When intelligence became infrastructure: Is AI the greatest bet in human history or the next Great Bubble?
On June 25, 2026, something happened that most Apple followers never imagined they would witness. Apple quietly increased prices across several existing products in the middle of a product cycle. There was no redesigned MacBook, no revolutionary iPad, no surprise keynote. Just higher prices. The explanation was even more surprising than the announcement itself. Apple pointed toward soaring memory chip prices, driven primarily by the unprecedented demand created by artificial intelligence infrastructure.
For decades, technology followed a familiar trajectory.
Better chips became cheaper every year. Consumers expected more performance at
lower prices. Moore's Law conditioned an entire generation to believe that
computing only gets faster and more affordable. AI has temporarily reversed
that assumption. This seemingly simple event serves as an entry point into a
much larger question: Are we witnessing the greatest technological investment
in human history, or are we watching history's next great capital bubble being
inflated in real time?
Does this predict doom or ask an uncomfortable question that
many investors have been avoiding: What if AI is unquestionably revolutionary,
yet still massively overpriced? That distinction matters. History teaches us
that transformative technologies and investment bubbles are not mutually
exclusive. In fact, they often arrive together.
One of the most striking aspects that caught my eye is the
explosion in capital expenditure by the world's largest technology companies. Before
ChatGPT existed, Amazon, Microsoft, Google, and Meta together spent roughly $90
billion annually on capital investments. Within six years, that figure has
reportedly grown to approximately $725 billion.
Pause for a moment and appreciate what that means. This is
not marketing expenditure. This is not research spending. This is physical
infrastructure. Entire cities worth of computing power are being constructed
every year. The modern AI race isn't primarily about algorithms anymore. It is
about concrete, electricity, fiber optics, transformers, cooling systems,
memory chips, networking equipment, and thousands upon thousands of GPUs housed
inside gigantic industrial buildings that resemble modern factories more than
software companies.
Every ChatGPT conversation, every Gemini response, every
Claude analysis begins inside one of these enormous data centers. When someone
asks an AI model to summarize a document, it may feel like software.
Economically, however, it resembles operating one of the world's largest
industrial machines. That changes everything.
One of the biggest misconceptions surrounding AI is that
software companies simply write code and collect subscription fees. Reality is
much more expensive. Training frontier AI models now requires tens of thousands
and increasingly hundreds of thousands of advanced GPUs connected by ultra-fast
networking. Each GPU may cost between $30,000 and $40,000 before considering
installation.
Then comes networking, Power distribution, Liquid cooling, Backup
generators, Land acquisition, Construction, Security and Maintenance. The
result is a data center whose cost can rival that of a major airport terminal.
Let’s shift attention away from AI itself and toward capital allocation. Every
dollar spent on infrastructure represents an expectation that future demand
will justify today's investment. And expectations can be dangerously
optimistic.
The framework is remarkably simple. High profits attract investment. Investment creates more supply. Eventually supply exceeds demand. Returns collapse. Only the strongest companies survive. This cycle has repeated throughout economic history.
- Railways
- Oil
- Shipping
- Telecommunications
- Semiconductors
- Real estate
None of these industries escaped the laws of economics
simply because they were important. AI may not either.
The comparison with the fiber-optic boom of the late 1990s
is particularly compelling not because history repeats perfectly, but because
human behavior often does. During the internet revolution, investors were
absolutely correct about one thing. The internet was going to change
civilization. They simply underestimated one variable - Timing. Billions of
dollars flowed into laying fiber-optic cables across America. Companies
borrowed enormous amounts of money. Valuations skyrocketed. Many firms achieved
market capitalizations larger than century-old industrial giants despite
generating almost no profit.
Then reality intervened. Demand grew rapidly but not rapidly
enough. Infrastructure was built faster than customers could use it. Bandwidth
prices collapsed. Telecom companies failed. The dot-com crash followed. Yet
here's the fascinating part. Those fiber cables never disappeared. They
remained underground. Years later, YouTube emerged. Netflix appeared.
Cloud computing became mainstream. Smartphones exploded. The
infrastructure eventually became indispensable. The investors who bought during
peak euphoria lost fortunes. Society, however, inherited the backbone of the
modern internet. That historical lesson deserves attention because it
illustrates something subtle:
Technological success does not guarantee investment
success.
The internet changed humanity. Thousands of internet
companies still disappeared. Both statements are simultaneously true.
AI could follow a remarkably similar trajectory. Even if AI
becomes as important as electricity, today's valuations still need to be
economically justified. This distinction often disappears in public debates. People
assume skepticism toward AI valuations equals skepticism toward AI itself. It
doesn't. One concerns technology. The other concerns price. Warren Buffett has
often observed that a wonderful company can become a terrible investment if
purchased at an unreasonable price. The same principle may apply to AI. No
serious analyst doubts that artificial intelligence will reshape industries. The
real disagreement concerns how much investors should pay today for profits that
may not fully materialize for another decade.
The strongest point is the gap between spending and monetization. According to the figures cited, major AI companies collectively generate roughly tens of billions in annual AI revenue while the broader ecosystem is investing several hundred billion dollars into infrastructure. Even if one debates the exact numbers and they certainly deserve scrutiny, the broader concern is legitimate. Infrastructure spending has accelerated faster than revenue realization. Economists call this the monetization gap. Technology adoption is often exponential. Business adoption rarely is. Companies must redesign workflows. Employees require training. Regulatory uncertainty slows deployment. Cybersecurity concerns emerge. Legacy software resists integration. The result is that technological capability often races ahead while commercial adoption walks. This is precisely where many investors become nervous.
Perhaps the most surprising insight is from enterprise
customers. The original AI investment thesis assumed businesses would gladly
pay increasing amounts for AI services because productivity gains would more
than justify the cost. Reality appears more complicated. Several enterprises
have begun aggressively optimizing AI expenses. Some startups reportedly
switched providers and reduced inference costs dramatically. Others are
experimenting with smaller open-source models rather than premium frontier
models.
This mirrors something that happened during the early cloud
computing era. Initially, businesses rented enormous server capacity because it
was convenient. Eventually they became sophisticated enough to optimize
workloads, negotiate pricing, and reduce waste. AI may be entering a similar
optimization phase much sooner than investors expected.
The irony is that falling AI costs create both winners and losers. Users celebrate cheaper inference. Businesses enjoy lower operating expenses. Innovation accelerates. Yet infrastructure providers may see lower margins precisely when they need higher revenues to justify record capital expenditure. That tension sits at the heart of today's AI debate. Not whether intelligence is valuable. But whether intelligence can become profitable quickly enough.
#ArtificialIntelligence #AI #ThinkSchool #NVIDIA #OpenAI #Microsoft #Google #Meta #Amazon #DataCenters #Investing #Technology #Economics #CapitalMarkets #Innovation #DigitalTransformation