Have you ever watched a technology boom race ahead so fast that everyone starts treating sky-high valuations as the new normal? I have, more than once, and each time the same uneasy feeling settles in. Right now that feeling is back with the artificial intelligence buildout. Massive sums are pouring into data centers, specialized chips, and power infrastructure. Some voices already warn of widespread capital destruction. Yet looking closer at history, the pattern feels less like a glitch and more like a recurring feature of how progress actually happens under competitive markets.
The Familiar Cycle Behind Every Major Technology Leap
Innovation rarely arrives in neat, measured doses. Capital rushes in, expectations soar, capacity gets built far ahead of proven demand, and then the inevitable reckoning arrives. After the dust settles, the physical and technical assets that remain often power the next long stretch of growth at much lower cost. This sequence has played out with railroads, electrification, and the fiber networks of the late 1990s. The current wave of spending on artificial intelligence infrastructure shows many of the same hallmarks. Understanding the cycle does not remove the pain of the downturn, but it does change how one might position for what comes after.
In my view the most useful way to approach these episodes is to separate the financial outcomes from the economic ones. Equity and debt holders frequently suffer large losses. The broader economy, however, tends to inherit a cheaper, more abundant foundation for future productivity. That distinction matters when deciding how much exposure still makes sense at elevated prices.
Lessons From The Post Civil War Railroad Mania
After the Civil War the United States embarked on an extraordinary railroad expansion. European capital joined domestic funds and flowed into thousands of miles of new track. Many of those lines stretched into sparsely populated territory where near-term traffic could never justify the investment. Speculative financing and generous land grants kept the money coming. Early projects that did earn solid returns only encouraged riskier ventures farther ahead of demand.
The turning point arrived when bond markets simply stopped absorbing new issues. The failure of a major banking house trying to place Northern Pacific securities triggered broader panic. More than a hundred railroads entered bankruptcy in the years that followed. Banks failed, investors lost fortunes, and a multi-year depression set in. Equity holders were not the only ones hurt; bondholders absorbed heavy losses as well.
Yet the steel rails themselves did not vanish. Reorganized and recapitalized at far lower valuations, the surviving network became the transportation backbone of American industry. Goods, people, and capital moved more efficiently than ever before. The overbuilt system, once stripped of its inflated financial claims, delivered decades of genuine economic value. The paper losses were real and painful. The residual infrastructure proved far more useful to the wider economy than those losses suggested.
I find this episode particularly instructive because credit markets, not equity prices, delivered the decisive blow. Once lenders refused to fund further expansion, the party ended abruptly. That detail will become relevant again when we examine the present environment.
Electrification And The Holding Company Excesses Of The 1920s
A similar script unfolded during the great electrification boom of the 1920s. Central station utilities expanded rapidly. Holding companies layered on top of operating businesses, creating complex, leveraged structures that facilitated access to capital and technical know-how. Electricity generation and revenues climbed steeply as the technology shifted from novelty to essential infrastructure. Utility shares became a large and fashionable part of the stock market.
The structures, however, were opaque and capacity was added aggressively. When the 1929 crash arrived and the Depression deepened, the leverage and accounting practices stood exposed. One of the largest empires collapsed in spectacular fashion. Utility averages plunged. Regulatory reforms later broke up the biggest holding companies and steered the industry toward more stable regional monopolies.
What remained was a dramatically expanded electrical grid and generation capacity. Homes, factories, and entire industries transformed under the influence of cheap, abundant power. The financial destruction for equity and debt holders was severe. The long-term productivity gains from the physical system were enormous. Once again the overinvestment left behind an asset base that the economy put to productive use for the rest of the century.
One subtle difference from the railroad era is worth noting. Equity markets led the decline this time, with credit stress following later. Timing of the signals can vary, yet the underlying sequence of mania, overbuild, collapse, and eventual productive reuse stays consistent.
The Telecom Fiber Overbuild And The Birth Of Modern Internet Platforms
The late 1990s offer the most recent and closely comparable example. Carriers and equipment makers poured hundreds of billions into fiber optic networks, switches, and related gear. Much of the financing came from debt markets, and optimistic forecasts of internet traffic doubling every few months kept the capital flowing. Capacity raced far ahead of actual near-term demand.
