Will TinyML Displace Multi-Access Edge Computing?

New technologies have a way of redesigning older markets out of existence or reshaping older markets. TinyML, for example, is already a possible functional replacement for mobile edge computing. 

By enabling on-device intelligence, TinyML replaces (and limits the size of) the MEC revenue opportunity for mobile operators, as it supports AI and other processing right on the device. For example, if TinyML enabled on-device processing, it could cut MEC revenue in half. 

MEC Scenario	Growth Rate Assumption	Estimated Revenue (2024-2028)
Low	Slow adoption of MEC, primarily driven by enterprise use cases with limited consumer uptake.	$10-20 billion
Moderate	Moderate adoption across both enterprise and consumer sectors, with MEC-enabled services like augmented reality and connected vehicles gaining traction.	$30-50 billion
High	Rapid adoption fuelled by widespread deployment of 5G and increased integration of MEC into essential services like healthcare and smart cities.	$50-100 billion

Tiny machine learning (TinyML) is machine learning (including hardware, algorithms and software) capable of performing on-device sensor data analytics at extremely low power, enabling a variety of always-on use-cases and targeting battery operated devices.

It therefore enables any number of AI inference operations on a device, eliminating the need to transmit data to an external processor located elsewhere. Though part of the broader “AI at the edge” possibility, it further decentralizes AI inference operations, reduces latency to the greatest extent and likely will be used to support  highly-specialized inference operations running lightweight language models. 

Some obvious use cases include:

• Wearables such as a fitness tracker that analyzes your movements in real-time, offering personalized coaching or detecting falls.

• Smart homes devices that monitor temperature, humidity, and air quality, adjusting settings automatically.

• Predictive maintenance on machinery to predict potential failures before they happen.

• Environmental monitoring

• Agricultural sensors to optimize irrigation, detect pests and diseases.

But note that such lists of use cases actually are substitutes for older categories such as “internet of things” sensors. In many cases, devices and software that support TinyML will be used by the same devices once touted as being in the  “IoT at the edge” category.

It is not a new development. In the past, we saw tablets and smartphones displace PCs.  Smartphones displaced watches, cameras, GPS sensors, home phones and pagers. Now watches have in many cases become wearable computers. 

By some estimates, at least $48 billion worth of global device and product sales are lost every year because smartphones have displaced them. Some estimates believe the substitution results in a lost $70 billion of legacy product and sales activity annually. 

Displaced Device	Annual Revenue Loss from Sales of Other Products Because Smartphone Replaces Them (USD Billion)	Assumptions
Home Phone Service	25-35	Includes landlines and traditional VoIP services. Assumes partial displacement, with landlines remaining in niche markets.
Digital Cameras (Standalone)	15-20	Considers point-and-shoot and high-end DSLR cameras. Excludes mirrorless cameras still maintaining market share.
Wristwatches (Traditional)	5-10	Accounts for lost sales of non-smart watches, with smartwatches capturing a new market segment.
GPS Devices (Standalone)	2-3	Considers dedicated car and portable GPS units. Increased smartphone navigation usage contributes to revenue loss.
MP3 Players and Portable Media Players	1-2	Includes digital audio players and video playback devices. Niche market remains for high-fidelity audio equipment.
Pagers	Negligible	Pagers are practically obsolete, with near-complete displacement by smartphones.

In the U.S. market, lost revenue likely was in the $8 billion range in 2023, for example. That arguably is a low estimate, as the loss of a single residential phone line is assumed to be only $30 a month worth of lost revenue. 

There also could be a loss of other revenue from cost recovery mechanisms and also customer churn when the phone line is part of a service bundle.

Displaced Device	Lost Revenue Estimate (USD Billion)	Notes
Home Phone Service	5.8	Based on decline in residential landline subscriptions and average monthly service fee of $30.
Camera Sales	1.9	Based on decline in point-and-shoot camera sales and estimated average camera price of ~$200.
Watch Sales	0.5	Based on decline in traditional watch sales and estimated average watch price of ~$100.

On the other hand, mobile phone service also creates new markets for mobile service providers. 

But you get the point: new technologies often can redefine older markets and in many cases can be substitutes for the legacy products and services. 

We have seen this process at work in estimates of revenue to be earned by mobile service providers using network slicing to create new types of virtual private networks. But traditional VPNs, private networks,  traffic prioritization or edge computing are substitutes for network slicing, for example.

