Monday, April 14, 2025

Telco AI Monetization on the Revenue Front Will be Difficult

Mobile executives these days are talking about ways to monetize artificial intelligence beyond using AI to streamline internal operations. Generally speaking, these fall into three buckets:

  • Personalizing existing services to drive higher revenue, acquisition and retention (quality of service tiers of service, for example)

  • Creating enterprise or business services (private 5G networks with AI-optimized performance,, for example)

  • AI edge computing services for autonomous vehicles, for example


Obviously, those are AI-enhanced extensions of ideas already in currency. But some of us might be quite skeptical that such “AI services” owned by telcos will get much traction. History suggests the difficulty of doing so. How many “at scale” new products beyond voice have telcos managed to create? Text messaging comes to mind. Mobile phone service also was a big success. So is home broadband. 


All those share a common characteristic: they are network services owned directly by the service providers. Generally speaking, other application efforts have not scaled well. 


Mobile service providers have been hoping and proclaiming such new revenue opportunities since at least the time of 3G. But many observers might agree there has been a disconnect between the technical leaps (faster speeds, lower latency, better efficiency) and the ability to turn those into new revenue streams beyond the basic "sell more data" model. 


That is not to say that service providers have had no other ways to add value. Bundling devices, content and other measures have helped increase perceived value beyond the core network features. 


But the core network as a driver of new products and revenue is challenging for a few reasons. 

  • Open networks mostly have replaced closed networks (IP versus PSTN) 

  • Applications are logically separate from network transport (layers)

  • Permissionless app development is the norm (internet is the assumed network transport)

  • Vertical control of the value replaced by horizontal functions (telcos had full-stack control of voice, but only horizontal transport functions for IP-based apps)


As I have argued in the past, modern telcos have a hybrid revenue model. They are full-stack “service” providers for voice and text messaging. But they are horizontal transport providers for most IP apps and services, and sometimes are app providers (owned entertainment video services, for example). 


The point is that most new apps and revenue cases can be built by third parties without telco or mobile operator permission, which also takes transport providers out of the direct revenue chain. 


So I’d argue there is a structural reason why telcos and mobile service providers do not directly benefit from most of the innovation that happens with apps. Think about all the customer engagement with internet-delivered apps and services, compared to service provider voice and messaging. 


In their role as voice and text messaging providers, telcos are “service providers” (they own and control the full stack). For the rest of their business, they are transport or access providers (capacity or internet access such as home broadband), a horizontal value and revenue stream. ISPs get paid to provide “internet access,” not the actual end user apps. 


And that has proven a business challenge for now-obvious reasons. Once upon a time, voice services were partly flat-rate and partly usage-based. In other words, telcos earned money by charging a flat fee for access to the network, and then variable usage based on number, length or distance of voice calls. 


In other words, greater usage meant greater revenue. But flat-rate voice and texting usage subverts the business model, as  most of the revenue-generating services become usage-insensitive. That is the real revolution or disruption for voice and texting. 


In their roles as internet access providers, some efforts have been made to sustain usage-based pricing. Customers can buy “buckets of usage” where there is some relationship between revenue and usage. 


Likewise, fixed network providers have used “speed-based” tiers of service, where higher speeds carry  higher prices. Still, those are largely flat-rate approaches to packaging and pricing. And the long-term issue with flat-rate pricing is that it complicates investment, as potential usage of the network is capped but usage is not.  


So as much as ISPs hate the notion that they are “dumb pipes,” that is precisely what home or business broadband access is. So internet access take rates, subscription volumes and prices are going to drive overall business results, not text messaging, voice or IoT revenues. 


To be sure, we can say that 5G is the first mobile generation that was specifically designed to support internet of things applications, devices and use cases. But that only means the capability to act as a platform for open development and ownership of IoT apps, services and value. And even if some mobile service providers have created app businesses such as auto-related services, that remains a small revenue stream for mobile service providers.  


Recall that IoT services are primarily driven by enterprises and businesses, not consumers. Also, the bulk of enterprise IoT revenue arguably comes from wholesale access connections made available to third-party app or service providers, and does not represent telco-owned apps and services (full stack rather than “access services”). 


Optimistic estimates of telco enterprise IoT revenues might range up to 18 percent, in some cases, though most would consider those ranges too high. 