After the equity bubble burst, market values collapsed. Many telecom companies filed for bankruptcy. Bandwidth prices fell sharply. Large stretches of newly laid fiber sat dark and unused. Through bankruptcies and subsequent recapitalizations, that physical infrastructure changed hands at deep discounts, often for pennies on the original dollar.
The resulting cheap broadband made business models viable that would have been prohibitively expensive under the old cost structure. Search engines, video platforms, social networks, smartphone ecosystems, streaming services, software delivered as a service, and cloud computing all scaled on that excess capacity. Without the prior overbuilding and the distress that followed, the second generation of internet services would have arrived more slowly and at higher cost.
Capital destruction was undeniable. Bondholders in many cases recovered only about twenty cents on the dollar. Yet the residual infrastructure seeded an entire wave of growth. The pattern held once more.
Where The Current AI Buildout Stands In The Cycle
Today we appear to sit squarely in the overinvestment phase of the artificial intelligence infrastructure cycle. Capital is flooding into data centers, advanced chips, power generation and transmission, and related equipment. The sheer scale of spending is turning advanced models into something closer to a commodity faster than many investors seem prepared for. Lower-cost open-source alternatives, both from overseas and from Western developers closing the performance gap, are already exerting downward pressure on pricing.
Several signals suggest the late stages of the party. Private credit funding for portions of the buildout has shown signs of stalling. Enterprise demand for the promised productivity leaps has not yet materialized at the scale needed to justify every contracted commitment. Power constraints are emerging as a tighter binding limit than chip supply in some regions. Circular financing arrangements, in which companies with negative free cash flow rely on continued debt-funded purchases of expensive hardware, have drawn increasing commentary. Analyses have appeared showing that certain major players may struggle to cover long-term compute contracts under more realistic revenue assumptions.
Narrative shifts toward speculative solutions such as space-based data centers sometimes feel like attempts to paper over terrestrial bottlenecks rather than genuine near-term answers. Growth rates embedded in current valuations look vulnerable once the second derivative of funding and actual utilization begins to slow. In short, the question is no longer whether a meaningful correction arrives but when and how severe it proves.
I have found it useful to watch the credit side of the market especially closely. Private credit has already shown hesitation. Off-balance-sheet structures are growing. Early stress has appeared in the lowest-rated segments of the junk market. Given the large presence of passive equity vehicles, it is at least plausible that credit markets could lead this time while equity prices lag the deteriorating fundamentals.
Why Credit Markets Tend To End These Buildouts
Across the historical cases, debt played a central role in funding the expansion and, later, in marking its end. During the railroad boom, bonds often supplied more than three-quarters of the capital. When new issues could no longer be placed, the credit system seized. Bank runs followed, the stock exchange closed temporarily, and within a couple of years roughly a third of outstanding railroad bonds had stopped paying interest.
In the utility expansion of the 1920s, layered debt and bank loans secured by the securities of the operating companies themselves created fragile pyramids. Once revenues softened and banks refused to roll the notes, the structures collapsed into receivership.
The telecom fiber episode saw carriers raise enormous sums in the bond market and lean on vendor financing and circular arrangements. Credit spreads eventually widened sharply, defaults clustered, and recoveries for bondholders were often modest. When lenders stopped funding incremental capacity, the overinvestment phase ended and the financial cleanup began.
The same dynamic is visible now. Whether equity or credit delivers the first decisive signal remains open, yet the historical record suggests that the willingness of lenders to keep extending capital is the true limiting factor on further overbuilding.
Bubbles As A Feature Rather Than A Flaw
Human nature has not changed. Periods of genuine technological excitement attract capital in volumes that eventually exceed prudent near-term needs. The resulting overinvestment produces financial losses and distressed assets. Those assets, once recapitalized at lower valuations, frequently become the cheap foundation for the next wave of economic activity.
This process destroys capital in the short run. Over a longer horizon it often leaves behind physical and technical capacity that the economy would not otherwise possess at comparable cost. The long-term outcome for productivity and living standards has repeatedly proven more favorable than the peak-to-trough equity charts imply.