"You Get to Keep Your Business" is the Value of 5G

Though we are early in its life cycle, some argue 5G remains “revolutionary,” a “transformative leap” that will “reshape” connectivity. Others say it is a disappointment, either as a driver of near-term enterprise use cases, or as an enabler of higher-value, higher-revenue consumer mobile services.  

The eventual “truth” is likely to be far more nuanced. Some valuable new lines of business eventually could emerge. By design, 5G supports device density quite a bit higher than was available on 4G or earlier networks. By design, 5G supports network slicing, which enables private networks with some quality of service features. 

But 5G was always going to be a bit of a disappointment for most consumer accounts, which drive roughly 60 percent of total mobile operator revenues, for reasons related to the dynamics of all internet access and transport services. 

At a high level, demand for data consumption does not have a revenue elasticity that matches the consumption elasticity. In other words, mobile and fixed network operators cannot assume that increases in supply will produce increases in average revenue per unit that match the rate of consumption. 

To be sure, typical recurring charges for home broadband, for example, have increased since 1990, and access speeds for home broadband have risen as well. The cost to supply, on a per-unit basis, arguably is less important than the ability to charge more for higher consumption or higher speed. 

In the U.S. home broadband market, for example, per unit prices have plummeted, but typical  home broadband speeds have grown by an order of magnitude about every decade. Retail prices for stand-alone home broadband have not increased that fast, taking five decades to grow an order of magnitude. 

Year	Typical Speed	Typical Data Consumption	Price per Unit	Monthly Consumer Subscription Charges
1990	14.4 Kbps	10 megabytes	$1 per megabyte	$20 per month
2000	1.5 Mbps	100 megabytes	$0.1 per megabyte	$30 per month
2010	10 Mbps	1 gigabyte	$0.01 per megabyte	$50 per month
2020	100 Mbps	10 gigabytes	$0.001 per megabyte	$70 per month
2023	1 gigabit per second	1 terabyte	$0.0001 per megabyte	$100 per month

The key business takeaway is that supplied capacity must continue to increase, but will happen faster than price increases to match. 

Mobile operators have arguably had better outcomes where it comes to capacity supply and retail prices. In the U.S. market, it has taken three decades for prices to increase by an order of magnitude, as capabilities have grown three orders of magnitude. 

Year	Typical Speed (Mbps)	Typical Data Consumption (GB)	Price per Unit (GB)	Monthly Consumer Subscription Charges (\)
1990	1	0.1	100	20
2000	10	1	10	50
2010	100	10	1	100
2020	1,000	100	0.1	150

For such reasons alone, revenue expectations for faster mobile or fixed network internet access are likely to remain challenging. The argument that 5G would bring significantly-higher revenues, in the form of the ability to “charge more” for access speed, always was going to be quite difficult. 

It remains to be seen how much incremental new revenue might be created by internet of things connections, private networks or edge computing, the services most-often cited as “new” enterprise features of 5G. 

As it has happened, an unexpected “new” revenue source of some significant magnitude--in the form of 5G fixed wireless for home broadband--has developed rather quickly. 

Mobile Operator	Subscribers (Q1 2023)	Annual Subscription Revenues (Q1 2023)	Growth Rate (Q1 2022 - Q1 2023)
T-Mobile	3.2 million	$1.2 billion	120%
Verizon	1.5 million	$600 million	100%
AT&T	0.5 million	$200 million	50%

Whether one views such new revenue sources as “revolutionary” or “merely” significant is the issue. Still, the growth of 5G fixed wireless for home broadband clearly is important for contestants in the home broadband space. 

And 5G fixed wireless remains, at the moment, the clearest “new” revenue source for mobile operators. It always is conceivable that other enterprise-focused revenue streams will emerge as well, though the magnitude of those revenue streams remains more uncertain.

The point is that we likely err when arguing either for “revolutionary” or “disappointing” outcomes for 5G. We still are early in global deployment. New revenue sources generally take time to develop. 

But 5G remains vitally important for other reasons. All internet access providers, all data transport providers and data centers must increase capacity on a sustained basis. Each new mobile generation is the way that capacity increase happens.

Mobile operators may indeed be disappointed at the revenue outcomes from 5G so far. But 5G is essential for protecting the value of the business, as will be true of 6G and subsequent platforms. 

“You get to keep your business” might sound like a rather-trivial outcome. It is not.

