US AI Infrastructure Investment Surpasses Railroads and Canals, Becoming the Largest Bet in History
The Brookings Institution released the latest estimates from economist Stijn van Nieuwerburgh: from 2025 to 2032, total investment in data centers and related AI infrastructure in the US is expected to reach $10.3 trillion, averaging about 3.6% of GDP per year, surpassing historical infrastructure waves such as canals, railroads, and power grids. Goldman Sachs estimates that AI investment in the US will account for 1.9% of GDP by 2026. The last time a single emerging industry had a higher share in the economy dates back to the railroad boom of the late 19th century. The US economy has never been so reliant on the expansion of a single industry.
The construction industry is the most direct beneficiary and also the area where the crowding-out effect is most evident. Data from the US Department of Commerce shows that in the first seven months of this year, seasonally adjusted private spending on data center construction was $37 billion, about $9 billion more than the same period last year; other private construction expenditures, such as residential, apartment, and shopping center projects, decreased by about $46 billion compared to the same period last year. A report from the Richmond Fed states that data center construction is crowding out local labor. Last year, Mississippi was once poised to secure an aluminum smelter, which would have brought about 1,000 permanent jobs; however, the announcement of a data center construction near a potential site for the smelter consumed the power needed for the smelter, and the project ultimately moved to Oklahoma. Site consultant Didi Caldwell bluntly stated that data centers are "crowding out manufacturing," not only competing for electricity but also driving up land prices.
Capital expenditures are increasingly relying on debt. Analyst forecasts compiled by FactSet show that five major cloud providers—Alphabet, Amazon, Meta, Microsoft, and Oracle—are expected to reach $4.2 trillion in capital expenditures over the next four years ending in 2029, with the proportion financed by debt continuing to rise. Van Nieuwerburgh pointed out that tech companies often use off-balance-sheet entities to borrow from banks and private credit institutions, and these transactions are rarely disclosed, making it difficult for outsiders to assess the real risk exposure. If AI revenues are insufficient to repay the debts incurred for building data centers, the impact could ripple through the financial system.
There is a noticeable high-salary differentiation in employment. LinkedIn estimates that since 2023, AI has driven the creation of over 750,000 new jobs in the US; the median salary for AI-related positions is about $180,000, while the median for all positions is about $80,000. Since 2024, data centers have added 117,000 operational positions, excluding a large number of temporary construction jobs. In the Washington D.C. area, the number of union electricians in IBEW Local 26 has increased from 9,000 to 17,500. A 23-year-old with a cybersecurity degree switched from an IT entry-level position to become an electrician due to concerns about being replaced by AI, earning about $30 per hour as an apprentice; a 28-year-old senior electrician supplying prefabricated components to Amazon's data center site earns $62 per hour plus overtime, often working 58 hours a week.
The wealth effect is amplifying consumption. Data from the Federal Reserve shows that as of the second quarter, US households held $63 trillion in stocks and mutual funds, nearly double the amount at the end of 2022. In the absence of real wage growth, stock market wealth has supported consumer spending, with gains primarily concentrated among high-net-worth individuals. Home sales across the US have been sluggish for four years, yet the luxury market in Silicon Valley is experiencing its best performance since 2000: a five-bedroom home listed at $9.9 million received seven offers and was contracted for over $13 million within two weeks, with the buyer being an AI entrepreneur.
In market mechanisms, the buyers are the five major cloud providers and their off-balance-sheet financing entities, with funds coming from corporate bonds, bank loans, and private credit; the sellers and beneficiaries are chip and storage suppliers, electrical equipment manufacturers, construction contractors, and skilled tradespeople such as electricians. Funds flow from bond markets and stock market valuations into the land, electricity, and equipment for data centers, and then back into consumption through high-paying jobs and stock market wealth effects. The pressured parties include the manufacturing sector losing projects due to occupied electricity and land, the residential and commercial real estate sectors facing reduced spending, and ordinary households dealing with rising electricity costs and mortgage rates.
