*****************************************************************************

*                       FIEND'S SUPERBEAR MARKET REPORT                     *

*                                October 20, 2026                           *

*                                                                           *

*                       e-mail: fiendbear@fiendbear.com                     *

*                    web address: http://www.fiendbear.com                  *

*****************************************************************************

Fiend Commentary
================

The Bubble That Depreciates

Technology stocks bounced back Monday following last week's bruising selloff. The Nasdaq gained 1.1%, the S&P 500 rose 0.4%, and the semiconductor index finally snapped an eight-session losing streak. The Dow slipped slightly, while market breadth remained negative.

Investors apparently decided that concerns about the enormous cost of artificial intelligence had gone far enough.

Perhaps. But there is a fundamental difference between the AI boom and many of the great investment bubbles of the past, and it deserves considerably more attention.

Much of the infrastructure being built today could become economically obsolete long before investors recover what they paid for it.

A Different Kind of Infrastructure

History is filled with speculative booms surrounding revolutionary technology.

The canal mania of the early nineteenth century produced financial disasters, but many canals continued transporting goods long afterward. Railroad speculation bankrupted countless investors, yet the tracks helped build modern economies. Electricity required enormous investments in generating plants and transmission networks that remained useful for decades.

The internet boom followed a similar pattern. Investors poured money into fiber-optic networks, much of which went unused when the bubble burst. But the fiber remained in the ground and eventually became valuable infrastructure for the digital economy.

In each case, investors could lose fortunes while society retained something enormously useful.

AI may produce the same long-term economic transformation, but a substantial portion of today's investment has a much shorter useful life.

A railroad bridge can last a century.

A generating plant can operate for decades.

A state-of-the-art AI processor may lose much of its competitive value within a few years.

That distinction could prove enormously expensive.

Moore's Law Meets the Balance Sheet

For decades, improvements in semiconductor technology have made computing dramatically more powerful and generally cheaper per unit of work. Moore's law originally described the growth in transistor density, not a guaranteed halving of processor or memory prices every two years. Nevertheless, the relentless improvement in computing efficiency is undeniable.

AI hardware has an additional problem: new chip architectures, specialized accelerators and increasingly efficient software can make existing equipment less economical even before it physically wears out.

A processor purchased at an enormous premium today may still operate perfectly five years from now. But if newer equipment performs the same calculations at a fraction of the electricity and operating cost, the older machine becomes much less valuable.

This doesn't mean every existing processor becomes worthless. Older GPUs can continue handling less demanding workloads for years. Buildings, power connections and cooling systems can also remain useful through several generations of equipment.

The real danger is economic depreciation, not physical deterioration.

The accounting assumptions are revealing. Microsoft, Alphabet and several other major technology companies generally depreciate server equipment over roughly six years. Meta uses approximately five and a half years.

What happens if the economic value of some of that equipment deteriorates considerably faster?

Depreciation expenses or asset write-downs could consume a meaningful portion of the profits investors are currently extrapolating into the future.

Microsoft's annual depreciation expense has already risen from approximately $22 billion to $34 billion in just one fiscal year.

And the investment boom is nowhere near finished.

The Spending Has to Keep Coming

UBS estimated that major hyperscalers would spend approximately $673 billion on capital expenditures in 2026, an increase of 76% from the previous year.

That spending is wonderful for chipmakers and equipment suppliers.

But much of it creates the need for more spending later.

Processors need replacement. Power-hungry systems need upgrades. New software may require different hardware. Competition forces operators to keep improving performance and lowering costs.

Meanwhile, investors expect the companies making these investments to earn substantial returns on their original expenditures.

This is where the mathematics can become unpleasant.

A company spending $100 billion on equipment that has a five-year economic life needs approximately $20 billion in annual earnings just to recover that investment before financing costs and other expenses, assuming no residual value.

And five years later, the replacement cycle begins again.

Strong demand for AI computing could make that investment worthwhile. The question is whether revenue and margins will grow fast enough.

Borrowing at today's interest rates makes the calculation even more demanding.

We are already seeing concern about highly leveraged AI infrastructure companies such as CoreWeave, whose financial obligations have become a growing focus for investors.

The danger isn't necessarily that AI demand disappears.

It is that computing becomes cheaper and more abundant faster than the companies financing today's infrastructure can earn adequate returns.

That would be excellent for users of AI.

It could be disastrous for some of its early investors.

Forty Percent of the S&P 500

There is another problem that makes this particular boom dangerous for ordinary investors.

The ten largest companies now account for approximately 40% of the S&P 500.

Not all are pure AI companies, but technology and AI-related businesses dominate the group.

That concentration creates a situation few investors are accustomed to considering.

Suppose several years from now the AI investment boom has run its course.

Computing costs have plunged, AI services have become inexpensive commodities, and profit margins at some of today's largest beneficiaries have contracted sharply.

Imagine the top ten stocks collectively losing 50% of their value while the remaining companies in the S&P 500 gain 20%.

Using today's approximate index weights, the result would still be an 8% decline in the S&P 500, despite substantial gains across the rest of the index.

That is an illustrative calculation, not a forecast. But it exposes the vulnerability.

The average stock could be doing reasonably well while the headline index remains trapped in an extended bear market.

And an index investor who believes an S&P 500 fund provides broad diversification may discover how heavily the portfolio depends on a handful of companies.

The Revolution Can Succeed While Investors Lose

There is a historical irony here.

The railroads transformed transportation, but many railroad shareholders were ruined.

The internet transformed commerce and communications, but numerous technology stocks never recovered their bubble valuations.

AI could follow the same path.

It might become indispensable to business, medicine, engineering and everyday life. Computing could grow dramatically more efficient while the cost of using AI continues falling.

Those developments would be tremendous achievements.

They would not necessarily justify the prices investors are paying today for companies whose profits depend on continued extraordinary capital spending and unusually high margins.

Monday's technology rebound showed that investors remain willing to buy the AI story after a sharp correction.

It did not resolve the larger problem.

The industry's leaders are spending hundreds of billions to build computing capacity that future generations of technology may make dramatically cheaper.

Some will earn extraordinary returns. Others may discover that the equipment depreciated faster than the profits arrived.

And because these companies now dominate the stock averages, their eventual performance could determine the direction of the S&P 500 for years.

The technology may last forever. The investment might not.

A few research details strengthen this argument:

  • The 40% concentration is real: S&P 500 ETF holdings dated October 15 put the top ten at approximately 40.1% of the index.
  • Depreciation deserves particular emphasis: Microsoft's 2026 annual report shows depreciation increasing from $22.0 billion to $34.3 billion. Server-life assumptions generally run around five to six years across major operators.
  • There is an important counterargument: Nvidia A100 chips introduced in 2020 are still being contracted for use through 2029. So obsolescence is not automatic; the risk is that revenue or resale values fall faster than anticipated.

The other point worth watching is UBS's projection that hyperscaler capital-expenditure growth could slow from 76% this year to just 6% in 2028. Even without an outright spending collapse, that deceleration could cause a major earnings and valuation adjustment among AI suppliers.

 

Weekly Market Summary Page
[Return to the Fiend's SuperBear Page]