might Still be Key

By 2025, there will be 41.6 billion connected internet of things devices, generating 79 zettabytes (ZB) of data, according to IDC. That is the sort of forecast that five years ago might have been ridiculed as excessively optimistic. Indeed, some forecasts suggest that could be high by an order of magnitude. 

source: SonicWall

All represent attack surfaces. About 98 percent of all IoT traffic is not encrypted, perhaps driving an increase in attacks of perhaps two orders of magnitude in the first half of 2019 alone. 

Some studies suggest the average IoT device was attacked once every two minutes at times of peak activity. But routers are the target of up to 75 percent of attacks, Symantec data suggests.