Wednesday, October 16, 2024

How Should Governments Subsidize Unlimited Internet Access?

A new Federal Communications Commission inquiry on data caps does raise some obvious issues, as the action is driven by complaints by internet users about the very existence of such caps, which might be likened to rationing of a resource, says FCC Chairwoman Jessica Rosenworcel. At least so far, the concerns seem directly related to data limits related to the cost of service plans.


In the press release announcing the inquiry, the FCC included complaints such as “We can't afford $190 a month for unlimited internet.” Other cited complaints revolve around “excessive costs” or present data caps being insufficiently capacious. 


On the other hand, differing prices based on differing consumption volume is a fairly-standard principle in most businesses and industries, though some subscription services, such as linear video, satellite radio or other audio streaming services routinely operate on an “unlimited consumption” model. 


So do many mobile phone voice and texting service plans, at least for domestic usage.


As a rule, consumers seem to expect volume-related pricing for physical goods and most intangible products, whether that is water, electricity, groceries, fuel or clothing. 


And, for providers, total costs of creating and providing a service or product matter, not only the cost of one particular part of the value chain. Some observers focus only on the marginal cost of providing the next unit of consumption, not full costs (capital investment and all operating costs). 


The total cost of providing an internet access service arguably differs for large, dominant providers and smaller local providers. 


Though network Infrastructure, bandwidth and transit costs are of high importance for all ISPs, smaller, regional ISPs might tend to find that bandwidth and infrastructure costs dominate, whereas larger ISPs might find that marketing, research and development costs, and regulatory compliance may take on greater importance.


Customer premises equipment, labor, marketing and customer acquisition costs generally are of medium importance for all ISPs. 


But large ISPs might find the costs of regulatory compliance, research and development as well as energy costs to be of more significance, compared to how those issues pertain to small ISPs. 


Study Title

Date

Publication Venue

Key Conclusions

"The Economics of Internet Traffic"

2002

Journal of Economic Perspectives

Found that marginal cost is generally low, especially for peak-time traffic.

"The Cost of Internet Traffic: A Study of Residential Broadband Access"

2005

Telecommunications Policy

Examined the cost structure of residential broadband access and concluded that marginal cost is relatively low.

"The Economics of Internet Traffic: A Critical Review"

2010

Telecommunications Policy

Provided a comprehensive overview of research on the economics of Internet traffic, highlighting the challenges of measuring marginal cost accurately.

"The Cost of Internet Traffic: A Survey of Recent Studies"

2014

Telecommunications Policy

Summarized key findings from recent studies on the marginal cost of Internet traffic, emphasizing the importance of network congestion and traffic patterns.


The point is that the marginal cost of providing the next unit of capacity or consumption might not be the only measure, or the best measure, of cost and its relationship to consumer pricing. Providers can affect some of their total costs. But many fundamental costs, including network infrastructure, are relatively inelastic. 


Other costs have some elasticity, but can be hard to contain in highly-competitive markets. So the actual marginal network cost of producing the next unit of capacity might not be the best metric for assessing the “fairness” of access pricing. 


The larger issue, perhaps, are the sustainable business models allowing internet service providers to continually expand capacity, providing the needed usage support for consumers, at prices those consumers consider fair and reasonable. All of that is dynamic.


To the extent that “cost of use” is a financial problem, governments routinely use subsidies of various types to support consumption of essential or important goods by some citizens who would not otherwise be able to afford such goods. 


As always, the issue of “who pays” has to be answered in the concrete. To the extent that ISP sustainability literally requires profits, providers have to keep working on efficiencies so they can keep costs “reasonable.” And observers debate the degree to which customer usage volume actually matters.


Logically, marginal costs exist when customers use more of a resource. But how much marginal cost exists is an issue. Fixed or sunk costs might actually predominate. But again, subsidy programs can be created that address the needs of specific populations deemed to require support.


Monday, October 14, 2024

ChatGPT is the VisiCalc of the AI Era

VisiCalc, a spreadsheet program, was the catalyst that enabled enterprises to adopt personal computers. providing the practical use case for the machines. Technology adoption often is triggered by one specific use case or capability.

So many could point to ChatGPT as the breakthrough app enabling widespread use of artificial intelligence by people as a retail application, rather than as a background feature. It is a common occurrence in computing echnology. 