Region/Group

Total Mobile Services Revenue 

IoT Connectivity Revenue (Enterprises)

Automotive IoT Apps Share of IoT Revenue

% of Total Revenue from Automotive IoT Apps

Global Average

$1.5 trillion (2025 est.)

10-15% (2025, growing to 20% by 2027)

25-35%

2.5-5.25%

North America (e.g., Verizon)

$468 billion (U.S., 2023, growing 6.6% CAGR)

12-18% (2025 est.)

30-40% (high 5G adoption)

3.6-7.2%

Asia-Pacific (e.g., China Mobile)

$600 billion (2025 est.)

15-20% (strong automotive industry)

35-45% (leader in connected cars)

5.25-9%

Europe (e.g., Deutsche Telekom)

$400 billion (2025 est.)

10-15% (CEE high IoT reliance)

25-35%

2.5-5.25%

Top 10 Mobile Operators

$1 trillion (2025 est.)

12-18% (based on 2.9B IoT connections)

30-40%

3.6-7.2%


Though automotive IoT revenues (again mostly driven by access services) arguably are higher for the largest service providers, their contribution to  total business revenues is arguably close to three percent or so, and so arguably contributing no more than 1.5 percent of total revenues, as consumer services range from 44 percent to 65 percent of total mobile service provider revenues. 


Category

Percentage of Total Revenue

Example products

Services to Consumers

55-65%

Driven by mobile data (33.5% in 2023), voice, and equipment sales; 58% in 2023

Services to Businesses

35-45%

Includes enterprise, public sector, and SMBs; growing at 7.1% CAGR

Business Voice

5-10%

Declining due to VoIP adoption and mobile data preference

Business Internet Access

15-25%

Rising with 5G, IoT (e.g., automotive apps at 2.5-9%), and enterprise demand


The point is that the ability to monetize AI beyond its use for internal automation is likely limited. Changes in the main revenue drivers (consumer and business mobile phone subscriptions and prices) are going to have more impact on revenue and profit outcomes than IoT as a category or automotive IoT in particular.


Sunday, April 13, 2025

China Approaches AI Diffusion as "Asian Tigers" Approached Economic Development

Among the differences between the U.S. and China frameworks for fostering widespread use of artificial intelligence are the roles of state sponsorship, which is in many ways similar to the state-led models for economic development followed in past decades by Japan, South Korea and Singapore, for example. 


Since at least the 1970s, the Asian Tiger economies (South Korea, Taiwan, Singapore, and Hong Kong) have shared similar “state-led” approaches to economic growth, compared to the more market-led U.S. approach. 


Country

Key Period

Role of the State

Industrial Policy

Key Mechanisms of State Involvement

Outcomes

South Korea

1960s–1980s

Highly interventionist

Targeted heavy industries and tech sectors

- State-directed credit via national banks

- Chaebol system fostered

- Export targets and subsidies

Rapid industrialization; major global brands (Samsung, Hyundai); GDP per capita rose from ~$100s to ~$30k

Taiwan

1950s–1980s

Strong state direction with private sector support

Promoted SMEs; focused on tech and electronics

- State-owned firms in key sectors

- Tech parks like Hsinchu

- Export processing zones

Tech powerhouse (TSMC, Foxconn); transition to high-tech exports; equitable land reform promoted broad growth

Singapore

1965–present

State-capitalist model

Moved from labor-intensive to high-tech, biotech, finance

- GLCs (Temasek, GIC) played a major role

- Strategic FDI attraction

- Skills training via ITEs, polytechnics

Among world’s richest countries per capita; GLCs still key; high-quality infrastructure and education

Hong Kong

1950s–1997 (British rule)

Minimal intervention – laissez-faire

No formal industrial policy

- Low taxes

- Free trade

- Rule of law and strong institutions



We might see a similar approach in the way China approaches mass adoption of artificial intelligence technologies. 