What has been will be again, what has been done will be done again; there is nothing new under the sun.
Those who accept the cyclical nature of these episodes may choose to reduce exposure where risk-reward has become unattractive. When the inevitable bust arrives, the same observers can look for opportunities among the durable assets that emerge from the wreckage. Timing the exact peak is notoriously difficult. Recognizing the phase of the cycle is more realistic and, in my experience, more useful.
Practical Implications For Capital Allocation Today
No one can say with certainty how large the eventual write-downs will be or how long the adjustment will last. What history does suggest is that the physical data centers, power infrastructure, and specialized equipment being installed today will not simply disappear. Some portion will change hands at lower valuations. Utilization rates that look disappointing at current capital costs may look far more attractive once the balance sheets have been cleaned up.
Open-source models and competitive pressure from multiple regions are likely to keep compressing margins on pure model development. The durable value may reside more in the underlying infrastructure and in applications that can exploit abundant, low-cost compute. Power availability already looks like a critical constraint in several markets. Solutions that address energy supply and transmission could therefore retain strategic importance even after a broader correction.
I tend to favor a measured approach. Reducing concentrated exposure to the most richly valued names while remaining open to opportunistic purchases of high-quality assets after a period of distress has served me better than trying to call exact tops or bottoms. The goal is not to avoid every loss but to participate in the longer-term productivity gains that these cycles have historically delivered.
Perhaps the most interesting aspect is how consistently the pattern repeats despite the different technologies involved. Railroads, electricity, fiber, and now artificial intelligence each seemed unique in their moment. The capital dynamics proved remarkably similar. Recognizing that similarity does not eliminate risk, yet it does provide a framework for interpreting the noise of the present moment.
Preparing For The Next Phase Without Predicting Exact Timing
Markets can remain elevated longer than sober analysis suggests. Liquidity, momentum, and narrative can sustain valuations well after the underlying fundamentals have begun to deteriorate. That reality argues against aggressive short positions for most investors. At the same time, continuing to add capital at peak optimism carries its own dangers.
A balanced stance might include gradual reduction of the most expensive exposures, careful attention to credit market signals, and the mental readiness to deploy capital when high-quality assets become available at more reasonable prices. The history of previous infrastructure cycles shows that the greatest long-term returns often accrue to those who can buy the residual capacity after the mania has ended rather than during its peak.
Power constraints, utilization shortfalls, and competitive pricing pressure are already visible. Circular financing arrangements add another layer of fragility. None of these factors guarantees an immediate collapse. Together they raise the probability that the current phase of unconstrained overinvestment is closer to its end than to its beginning.
In the end the lesson remains straightforward. Bubbles are not an aberration in market economies. They are a recurring mechanism through which ambitious infrastructure gets built, financed, written down, and then put to productive use. The short-term cost falls on those who provide capital at the most optimistic prices. The longer-term benefit often accrues to the broader economy and to those positioned to acquire the resulting assets on more favorable terms.
Watching the present cycle unfold with that perspective in mind makes the volatility less surprising and the eventual opportunities clearer. The physical foundation being laid today will likely matter long after today’s valuations have been adjusted. That, more than any single forecast, is the practical takeaway from more than a century of similar episodes.
The current enthusiasm around artificial intelligence is understandable. The technology is genuinely transformative. Capital is responding exactly as it has in every prior transformative wave. The overbuild that follows is equally predictable. Accepting both sides of the equation allows for clearer decisions about exposure, timing, and the search for value once the inevitable recalibration begins.
History does not repeat with perfect precision, yet the broad contours of these capital cycles have proven stubbornly persistent. Recognizing the pattern does not require perfect foresight. It simply requires the humility to acknowledge that human nature, competitive dynamics, and the incentives of capital markets have not changed as much as the underlying technologies themselves.
For those willing to look past the immediate excitement, the coming years may offer both cautionary lessons and genuine opportunity. The infrastructure being created at great cost today will almost certainly find productive uses tomorrow. The only open questions are the price at which that capacity changes hands and the patience required to wait for that moment.