Marketing Claims Aside, How Much Capacity Do Home Broadband Users Really Need?

How much internet access speed or usage allowance does a customer really need? It actually is hard to say. U.S. data suggests there are clearer answers about what customers expect to pay, which is about $50 a month, on average, even if some studies suggest wildly higher prices.  

source: Broadband Now 

Average prices are lower or higher than $50 a month, depending on what adjustments are made, such as adjusting for currency differences or cost of living differences between markets. Adjustments of that sort tend to show rather uniform global pricing of internet access, also adjusting for “quality” (speed, for example) differences. 

Also, any assessment should be based on service plans people actually buy, not posted retail prices for any particular tier of service. Bundle pricing adds another layer of complication. 

Internet service provider business models always require matching supply with demand; deployment speed and cost. What is needed to market effectively against competitors also matters. 

Time to market does matter. “It took us 22 years to pass 17 million households with fiber: 22 years,” says Hans Vestberg, Verizon CEO. “That’s how hard it is.”

“We basically had 30 million households covered with fixed wireless access in less than one year,” he also notes. 

So there is the trade off: rapid deployment of a lower-cost network versus slower deployment of a higher capacity network; wide coverage now versus higher capacity later; lower capital investment versus high. 

As typically is the case, wireless platforms can be provisioned faster than cabled networks, at lower cost. The Verizon data illustrates that fact. 

Cost also matters, as no internet service provider--especially those in competitive markets--can afford to spend unlimited sums on its infrastructure. Verizon and T-Mobile tout fixed wireless access in large part because they can afford to supply itt and can supply it fast, at lower costs than building fiber-to-home would cost. 

But marketing also matters: internet service providers do compete on the basis of speeds and feeds; do compete on price; do compete on perceptions of quality; terms and conditions and value. 

In that regard, even as home broadband speeds continue to rise, marketing claims are a battleground. Cable executives, for example, make light of fixed wireless as they claim it will not scale the way hybrid fiber coax and fiber-to-home can. FWA proponents argue that the platform does not have to scale as fast as FTTH or HFC to provide value for segments of the customer base. 

For example, even households that buy the fastest tiers of service rarely have a “need” for all that capacity. According to a survey by HighSpeedInternet.com, survey respondents say the “perfect plan” features a “610 Mbps fiber connection for $49 per month.”

In the third quarter of 2022, about 15 percent of U.S. households bought service operating at 1 Gbps, while 55 percent purchased service running from 200 Mbps to 400 Mbps. 

source: OpenVault 

 

The point is that, no matter what they tell researchers, U.S. home broadband customers do not seem especially eager to buy gigabit services at the moment, or services running at about half that speed. 

Speed demands will keep climbing, of course. But it does not appear, based on history, that most consumers will switch to buying the fastest tiers of service, or the lowest tiers of service, either. Historically, U.S. consumers have purchased internet access costing about $50 a month, with performance “good enough” to satisfy needs.

In fact, one might make the argument that is consumption (gigabytes consumed) that matters more than speed. Average data consumption stood at about 500 gigabytes per month in the third quarter of 2022, according to OpenVault. But the percentage of power users consuming a terabyte or more was growing fast: up about 18 percent, year over year, and representing about 14 percent of customer accounts. 

So speed claims are about marketing, as much as customer requirements. “It's turned into really a marketing game,” adds Kyle Malady, Verizon Communications EVP. ISPs compete on claimed speeds, even if there is little evidence most households require gigabit speeds at the moment. 

Beyond a certain point of provisioned capacity per user and device in any household, additional speed brings subtle if any benefits. Consumption allowances do matter, especially for households that rely on streaming for video entertainment. 

Nobody can give you a convincing answer why gigabit per second or multi-gigabit per second networks are required, beyond noting that multi-user and multi-device households need a certain amount of capacity if all are using the ISP connection at the same time. 

No single application, for any single user and device, requires a gigabit connection. So the real math is how much total bandwidth, at any moment, is needed to support the expected number of users, apps and devices in simultaneous use. 

For a single user or two, using one or two devices each, simultaneously, it is hard to see how a gigabit or faster connection is required. 

Some version of that argument--that a customer “does not need” a particular capability, is at the heart of much ISP marketing. ISPs whose platforms have some speed limitations point out that the limits do not matter for some customers, or that the price paid for higher-speed services does not provide value, commensurate with cost.