Model

Financial Support

Infrastructure, Data Access

Regulatory, Strategic Support

Public Sector Adoption

DeepSeek (Hangzhou)

Indirect: Benefited from government-backed cloud platforms

Access to state-supported GPU clusters; open-source model strategy

Endorsed by local governments; integrated into national AI initiatives

Deployed in hospitals, local governments, and state-owned enterprises

Qwen (Alibaba Cloud)

Participated in Beijing's AGI Innovation Partnership Program

Utilized Alibaba Cloud's infrastructure; part of government-supported computing power initiatives

Received government approval for public release; aligned with national AI development goals

Integrated into Alibaba's consumer services; supports various enterprise applications

Ernie Bot (Baidu)

Not specified

Leveraged Baidu's infrastructure; part of national AI development efforts

Approved under China's generative AI regulations; contributes to national AI objectives

Claimed 200 million users; used in various public-facing services

Doubao 1.5 Pro (ByteDance)

Not specified

Utilized ByteDance's infrastructure; aligned with national AI strategies

Operates within China's regulatory framework for AI; contributes to national AI initiatives

Integrated into ByteDance's platforms; supports content creation and user engagement

Kimi k1.5 (Moonshot AI)

Not specified

Details not publicly available

Operates within China's regulatory framework for AI; contributes to national AI initiatives

Specific public sector adoption details not available

Xinghuo (iFlytek)

State-backed: iFlytek is partially state-owned; received government funding

Trained on Huawei's computing platform; aligned with national infrastructure goals

Designated as an "AI champion" by the government; operates within regulatory frameworks

Integrated into educational tools and public services; supports various government applications

ChatXiPT (Cyberspace Administration of China)

Fully government-funded

Developed using state resources; aligned with national infrastructure

Designed to promote Xi Jinping Thought; operates under strict regulatory oversight

Used for ideological education and public information dissemination

Saturday, April 12, 2025

AI Stack Will be Based on Layers, Just Like All Other App Ecosystems

“Digital infrastructure” tends to refer to physical assets like data centers, fiber optic networks, cell tower networks and cloud computing “as a service” providers, at least as seen by investors in and operators of such assets. 


And even the cloud computing providers (AWS, Zaure, Google Cloud and so forth) most often viewed as the key customers for digital infra providers, rather than core parts of the digital infrastructure business itself. 


However, as AI capabilities become more integral to business operations, there is an argument that digital infrastructure should also encompass AI as a service (AIaaS) capabilities which extend beyond hardware to include software models, language models, and AI platforms. 


Right now that is perhaps not the case, as investors and operators of language models tend to be distinct from traditional “digital infra” providers and investors. 


The “AI ecosystem” is different from the “digital infra” ecosystem, as the former necessarily includes chips, servers, language models, other AI systems and apps. 


As a practical matter, the expanded definition probably will not happen, for several reasons. For starters, financial analysts and operators of “computing” businesses tend to be different from “digital infra” analysts and operators. The former tends to be anchored by “software and hardware” interests, while the latter tends to be dominated by “real estate” interests. 


Foundation models and language models such as  GPT-4, Gemini, and Claude will tend to be the province of “software” industry analysts and practitioners; “hardware” analysts and practitioners such as Nvidia as well as the traditional “computing” ecosystem participants and analysts. 


In other words, the analysts and businesses that are in “middleware” and software stacks and tools, plus applications, are distinct from the analysts and businesses that supply “real estate” functions such as data transmission and data center operations. 


“Cloud computing” might be the function of many data centers, both including AI operators and all other software hosting, but there still are key differences between the businesses of computing and connectivity real estate and the actual software applications that use those real estate assets and platforms. 


So “infrastructure” and “applications” (including language models and AI as a service) will likely continue to remain separate areas of interest. 


Thursday, April 10, 2025

How Much Will AI Compute Grow to 2030, Compared to Electricity Consumption?

A report by the International Energy Administration estimates data centers accounted for around 1.5 percent of the world’s electricity consumption in 2024, and might “more than double” to three percent of electricity consumption by 2030. 


What that means partly depends upon one’s perspective. Consider that estimates of AI-related compute cycles between 2025 and 2030 can range from growth of several hundred percent to 1000 percent (five times  to 10 times).


In that sense, an increase in use of electricity to power the servers handling those operations is relatively low, in relation to the increase in compute volume. 


Of course, the report also emphasizes that the outlook is “highly uncertain.” As always, much hinges on the assumptions about consumption growth and efficiency gains. 


source: IEA 


Ther study pegs data center electricity demand growth at more than space and water heating but less than space cooling or appliance use. 


source: Nature, IEA

D-Day Plus 81

Some things should not be forgotten.  D-Day, 81 years ago  is among them.