Today's AI infrastructure building frenzy is like the historical cycle of the railway building frenzy 150 years ago.
The data center being built by OpenAI
In 1865, just as the smoke of the American Civil War was clearing, new ambitions stirred on the war - torn land. In an office on Broadway in New York, Cornelius Vanderbilt, in his seventies, was drawing thick black lines on the map. These lines would connect the scattered railways into the transportation empire of his dreams.
One and a half centuries later, on a certain night in early 2023, Sam Altman presented a grander blueprint to investors in the OpenAI conference room: a plan to raise $7 trillion to build an AI chip empire, a figure that even exceeds the GDP of most countries.
These two scenes across time and space outline the astonishing similarities in the capital cycle during technological revolutions -
The railway and AI, two infrastructure construction booms separated by 150 years, are playing out almost the same script: Technological breakthroughs ignite imagination, capital influx spurs prosperity, over - construction leads to oversupply, and finally, a reshuffle occurs in the burst of the bubble.
01 The Railway Great Leap Forward: The Rise and Pains of the Iron Arteries
The railway construction after the end of the American Civil War can be regarded as the first large - scale infrastructure construction boom in human history.
At that time, the achievements of the First Industrial Revolution were being rapidly applied. Amid the roar of steam engines, railway tracks spread like a spider's web, and investors flocked in, dreaming of reshaping the economic geography of the North American continent.
Between 1865 and 1873, an average of 20 miles of new railway tracks were laid in the United States every day.
The microscopic details of this construction boom reveal the true nature of capital frenzy. In a small town in Kansas, a journalist recorded such a scene: "Ten trains full of railway tracks, spikes, and workers arrive every day. They are like army ants gnawing at the prairie. Irish laborers wave hammers under the scorching sun, and Chinese coolies work hanging on the cliff edges. Their tents stretch for miles, and the campfires at night are like falling stars."
Behind this construction speed was the huge subsidy from the federal government. The Pacific Railway Act not only provided government loans of $16,000 to $48,000 per mile but also granted railway companies 6,400 acres of land per mile along the railway line.
The Union Pacific Railroad ultimately received 12 million acres of land, and the Central Pacific Railroad received 9 million acres - equivalent to the combined area of Massachusetts, Connecticut, and Rhode Island.
It is estimated that railway investment accounted for 7% - 10% of GDP at its peak, equivalent to trillions of dollars today.
In this era of wild growth, a group of railway tycoons stepped onto the historical stage with courage, shrewdness, and luck.
Vanderbilt, the "Captain" who started as a Staten Island ferryman, decided to enter the railway industry at the age of 70. He adopted the classic "encirclement" tactic: when the rival Erie Railroad refused to be acquired, he immediately built a new railway on a parallel line and cut the freight rates in half until the other side surrendered.
By 1873, Vanderbilt controlled a continuous railway line from New York to Chicago, with a total length of over 1,100 miles.
Even more legendary is Jay Gould. This thin, often - coughing speculative genius turned his attention to the railway after triggering the "Black Friday" panic in 1869 by attempting to monopolize the gold market.
He controlled railway companies by issuing "watered - down stocks" - stocks without real asset support. He once manipulated the stock prices of 12 railway companies simultaneously in one day, and his operation methods were so complex that even his brokers could hardly fully understand them.
During this period, a large amount of British capital also poured in. By 1873, British investors held nearly 30% of American railway bonds.
However, beneath the prosperity, a crisis was quietly brewing. In Kansas, three parallel railway lines competed for scarce freight resources; in Minnesota, a railway leading to a "ghost town" had only one train per day, and the carriages were empty.
By early 1873, the idle rate of the total railway capacity in the United States had exceeded 30%, but new railway tracks were still being extended.
...
On September 18, 1873, Jay Cooke & Company, which dominated the financing of the Northern Pacific Railroad, declared bankruptcy, triggering one of the most severe economic crises in American history. When it could no longer sell more railway bonds, the capital chain finally broke.
Panic spread like a plague. The New York Stock Exchange closed for the first time in its history, lasting for 10 days. In the subsequent chain reaction, 89 railway companies went bankrupt, 18,000 enterprises closed down, the unemployment rate soared to 14%, and the economy plunged into a 65 - month "Great Depression."
The essence of the crisis was the fatal mismatch between over - investment and insufficient demand.
From 1865 to 1873, the total investment in American railways reached $2 billion, while the total national income of the United States during the same period was about $9 billion - more than one - fifth of the national savings was invested in railway construction.
However, the growth of freight revenue far lagged behind the extension of railway tracks. The railway freight rate dropped from 2.5 cents per ton - mile in 1870 to 1.5 cents in 1875, a decrease of 40%. The return on equity for shareholders plummeted from the expected 15% to less than 5%, and the bonds of hundreds of railway companies became worthless.
Ironically, the burst of the bubble marked the birth of a new era: although the railway destroyed many investors, it brought profound changes to the American economy. The cost of transporting wheat from Kansas to New York by railway dropped from 50 cents per bushel to 10 cents, which enabled American wheat to be exported to Europe at a low price, changing the global grain trade pattern.
In Chicago, Philip Armour used the railway network to build a modern meat - processing empire. The refrigerated carriages he invented could transport beef from Chicago to New York without spoilage. By 1880, Armour's factory processed 15,000 livestock per day, becoming the world's largest meat supplier.
02 The AI Infrastructure Construction Boom: A New Script 150 Years Later
If the core of infrastructure construction in the 19th century was railway tracks and steam locomotives, then in the 21st century, it is undoubtedly data centers and AI chips.