Inflationary pressures have already emerged. Due to the demand for data center equipment such as storage chips, the prices of computers, peripherals, and semiconductors imported into the US rose by 20% year-on-year in August, which has been passed on to consumer goods like iPhones and gaming consoles. Chicago Fed President Goolsbee warned that data center investment is driving up wages in related fields; Federal Reserve Chair Waller identified the large-scale borrowing by major cloud providers as one of the reasons for rising long-term interest rates, which undermines the home-buying ability of millions of Americans. In areas dense with data centers, electricity prices have already surged.
Source: Public Information
ABAB AI Insight
The main players in this round of investment have a history of aggressive bets. In January 2025, OpenAI, SoftBank, and Oracle announced the "Star Gate" plan, aiming to invest $500 billion; in September of the same year, Oracle signed a power contract worth about $300 billion with OpenAI, subsequently issuing about $18 billion in bonds, with credit default swap prices rising significantly, and the market began to price Oracle's debt repayment ability. Meta financed about $27 billion for its Hyperion data center in Louisiana through a joint venture with Blue Owl, with the debt not counted on Meta's balance sheet, serving as a typical example of "off-balance-sheet financing." Microsoft has cumulatively invested about $13 billion in OpenAI since 2019, exchanging equity for binding computational demand.
The funding path shows a clear characteristic of "circular financing." Nvidia announced it would invest up to $100 billion in OpenAI, which in turn uses this money to purchase Nvidia chips; AMD issued warrants to OpenAI in exchange for procurement commitments; new cloud providers like CoreWeave expand by borrowing against GPUs, with customers also being their shareholders. Meanwhile, Meta issued $30 billion in bonds in October 2025, with order sizes exceeding $120 billion, and the bond market's appetite for tech giants has fueled this round of expansion. The underlying motivation is game-theoretic: the risk of under-investing is total exit, while the risk of over-investing is merely a decline in return rates, and no major player dares to stop.
The most direct comparison is with the railroad boom of the late 19th century and the fiber optic boom of the late 1990s. In 1873, Jay Cooke & Company went bankrupt due to failed financing of Northern Pacific Railroad bonds, triggering a panic; during the panic of 1893, a large number of railroad companies, including Union Pacific, Northern Pacific, and Reading Railroad, went into bankruptcy. However, the tracks remained, becoming the backbone of American industrialization. During the fiber optic boom, WorldCom and Global Crossing went bankrupt in succession, and less than 5% of the laid fiber was lit at one point; this excess capacity later contributed to the success of Google and YouTube. AI infrastructure is currently in the latter stage of its expansion: financing sources are shifting from free cash flow to debt and off-balance-sheet structures, which is a common signal seen before the peak of every historical infrastructure boom.
At its core, this is about capital concentration. Four scarce resources—electricity, skilled labor, land, and credit—are being concentrated by one industry, with three layers of mechanisms. First, there are strong economies of scale in the competition for large models, where computational power serves as a moat, and leading firms must continue to invest heavily. Second, the cash flow and credit ratings of the five giants allow them to borrow at low costs, creating a capital siphon that other industries cannot compete with. Third, the rise in AI stock prices creates a wealth effect, which supports consumption and valuations, and lower valuations reduce financing costs, forming a self-reinforcing cycle. The costs are borne by the manufacturing sector that is being crowded out, the housing market, and households facing higher interest rates, putting the Federal Reserve in a dilemma: raising interest rates to curb investment inflation may burst the entire cycle.
ABAB News · Cognitive Laws
The foundation of prosperity is debt, and the starting point of collapse is also debt.
Every infrastructure frenzy ultimately pays for excess capacity.
Electricity, workers, and land are the true currencies of the computational era.
Two points to note:
Chart Numbers: In the chart, the proportions and names of canals, railroads, electrification, highways, and telecommunications do not match, so the newsflash only used AI's 3.6%, and historical comparisons only noted that "the railroad boom was the last time a higher proportion was recorded."
Background in AI Interpretation: The Star Gate, Oracle and OpenAI's contract, Meta and Blue Owl's financing, Nvidia's investment in OpenAI, and the history of the railroad and fiber optic booms were all written based on publicly available English information I have, and this round was not verified line by line online.