The pitch sounds perfectly reasonable until you do the math. A student at Stanford - or really anyone with a decent slide deck and a functioning internet connection - can walk into a room today and raise $10 to $20 million for an AI idea. No revenue. No product. Just the idea. OpenAI, the most visible name in the space, recently completed one of the largest fundraises in corporate history, pulling in capital to build data centers while running revenues of around $13 billion against spending commitments that dwarf it. According to Fortune, OpenAI's CEO has projected spending of over $1 trillion across the next several years, creating a financing gap of roughly $1.2 trillion against current cash reserves.
In a conversation published today on The Diary of a CEO, NYU Stern professor Scott Galloway laid out the argument plainly: America has placed a single, enormous bet on artificial intelligence - and the historical pattern for bets like this one is a crash.
The pattern nobody wants to look at
Every major infrastructure boom in US history has followed roughly the same arc. A transformative technology arrives. Capital floods in. Spending climbs beyond what the underlying economics can support. Then something gives.
Galloway's rule of thumb - drawn from economic history rather than speculation - is that when infrastructure spending on a new technology exceeds 2 to 3 percent of GDP (the total value of everything a country produces in a year), a correction almost always follows. It happened with the railroads in the 1880s, with electrification in the 1920s, with the internet in 2000, and with the global telecom buildout at the turn of the millennium. In each case, the technology itself survived and eventually transformed the economy. The investors who financed the initial surge mostly did not.
The numbers now are eye-watering. According to Fortune, AI capital expenditure - spending on building the physical infrastructure that runs AI systems, mainly data centers and the chips inside them - has in 2025 contributed more to US GDP growth than all consumer spending combined, which is the first time that has ever occurred. The five major cloud companies - Google, Amazon, Meta, Microsoft, and Oracle - are collectively expected to invest around $700 billion this year to support AI-related buildout. That is an 80 percent increase from one year ago. A Galloway analysis cited in his own newsletter found that AI investments account for approximately 92 percent of US GDP growth this year. As Harvard economist Jason Furman noted in the same analysis: without the AI investment, growth would be flat.
That concentration is the problem. Not the technology - the concentration.
What is actually happening
The argument Galloway makes is not that AI is a fraud. He says clearly it is a genuine breakthrough. The argument is that genuine breakthroughs and lucrative investments are not the same thing, and American markets are currently pricing them as if they were identical.
History keeps producing examples of the gap between the two. Jet transportation is one of the most consequential technologies in human history - it has connected people, opened economies, and transformed how the world moves. Add up all the airline profits, subtract all the airline bankruptcies, include all the government subsidies paid to plane manufacturers, and the entire industry across its entire history is roughly at break even. It has never, collectively, made money. Personal computers reshaped civilization. The companies that built them - Gateway, Compaq, dozens of others - are mostly gone or hollowed out. The company that survived and captured enormous value was Apple, but Galloway's point is that Apple made its money from the iPhone, not from PCs.
According to Fortune, AI has been responsible for 80 percent of stock market returns since ChatGPT launched in late 2022. Roughly 40 percent of the S&P 500 - the index that tracks the 500 largest companies in the United States - is now directly or tangentially tied to the AI bet. If that spending slows, or if the market revises its expectations for AI returns downward, the knock-on effects reach well beyond the technology sector.
The additional structural risk, which Galloway calls the most underappreciated, is depreciation. According to reporting from Calcalist, the four largest AI infrastructure investors - Microsoft, Amazon, Google, and Meta - recorded combined depreciation expenses of $41.6 billion in a single quarter, a 36.9 percent increase year-on-year. Data centers built on cutting-edge chips age quickly. Analysts project that annual depreciation costs for these four companies could exceed $430 billion within five years - higher than their combined revenues of $372 billion in 2025. The infrastructure being built today will begin weighing on profits before its value is proven.
The money trail
The financial logic running underneath the AI boom has a specific shape. Companies raise money at valuations justified by the expectation that AI will generate massive returns. That money funds data centers, chips, and talent. The data centers consume electricity and depreciate. The returns, so far, have been limited. According to Goldman Sachs analysis cited by Fortune, $700 billion in AI investment in 2025 contributed essentially zero to US GDP growth in productivity terms. A Goldman study found no meaningful relationship between AI and economy-wide productivity, though isolated gains of around 30 percent appeared in specific tasks where companies were actively measuring.
The companies that have found a way to make money from AI so far are largely doing it through subscription price increases - raising the monthly cost of software and blaming it on AI feature upgrades. Microsoft raised its M365 personal subscription price by 30 percent in January 2025, the first such increase in its history. Adobe raised Creative Cloud Photography prices by 50 percent. Intuit hiked QuickBooks by 45 percent in August 2025. The AI capability is real; the value to the customer is debatable.
The more alarming scenario in the Galloway thesis involves China. The basic idea draws on a historical tactic called dumping - pricing a product below its true cost in order to undercut competitors and capture market share, then raising prices once competition is destroyed. China attempted this with steel in the 1980s and 1990s. Amazon and Netflix did a version of it in their own industries - selling goods and subscriptions below cost until they had locked in enough customers that raising prices became straightforward.
Galloway's argument is that China may be doing the same with AI. According to DeepSeek's own technical documentation, cited in multiple analyses, the company trained its R1 model for approximately $6 million - against the hundreds of millions typically spent on comparable US models. R1's capabilities, when released in January 2025, were broadly comparable to leading American models and triggered one of the largest single-day losses in Nvidia's stock market history. The V4 model, released last month, runs entirely on domestic Chinese chips.