VisiCalc, released in 1979, was a watershed moment for personal computing. As the first electronic spreadsheet program, it provided a compelling reason for businesses to invest in personal computers. It paved the way for future spreadsheet programs like Lotus 1-2-3 and Microsoft Excel.


Microsoft Office, particularly Excel and Word, has dominated knowledge work for decades, and is a development based on early spreadsheet and word processing use cases. 


In similar ways the BASIC Programming Language (Beginner's All-purpose Symbolic Instruction Code), developed in 1964, played a crucial role in democratizing programming, making coding accessible to a wider audience, including non-specialists.


The Unix operating system, developed in the late 1960s and early 1970s, laid the groundwork for open-source development models.


Some would say the same for the Oracle database management system, considered a standard for enterprise relational database systems. 


The World Wide Web, HTML and Netscape Navigator enabled the multimedia Web, making the internet experience useful to the general public. 


The SABRE computerized reservation systems enabled real-time transaction processing systems for airlines, and was a forerunner of online e-commerce. Likewise, the IBM Customer Information Control System enabled real-time transaction processing. 


The Java Programming Language, released by Sun Microsystems in 1995, introduced the concept of "write once, run anywhere," enabling cross-platform development and reducing the need for platform-specific code. 


The Linux Kernel further enabled open source. Likewise, the Apache web server democratized web hosting, allowing more individuals and organizations to establish an online presence.

It's hard to disagree with Jensen Huang, Nvidia CEO, about OpenAI's artificial intelligence influence or role. And its strategy--"insane" levels of capital investment--seem calculated to destroy the ability of would-be competitors to keep up. The analogy of an armaments race is apt. 

Immediate or near-term payback is not likely, and perhaps not possible, using that strategy. The possible outcome is that most of the other would-be competitors have to concede some level of defeat as their investors punish firms for investing without reward. 

Friday, October 11, 2024

Warning Labels for GenAI are Really Important

Liability is by definition a contentious matter, and will have to be updated with the rise of generative artificial intelligence content. Who is responsible for hallucinated or incorrect material, for example? Some might argue it is the language model provider, but that seems unlikely to happen, as a rule. Instead, users will likely still be held liable. 


LinkedIn, for example, is updating its user agreements to make clear that the site might show some artificial intelligence generated content that is inaccurate. 


Some might argue that product liability frameworks apply while others might see service contract frameworks as a possible model. In the former case, suppliers could be held liable for product defects or “failure to warn” of misuse. The “product defect” defense might be hard to prove, as it requires some proof of faulty design or production. 


The latter should be easy to defend: just make sure warnings about possible inaccuracies are prominent.


In that sense, the "warning labels" are really important, as they offer liability protection for providers of large language models.


Will Alphabet Antitrust Even Matter, By the Time is is Finally Resolved?

Potential antitrust action against Alphabet could include asset divestitures, though some believe  more-likely outcomes are behavioral measures that might temporarily slow Alphabet growth, but could  have fewer longer-term negative consequences--with one glaring exception.


If Alphabet leaders are consumed with defending themselves against antitrust, it is possible they will be constrained from moving forward in some key new area (possibly related to artificial intelligence), much as some believe Microsoft fell behind in mobility because of its antitrust efforts. 


The precedent there is the Microsoft antitrust action of 2001, which caused some initial changes in business practices intended to benefit competitors, but which arguably did not slow down Microsoft growth in other areas, with the notable exception of mobility and smartphones. 


Indeed, some might argue that Microsoft sought growth in other areas precisely because of the behavioral remedies. In other cases, even asset divestitures have had complex outcomes. 


The breakup of the AT&T system in the early 1980s was intended to promote competition, and did so. But consolidation followed and AT&T essentially was reassembled. Competition--the intended outcome of the breakup--did increase. 


But AT&T arguably was more affected by the emergence of the internet and the mobile communications revolution than by the antitrust actions. AT&T’s legacy businesses are a much-smaller part of overall revenue compared to mobility, which generates more than half of total revenue. 


Indeed, the long-term impacts on industry structure and dominant firm performance are complex. Standard Oil was broken up into 34 different competing firms. But consolidation followed and the surviving firms arguably were not harmed. 