At the NVIDIA headquarters, engineers created the modern - day "steam locomotive" - the H100 GPU. This chip has 80 billion transistors, and its efficiency in training large - language models is 9 times higher than that of its predecessors. Its price is equally astonishing - each costs up to $30,000, but it is still in short supply.
It is the hyperscale data centers that are building the "railway tracks" for these "digital locomotives."
In the desert of Arizona, Meta is building a data center park covering 2.7 million square feet. Its power consumption will exceed the total household power consumption of 500,000 nearby residents. Microsoft is investing $3.3 billion in building a data center in Wisconsin and has received decades of tax breaks from the local government - which is similar to the land grants given to railway companies back then.
In just three years, the scale of AI investment in the United States has reached an unprecedented historical level. According to industry analysis, the planned capital expenditure for global AI data centers in the next five years is estimated to reach an astonishing $4 trillion.
More fundamentally, the short life - cycle of AI chips exacerbates capital consumption. Compared with the 30 - year service life of railway tracks, the effective life of an AI GPU is only 3 - 5 years. This means that AI infrastructure requires continuous reinvestment, creating a "capital black hole."
Historical data shows that companies with high capital expenditure growth often underperform their capital - conservative peers in the long run. Since 1963, the annual average return of the portfolio of companies with the fastest asset growth has been 8.4 percentage points lower than that of the portfolio of companies with the slowest asset growth. This rule was already evident in the railway era: In the 1870s, the average return on invested capital of railway companies dropped below 5%, far lower than investors' expectations.
The rule is clear, but no one can stay out of it. Contemporary AI leading enterprises are caught in a typical "Prisoner's Dilemma."
Satya Nadella, the CEO of Microsoft, privately told the board of directors: "We may have over - invested by $200 billion, but the risk of not investing is even greater." Google executives also admitted that they had to match the investment scale of their competitors, even though they knew it might lead to industry - wide overcapacity. Mark Zuckerberg's remarks at the Q4 2023 earnings conference were quite representative: "If we end up misusing hundreds of billions of dollars, I think it will be very unfortunate, but the risk of missing the AI era is even higher."
This mindset is exactly the same as that of the 19th - century railway tycoons. Vanderbilt once said when building Grand Central Terminal in New York: "If they are all building, I have to build bigger." As a result, by 1900, New York City had three grand railway stations belonging to different companies. This redundant construction greatly increased the cost.
Historical experience shows that the biggest beneficiaries of infrastructure construction are often not the builders but the users. In the railway era, although many railway companies went bankrupt, the railway infrastructure gave birth to the modern logistics system. The railway increased the efficiency of American manufacturing by about 25%, and created a national market for mail - order enterprises such as Sears, Roebuck and Co.
In the AI era, a similar pattern is emerging.
Siemens has used AI to optimize its global supply chain, reducing inventory costs by 18%. Merck Group uses AI to accelerate drug discovery, shortening the time of some R & D stages from years to months. These enterprises do not need to invest hundreds of billions of dollars in building AI infrastructure but can enjoy the efficiency improvement brought by AI.
03 Lessons from History: Signals to Identify the Cycle Stages
From the railway boom 150 years ago, we can summarize several key signals to identify the cycle stages:
Characteristics of the Frenzy Period: The proportion of capital expenditure in GDP is extremely high (7% - 10% in the railway era); a large number of new entrants emerge (hundreds of new railway companies were established in the 1870s); the leverage ratio rises rapidly (the debt - to - equity ratio of railway companies exceeded 3:1).
Signals of the Turning Point: The capacity utilization rate decreases (the idle rate of railway capacity exceeds 30%); price competition intensifies (freight rates continue to decline); the financing environment tightens (bond issuance becomes difficult).
Current AI investment already shows the characteristics of the frenzy period but has not yet reached the decisive turning point. Closely observing the utilization rate of data centers and the changes in AI service prices will be the key indicators for judging the position of the cycle.
In the infrastructure construction boom, the transfer of value follows a predictable path:
First Stage: Value concentrates on equipment suppliers (steel mills in the railway era, chip manufacturers in the AI era);
Second Stage: Value concentrates on the most efficient operators (large railway companies in the railway era, cloud service providers in the AI era);
Third Stage: Value spreads to users (logistics enterprises and manufacturers in the railway era, users in various industries in the AI era).
Currently, AI investment is still in the transition period from the first stage to the second stage.
04 Conclusion: We Are at a Similar Crossroads Again
The capital cycle theory, systematically expounded by economist Edward Chancellor in Capital Returns, reveals the fate of prosperity: new technologies trigger investment booms, supply floods in and chases demand, often outpacing it, and finally leading to oversupply, price collapse, and a long - term adjustment.
150 years ago, the railway Great Leap Forward was a classic example of this theory. 150 years later, we are at a similar crossroads again.
AI will transform the world like the railway, connecting intelligence and driving efficiency improvement. However, if the trillions of dollars in capital expenditure are not matched by demand, it may lead to overcapacity, just like the idle railway tracks in the 1870s.
From Vanderbilt to Altman, from railway tracks to chips, technology is constantly evolving, but the human psychology, capital logic, and market laws driving these infrastructure construction booms are so similar.
After experiencing the painful burst of the bubble and industry restructuring, the railway boom finally shaped the modern American economy. By 1900, the cost of railway travel dropped to one - tenth of that in 1870, and the freight efficiency increased five - fold, truly unleashing its economic potential.
Similarly, today's AI infrastructure construction boom will surely unleash the full potential of the intelligent era after necessary adjustments and integration. It's just that we still don't know when that day will come.
This article is written based on public information and is only for information exchange purposes and does not constitute any investment advice.
This article is from the WeChat official account "Jinduan Research Institute." Author: Mu Yang. Republished by 36Kr with permission.