Why AI is Like Alta Vista in 1995

One of the issues for our understanding of artificial intelligence is that we are so early in the process. 

I think back to 1995 and the Alta Vista search engine, which I used for work, but which was not very useful, in large part because the number of indexed websites and pages was still low, and in part because the search architecture and user interface was not as "good" as Google search provided. 

But the biggest problem was simply that if indexing of websites is a proxy for access to the sum total of human knowledge, there simply wasn't that much total knowledge available on websites. In 1995, for example, there were probably only about 16,000 websites in existence. 

Compare that to the present, where more than two billion sites actually exist. 

The point is that we are still quite early in development of generative AI, for example, and it is hard to envision where and how it will develop. In 1995 I could not have imagined what would be possible for search, for example. 

One issue is that AI might take so many different forms.

One of the issues investors, entrepreneurs, end users and suppliers have to figure out about the various forms of artificial intelligence is “what is it?” In other words, using past computing analogies, AI might sometimes be akin to an operating system.

In other roles it might be more like a browser, a user interface, a platform or application or feature of an app. 

So what it means can vary for various participants in the value chain. For suppliers of computing capabilities, AI “as OS” could mean creation of a new computing ecosystem and paradigm. 

As a “browser” substitute, AI could change the way people search for information. As a UI, AI probably will change the way humans interact with computing machines. 

As a “platform” AI could be the foundation for whole new categories of software and ecosystems built on specific platforms. 

The least controversial analogy might be AI “as an app or feature,” since it will be integrated with most existing software and hardware in some way. 

Analogy	Key Points	Implications
Operating System	- Acts as a kernel process orchestrating other applications2 - Manages input/output across modalities - Controls code execution and browser access - Handles memory storage and retrieval	- AI could become the foundation of a new computing paradigm2 - May lead to more adaptive, customizable, and flexible systems
Browser	- Can automate web-based tasks like research and shopping3 - Enables AI to interact directly with web content - Allows for autonomous browsing and information retrieval	- Could revolutionize how we interact with online information and services - May lead to more efficient and personalized web experiences
User Interface	- Understands and generates multiple modalities (text, audio, vision)2 - Can replace traditional GUI-based interfaces - Enables natural language interaction	- Could make computing more accessible and intuitive - May lead to new forms of human-computer interaction
Platform	- Serves as a foundation for running AI models and applications2 - Manages AI-specific workloads - Integrates AI capabilities across various services	- Could become the basis for future software development - May lead to widespread AI integration in all digital services
App/Feature	- Enhances existing applications with AI capabilities3 - Automates specific tasks within larger systems - Improves user experience in targeted areas	- Could lead to incremental improvements in existing software - May result in more intelligent and responsive applications

Verizon's Home Broadband Scale Gambit

Though some might criticize the debt implications or the strategy, there is a reason Verizon is pursuing an acquisition of Frontier: it is one way to gain scale in the home broadband market.

Consider that although all telcos trail the two leading cable providers (Comcast and Charter) in national market share (those two firms have at least 63 percent national share, Verizon has just nine percent share compared to AT&T at 23 percent share. 

That is a result of the smaller geographic footprint Verizon has, relative to AT&T, Comcast and Charter. 

ISP	Subscribers (millions)	Market Share (%)
Comcast (Xfinity)	32.1	32.6
Charter (Spectrum)	30.4	30.9
AT&T (Fiber)	22.6	23
Verizon (Fios)	9.2	9.3
Lumen (CenturyLink)	4.8	4.9
Cox	7	7.1
Altice USA	4.7	4.8
Other (including smaller ISPs)	1.6	1.6
Total	98.5	100

U.S. internet service providers compete on a geographic basis and not all providers face all other providers. Comcast and Charter, both cable companies, generally do not compete head to head. Neither do AT&T, Verizon and Lumen Technologies. 

But sheer numbers of homes and other locations passed vary as well, with Comcast and Charter passing the most U.S. homes. 

ISP	Estimated Homes Passed (Millions)
Comcast	60
Charter	55
AT&T	30–35
Verizon	15–20
Lumen	10–15
Frontier	10–15
Altice USA	8–10
Windstream	6–8

ISPs also generally count small business broadband accounts within their “home broadband” totals, as well. 

ISP	Estimated Homes & Small Businesses Passed (Millions)
Comcast	65–70
Charter	60–65
AT&T	40–45
Verizon	20–25
Lumen	15–20
Frontier	12–15
Altice USA	10–12
Windstream	7–9

Also, differences in “homes and businesses” passed by any single ISP’s network long have mattered for assessments of the degree of competition. For example, when looking at telco fiber-to-home competition for cable hybrid fiber coax networks, the actual degree of competition has been shaped by the huge cost of upgrading telco copper access networks to fiber. 