Chinese open-weight models now account for 30 percent of all AI downloads globally, ahead of the United States at 15.7 percent. An open-weight model is one where the underlying code is made public and anyone can download, modify, and run it for free. In a world where capable AI is available without a subscription, the multi-million dollar site licenses that Anthropic and OpenAI sell to large corporations become very difficult to justify.
If a meaningful number of large enterprises cancel those contracts - not out of ideology but out of basic cost management - the revenue projections underpinning current AI valuations collapse. And because 40 percent of the S&P 500 is tied to those valuations, the market falls with them.
What people are doing about it
Institutional investors are already beginning to move. Galloway has disclosed publicly that he is diversifying his own portfolio out of US markets into Latin America and European equities, limiting exposure to any single asset to no more than 3 percent of net worth. The logic is blunt: diversification is the only protection against a risk nobody can precisely time.
Younger workers in companies dependent on AI spending are facing a different kind of pressure. According to a Gallup poll from April 2026 cited by Fortune, Gen Z workers' excitement about AI dropped 14 percentage points in a single year to just 22 percent, while anger rose 9 points to 31 percent. Forty-four percent of Gen Z workers admit to actively sabotaging their company's AI rollout by entering proprietary data into public tools or simply refusing to use the technology.
At the policy level, the White House issued a memo in late April accusing unnamed foreign entities - primarily Chinese - of conducting industrial-scale campaigns to replicate US AI models through a process called distillation, where a smaller, cheaper model is trained to mimic the outputs of a more expensive one. According to CNN, DeepSeek's new V4 model, which uses domestic Huawei chips, was the subject of renewed security scrutiny as it launched - yet it continues to spread, because it is free and it works.
Some governments have moved to ban Chinese AI models from public sector systems. Multiple US states, Australia, Taiwan, South Korea, Denmark, and Italy all introduced restrictions on DeepSeek's R1 after its January 2025 release, citing data privacy and national security concerns. Private sector adoption, however, has continued regardless.
The bottom line
The AI boom is not a hoax. The technology works, and it is changing things. But the price being paid for that technology - in stock valuations, in capital expenditure, in the share of US GDP it now represents - is built on the assumption that a small number of American companies will capture permanent, outsized profits from it. History does not support that assumption. Every other infrastructure revolution, from railroads to the internet, eventually became a commodity that everyone used and nobody monopolized. Meanwhile, cheap Chinese alternatives are already eroding the pricing power of US AI companies. When 40 percent of the stock market is riding a single story, the question is not whether the story will be tested. It is when.
Timeline
- 1880s - US railroad boom exceeds 2 to 3 percent of GDP; market correction follows
- 1920s - Electrification investment surge; correction follows the same pattern
- 1999 to 2001 - Amazon stock falls approximately 97 percent during the dot-com crash; the internet survives, most investors do not
- 2000 to 2001 - Global telecom buildout collapses; Global Crossing and others go bankrupt
- 2017 - China launches its New Generation Artificial Intelligence Development Plan, treating AI as an instrument of national power
- November 2022 - OpenAI launches ChatGPT; AI-related stocks begin a run that accounts for 80 percent of S&P 500 returns over the following two years
- January 2025 - DeepSeek releases R1, claiming training costs of approximately $6 million; Nvidia suffers its largest single-day stock market loss in history
- February 2025 - Multiple governments including several US states, Australia, and several European countries ban DeepSeek from public systems on security grounds
- January 2025 - Microsoft raises the price of its M365 personal subscription by 30 percent, its first ever such increase; Adobe raises Creative Cloud Photography by 50 percent
- August 2025 - Fortune reports AI capital expenditure has surpassed consumer spending as a driver of US GDP growth for the first time in history
- August 2025 - Intuit raises QuickBooks pricing by 45 percent
- October 2025 - Galloway publishes analysis showing AI investments now account for approximately 92 percent of US GDP growth; Harvard economist Jason Furman notes growth would be flat without them
- November 2025 - Fortune reports Galloway's warning that an OpenAI collapse would leave investors "nowhere to hide"; OpenAI's financing gap disclosed at approximately $1.2 trillion
- April 2026 - DeepSeek launches V4, its first model running entirely on domestic Huawei chips, matching near-frontier US performance
- April 2026 - White House issues memo accusing Chinese entities of "industrial-scale" distillation campaigns against US AI models
- May 2026 - Goldman Sachs analysis finds $700 billion in 2025 AI investment contributed essentially zero to US productivity growth; combined depreciation for the four largest AI infrastructure companies projected to exceed their combined revenues within five years
- May 6, 2026 - Scott Galloway outlines the full crash thesis on The Diary of a CEO, arguing the US market faces a 40 to 50 percent correction risk within 24 months
Summary
Who: NYU Stern professor Scott Galloway, speaking on The Diary of a CEO, with additional context from Goldman Sachs, Fortune, DeepSeek, and Galloway's own Prof G newsletter.
What: A detailed argument that the current AI investment boom follows the same historical pattern as the railroad, electrification, internet, and telecom bubbles - all of which exceeded 2 to 3 percent of GDP before crashing. With 40 percent of the S&P 500 tied to AI bets and AI capex now the single largest driver of US GDP growth, a slowdown or a loss of confidence could trigger a market correction of 40 to 50 percent. The additional mechanism is Chinese AI dumping - the deliberate release of cheap, capable open-weight models that undercut the pricing power of US AI companies.
When: The warning is current, published May 6, 2026, though the underlying data and historical pattern span from the 1880s railroad boom to the DeepSeek V4 release last month.
Where: The United States economy, with secondary implications for global markets. The AI infrastructure investment is concentrated in the US; the competitive pressure from Chinese open-weight models is global.
Why: Because the valuations of US AI companies are built on the assumption that a small number of firms will capture permanent, disproportionate returns from the technology - an assumption that every prior infrastructure revolution has eventually destroyed.