Exxon; Mobil (eventually combined to form ExxonMobil; Chevron; Amoco (later acquired by BP) and Marathon Oil were formed by state-level Standard Oil divisions, for example. 


IBM’s antitrust suit eventually was dropped, but that firm’s fortunes were arguably shaped more by the emergence first of the minicomputer and then by personal computing than regulatory action. 


In fact, one possible outcome is that the case drags on long enough that market dynamics already have shifted, lessening the importance of any proposed remedies. It is possible that Google’s search dominance already will have declined because of generative AI alternatives, long before the antitrust action is applied. 


Much product search and other forms of discovery already have shifted to Amazon or social media, while AI-powered “answers” are poised to disrupt other forms of search as well. In other words, the Department of Justice antitrust action might be coming just at the point where it becomes almost irrelevant, by the time it is settled, if any action occurs at all. 


The Microsoft antitrust action lasted a decade before being resolved, and some might argue that shifted the playing field to mobility and phones, making the original reasons for antitrust actions around personal computers and browsers somewhat moot. 


Also, Alphabet also faces antitrust action in other countries, which could lead Alphabet to take voluntary actions that alleviate those concerns in ways that lessen Alphabet’s market dominance, but without huge structural changes such as breaking Alphabet up into smaller “bets.” 


Some have suggested Android and Chrome, or perhaps YouTube wind up becoming products owned by separate companies. What remains unclear are possible changes--or continuity--of consumer behavior. Users might not switch from using Google for search, Android for the operating system of their mobile devices or Chrome for their preferred browser, anymore than they’d switch from using YouTube to some other app. 


Rapid technology change can upend even well-intentioned reform. The Telecommunications Act of 1996, for example, aimed to increase competition in the communications market largely around voice services. 


Though arguably succeeding in that sense, the Act missed the arrival of the internet, and exempted mobile services, both of which had an even more profound impact on connectivity, content and communications than the effort to promote competition in fixed-network voice services. 


To some genuine extent, the Telecom Act focused on the wrong problem, or perhaps on an unnecessary problem, given the disruptive changes the internet and mobility were unleashing. 


One might have a disquieting--and similar--feeling about the antitrust action against Google. By the time it is resolved, it might not matter so much.


Thursday, October 10, 2024

DT Revenue Growth: Scale or Scope?

Deutsche Telekom says it plans to boost revenue growth by increasing economies of scale and using artificial intelligence. The promise of AI to reduce costs is likely understood by all observers. The “economies of scale” might be more complicated, as that term implies wringing cost out of existing operations by selling more, or to more customers, using the same assets. 


Strictly speaking, the latter phrase (“scale”) refers to selling at higher volumes (to more customers). But some of DT’s stated plans might involve selling new or different products to the same customer base, which, strictly speaking, is “economy of scope.” 


In other words, “scale” means selling a product to more customers. “Scope” means selling additional things to existing customers. As a practical matter it might not matter whether what DT intends are examples of scale or scope. It is likely both will be at work.

  

DT expects to sell “additional products and services ranging from payment services for cell phone insurance services and platforms for payment services through to AI solutions for consumers” in its mobile business, which is a clear example of scope economics. 


In the global business markets, DT seems to suggest gains will come from higher sales to more customers, which is a “scale” economy. 


In the telecom industry, “economies of scale” can be operationalized as instances where the average cost per user decreases as the volume of services provided increases. That generally arises from spreading large fixed infrastructure costs over a growing number of subscribers; increasing sales to those customers or otherwise optimizing network usage to reduce cost per unit.


So, compared to some other industries, scale economies are more difficult, as the physical network footprint generally has to be increased to reach more potential customers (acquisitions of other telcos, for example; or building out new networks outside the present geographic footprint). 