That has limited the actual degree of competition between telcos and cable companies for decades, as it rarely is the case that a given telco has FTTH deployed ubiquitously in all its geographies. 

ISP	FTTH Homes & Small Businesses Passed (Millions)	Total Homes & Small Businesses Passed (Millions)	FTTH as % of Total Passings
AT&T	25–30	40–45	60–67%
Verizon	17–20	20–25	80–90%
Lumen	5–7	15–20	25–35%
Frontier	6–8	12–15	50–53%
Windstream	3–4	7–9	35–45%
Consolidated	1.5–2	4–5	30–40%

Traditionally, the “best” data we have had on the market share positions of cable and telco competitors has come from Verizon areas, as that is were FTTH facilities are most-ubiquitously deployed. And in those areas, Verizon has been able to gain a bit more than 40 percent market share, while the local cable operator has been able to hold on to 45 percent to 55 percent of the market, with other independent providers holding generally single-digit shares but growing. 

In a growing number of markets third-party providers have targeted areas where telco FTTH is not available, and in such areas have generally been able to garner up to 20 percent share. 

In some instances, where a cable company mostly competes with a municipal fiber network, and the local telco has no appreciable residential and small business fiber coverage, the municipal provider tends to get 20 percent to 30 percent market share. 

Provider Type	Estimated Market Share (%)
Cable Company	60–70%
Independent ISP	20–30%
Telco (non-FTTH)	5–15%
Other ISPs	2–5%

The degree of “other ISP” market share is shaped by the coverage area selected by the attacking independent ISP. Generally speaking, such ISPs will choose portions of an incumbent’s territory to operate in, rather than overbuilding an entire city or town, for example. 

As in the case of telco-cable competition, that necessarily restricts the degree of head-to-head competition across an entire market area, and is reflected in the lower take rates we generally see when a cable company competes against any fiber provider that does not cover the whole local market.

AI Agents Might be Key to Creating Whole New Industries

Think about the content creation chores you find most tedious and then imagine those tasks fulfilled by an artificial intelligence agent, such as Google is prepping with the release of Gemini 2.0.

For me, creating presentations has always been an unpleasant chore. So I’d welcome an agent that create such slide shows automatically, without me having to create outlines, create designs, find graphics and orchestrate all that. It’s coming.

If AI emerges as a “general-purpose technology,” the capabilities are likely to dwarf anything we can presently conceive of. That has tended to be the case with all prior GPTs as well, including:

• Spoken language
• Mastery of fire
• The wheel
• Writing
• The printing press
• The steam engine
• Electricity
• The railroad
• The automobile
• The telephone
• The computer
* The internet

“Imagine a world in which you just write a SQL statement that goes to act on a PDF and produces a bunch of structured information out on the other side,” says Snowflake CEO Sridhar Ramaswamy.

Still, it is worth noting that we are likely unable to envision many ways AI, as a GPT, could create new possibilities. Most often, we can conceive of how AI can automate and therefore improve the utility of any number of processes.

What generally seems to escape us are the new things we never imagined before. Sure, most of us can imagine an artificial intelligence agent that takes care of other consumer chores as well. The agent might find products and negotiate prices on behalf of a consumer user; find vacation destinations and creates your plans.

Also conceivable are AI-enhanced methods for managing home heating, cooling or lights, health monitoring that additionally schedules appointments when needed.

But those are examples of things we can imagine. The big breakthroughs tend to be things we have not encountered before, and mostly cannot imagine. Search and social media are examples. Ride-hailing likewise is something most of us would not have predicted.

Agent functions seem to be the sorts of experiences that could fuel entirely-new ways of doing things, and hence create possible new firms and industries.

Business uses are likewise areas where our own imaginations don’t help much. Sure, we can see how AI can further automate what we already do, improving:

• Real-time customer service
• Inventory management
• Marketing campaign management and content generation
• Vehicle fleet performance monitoring and maintenance scheduling
• Automated task delegation and workflow management
• Fraud detection
• Quality control in manufacturing processes
• Predictive maintenance for equipment and machinery
• Automated customer sentiment analysis

But that is largely predictable.

What always is difficult are envisioning entirely-new value propositions and the impact on existing ways of doing things. That, in turn, means the possibility of disruption of firm and industry market positions, unit economics, competitive dynamics, and business strategies.