/

$AAPL $META $GOOGL $AMZN $NVDA $MSFT

Most investors compare artificial intelligence to the internet. We believe the better analogy may be railroads.

At first glance, the comparison seems odd. Railroads moved goods. AI processes information. Yet from an economic and investment perspective, the similarities are striking.

Both technologies required massive upfront infrastructure investments before their full benefits could be realized. In the 1830s and 1840s, investors poured enormous sums into tracks, bridges, locomotives, and rail networks. Today, technology companies are investing hundreds of billions of dollars into GPUs, datacenters, power infrastructure, cooling systems, and AI models.

The railroad boom ultimately transformed the global economy. Transportation costs collapsed, trade expanded, and productivity surged across nearly every industry. Yet many early railroad investors lost substantial amounts of money as speculation, overbuilding, and fierce competition overwhelmed returns on capital.

That is the critical lesson for today's AI investors.

Transformational technologies do not automatically translate into transformational shareholder returns.

The key question is not whether AI will change the world. It almost certainly will. The more important question is whether the companies spending hundreds of billions of dollars building AI infrastructure will ultimately earn attractive returns on that investment—or whether AI becomes another railroad, creating enormous value for society while delivering disappointing returns for many of its earliest investors.

Understanding that distinction may be one of the most important investment questions of the next decade.For most of the last two decades, the MAG7 owned some of the greatest businesses ever created.

They possessed characteristics Warren Buffett has historically prized:

In many cases, these businesses looked less like traditional corporations and more like economic monopolies disguised as software companies.

The AI era may be changing that equation.

The critical question for investors is not whether AI is powerful.

The critical question is whether AI strengthens existing moats—or forces the world's best businesses to spend enormous amounts of capital merely to defend them.

Every additional search query cost virtually nothing.

Google's economic engine was among the most efficient businesses ever created.

Great Business (10/10)

Network effects dominated the ecosystem.

Every additional user made the platform more valuable for every other user.

Advertisers followed attention.

Attention followed users.

The flywheel reinforced itself.

Capex remained modest relative to profits.

Great Business (9/10)

Microsoft built one of the strongest enterprise moats in modern history.

For most customers, abandoning Office was more expensive than continuing to use it.

Great Business (10/10)

Apple created perhaps the strongest consumer ecosystem ever assembled.

Hardware drove software.

Software drove services.

Services reinforced hardware.

Every product strengthened the ecosystem.

Great Business (10/10)

The fascinating question is simple:

Does AI strengthen moats or weaken them?

Many investors assume AI automatically strengthens dominant incumbents.

The answer is not nearly that obvious.

In many cases, AI appears to weaken the very forces that made these businesses extraordinary.

Building a Google competitor required decades.

Building a Meta competitor required billions of users.

Building a Microsoft competitor required enterprise distribution.

The barriers were enormous.

Today, a small startup can leverage foundation models and cloud infrastructure to launch products that previously required thousands of engineers.

The cost of software creation has collapsed.

The moat shifts from software creation to distribution and data.

Moat Weakens

10 → 7

Google built software.

Meta built software.

Microsoft sold software.

Their suppliers had limited influence.

The AI stack increasingly depends on:

The economic leverage moves away from software owners and toward infrastructure providers.

Suddenly, suppliers matter.

A lot.

Much Worse

9 → 5

Users had very few alternatives.

Search meant Google.

Productivity meant Office.

Social meant Meta.

Users increasingly move between:

Switching costs are falling.

When answers become interchangeable, customers gain negotiating power.

Customer Power Rises

8 → 6

Search had no true substitute.

Office had no true substitute.

Social media operated within clearly defined categories.

AI creates substitutes everywhere.

Search competes with AI agents.

Office competes with AI-generated workflows.

Social platforms compete against infinitely generated personalized content.

Entire categories begin overlapping.

Substitutes Increase

9 → 6

Competition existed but was manageable.

Google vs Bing.

Meta vs Snap.

Microsoft vs smaller software vendors.

The battlefield was relatively stable.

Everyone competes against everyone.

Competitive boundaries are disappearing.

Search is becoming software.

Software is becoming AI.

AI is becoming infrastructure.

The number of credible competitors expands dramatically.

Rivalry Intensifies

8 → 5

The real concern isn't AI.

The real concern is capital intensity.

Hypothetical Economics

Fantastic economics.

Every incremental dollar of revenue produced enormous shareholder value.

Hypothetical Economics

Free cash flow compresses.

The business begins looking less like software and more like infrastructure.

Less like Google.

More like:

Ironically, Buffett eventually loved railroads.

Why?

Because once the network existed:

The same outcome could emerge in AI.

The question is:

Who owns the AI network?

Google, Meta, Microsoft and Amazon successfully control:

AI becomes a toll bridge.

Every user and business must cross it.

8–10/10

Still Great Businesses.

Models become commoditized.

Competition increases.

Everyone earns similar returns.

Capex continues rising.

Margins compress.

5–6/10

Good Businesses.

Not Great Businesses.

Massive spending.

Little differentiation.

Price competition.

Weak returns on capital.

The industry spends enormous amounts merely to maintain market share.

2–4/10

Gruesome Businesses.

Google (10 → 7.5)
Search remains dominant, but AI introduces the possibility that answers replace clicks. Simultaneously, inference costs rise dramatically. Google may spend far more capital to generate the same economics that search once produced almost effortlessly.

Meta (9 → 8)
Meta's distribution advantage remains extraordinary, but AI lowers content creation barriers and increases competition for attention. The company now spends tens of billions annually to remain at the frontier.

Microsoft (10 → 8.5)
Microsoft retains enterprise lock-in, but AI threatens to shift value away from applications and toward models and agents. Office remains sticky, but future workflows may not look like today's software.

Apple (10 → 9)
Apple's ecosystem remains the strongest in technology. However, if AI agents become the primary interface, some value may migrate from hardware toward software intelligence, reducing Apple's historical advantage.

Amazon AWS (9 → 8)
AWS may become one of AI's largest beneficiaries, but cloud infrastructure is becoming increasingly capital intensive. Future returns may depend on maintaining utilization and pricing power despite massive investment requirements.

NVIDIA (8 → 9)
NVIDIA is the rare company whose moat may actually strengthen. It currently occupies the most important bottleneck in the AI supply chain and enjoys extraordinary supplier power.

Tesla (6 → 6)
Tesla's AI opportunity is significant, but so is execution risk. The company remains heavily capital intensive and highly dependent on achieving autonomy at scale.

The AI debate may ultimately be less about technology and more about economics.

AI may be converting some of the greatest asset-light businesses ever created into more capital-intensive infrastructure businesses.

The bull case is that AI creates so much incremental economic activity that the returns justify the investment.

The bear case is that the MAG7 spend hundreds of billions defending existing profit pools while GDP continues growing only 2%–3%, causing AI to resemble telecom, utilities, or railroads rather than software.

That distinction—whether AI becomes a toll bridge or a commodity utility—may be the single most important valuation question in global equities over the next decade.