Industry

Economies of Scale Potential

Fixed Costs

Marginal Costs

Scalability

Barriers to Scaling

Virtual Products (e.g., SaaS, streaming)

Extremely High

High (development, initial infrastructure)

Near zero (reproducing digital products)

Unlimited (global reach)

Low (mainly infrastructure scaling, user acquisition)

Telecom Networks (e.g., Fiber, Cellular)

Moderate

Very High (infrastructure: cables, towers)

Significant (capacity upgrades, maintenance)

Limited (capacity constraints, physical coverage)

High (geography, regulation, infrastructure costs)

Manufacturing (e.g., Electronics)

High

High (factories, machinery)

Low (once economies of scale are achieved)

High (limited by supply chain and logistics)

Moderate (supply chain constraints, capital investment in machinery)

Automobile Production

Moderate to High

High (factories, R&D, supply chains)

Moderate (labor, raw materials, logistics)

High (dependent on supply chain, market demand)

Moderate (complex supply chain, regulation, capital intensive)

Retail (e.g., E-commerce)

Moderate

Moderate (warehousing, logistics)

Low (online distribution, logistics costs decrease with scale)

High (digital platforms scale easily)

Moderate (logistics, competition, last-mile delivery costs)

Healthcare (e.g., Hospitals)

Low to Moderate

Very High (equipment, staff, real estate)

High (labor, equipment usage, pharmaceuticals)

Limited (physical capacity, staffing limitations)

High (regulation, physical constraints, capital-intensive infrastructure)

Energy (e.g., Renewable energy production)

Moderate

Very High (plant construction, grid integration)

Low to Moderate (depending on energy source)

Moderate (limited by physical infrastructure)

High (regulatory barriers, physical infrastructure expansion)

Education (e.g., Online platforms)

High

Moderate (platform development, content creation)

Near zero (digital content distribution)

Very High (global reach, online scalability)

Low (content development, digital infrastructure scaling)

Logistics (e.g., Delivery services)

Moderate

High (transportation, warehousing)

Moderate (fuel, labor, vehicle maintenance)

Moderate (dependent on infrastructure and efficiency)

Moderate (geography, labor, fleet expansion)

Financial Services (e.g., Banking, FinTech)

High

Moderate (technology, regulatory compliance)

Low (digital transactions, account maintenance)

High (digital services can scale globally)

Moderate (regulation, cybersecurity, trust building)


Still, some might argue that telco potential for economies of scale is less than might be expected. When a new geography is to be served, additional capital investment is required. So, by definition, the additional customers and revenue are not generated by the “same” assets, which would imply lower cost per customer. 


To be sure, there are possible economies in other areas (back office, overhead), but telco geographic expansion on a facilities basis requires additional investment in plant. 


So DT’s possible upside is more likely to come from “scope” in its consumer business, but possibly “scale” in its global business customer segment.


Wednesday, October 9, 2024

How Much Might Generative AI Boost Productivity Across Industries?

According to Bank of America equity analysts, AI impact on productivity is going to vary. Most industries--but not all--should see productivity gains of two percent or less over the next five years, with a handful of industries supplying infrastructure expected to outperform.


As you might expect, software and semiconductor industries will lead the list of winners, with software profit margins gaining as much as five percent and semiconductors gaining nearly five percent. 


source: Bank of America Institute 


Healthcare and telecom were laggards, despite some claims that telcos are deploying generative AI faster than the other industries, at least according to telco firm survey respondents who were technology C suite or IT heads at about 1600 global organizations. 


Ignoring the obvious self interest technology leaders have in claiming they are moving rapidly to adopt generative AI and AI, the point made by the Bank of America analysis, which was produced by industry-specific financial analysts, is that actual outcomes related to productivity might be relatively modest in most instances, at least over the next five years. 


One problem is that some industries are likely positioned to improve productivity at faster rates than others, with or without GenAI, perhaps because they already are better positioned to deploy new technology to boost outcomes. 


 

source: SAS


The other caveat is that since knowledge worker productivity is notoriously difficult to measure, such surveys might be looking at other matters, such as firm agility, industry adaptiveness to new technology or industry growth rates in general, which are higher in some industries than others. 


Industry

Revenue Growth Rate

Technology

8-12%

Healthcare

5-10%

E-commerce

10-15%

Financial Services

4-8%

Renewable Energy

12-20%

Real Estate

3-6%

Consumer Goods

2-5%

Telecommunications

1-4%

Automotive

2-3%

Travel and Hospitality

6-10%


Directv-Dish Merger Fails

Directv’’s termination of its deal to merge with EchoStar, apparently because EchoStar bondholders did not approve, means EchoStar continue...