Not many want to remember the economic disaster of 1929, but we learned something extremely crucial from it: Speculation isn’t the enemy. Hiding it is.
After the crash, the US didn’t ban risk-taking. Instead, we made it more transparent with the formation of the SEC, the FDIC, margin rules and disclosure requirements. Andrew Ross Sorkin spent a decade studying what happened and his takeaway about these regulations was simple: “They’ve mostly worked. The American Century followed.”
Those guardrails gave us FedEx, Amazon, Tesla, SpaceX and Moderna. These companies bet big in public markets where quarterly SEC filings showed everyone the real numbers, revenues, margins, customer counts and research and development failures. So, in the end, investors could decide for themselves whether the risk matched the reward. AI companies have made a different choice: They’ve decided to take private money with private accountability while demanding public market valuations.
What Will Happen When the AI Bubble Pops?
The AI bubble is destined to pop because private funding structures hide colossal failure rates — 95 percent of pilots crash and 40 percent of projects are canceled — while demanding public market valuations. This risk is concentrated in the Magnificent Seven companies, which control 37 percent of the S&P 500, making the bubble 17 times larger than the dot-com crisis. When they correct, it will trigger a simultaneous market collapse, legal liability and a widespread trust crisis.
Infrastructure Investment Exceeds Adoption Rates
When OpenAI raised $6.6 billion from sovereign wealth funds, it bypassed the entire public disclosure system. No SEC registration statement details the use of proceeds. No quarterly earnings calls explain burn rate or customer acquisition costs. No mandatory 10-Q filings showing which enterprise pilots converted to revenue and which failed.
So, when AI projects fail, no one is explaining why. Instead, we see the aftermath: 130,981 jobs that vanished by July 2025, or the news that more than 700 organizations were breached through Drift AI. We see trust in AI hemorrhaging, with 34 percent of respondents “concerned” about AI versus just 16 percent who say they’re “excited” globally.
Companies will likely spend an estimated $1.4 trillion on AI data centers by 2027, but actual production deployment at scale is only five percent. That gap exists because private money hides failure rates that public markets would not accept.
If these numbers showed up in quarterly SEC filings, the market would demand answers. Gartner reports that 40 percent of agentic AI projects get canceled before they even launch. MIT found that 95 percent of pilots crash and burn. McKinsey reports that only 10 percent of companies have successfully implemented generative AI at scale for any use case. Despite these dismal numbers, 55 percent of companies fired workers to make room for AI, then admitted they made a mistake.
Why the AI Bubble Is Worse Than the Dot-Com and Subprime Crises Combined
The MacroStrategy Partnership recently published research identifying that the AI bubble is 17 times larger than the dot-com bubble and four times bigger than the 2008 subprime meltdown. The 2008 subprime crisis required $700 billion in taxpayer bailouts. The dot-com crash took 15 years for the NASDAQ to recover.
This time, the bubble is concentrated in seven companies known as the Magnificent Seven (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla). Together, these seven companies control 37 percent of the S&P 500. That means, if they crash, every index fund and retirement account goes down with them.
Large language models (LLMs) have slammed into scaling limits. For example, ChatGPT-3, with 175 billion parameters, had training costs estimated between $4.6 million and $12 million, depending on the hardware and optimization techniques used. The GPT-4 training cost OpenAI more than $100 million and took 100 days. And then GPT-5 got delayed, and when it finally launched, users couldn’t tell much difference. The individual six-month training runs can cost over $500 million in computing costs alone, with total training costs estimated between $1.25 billion and $2.5 billion. This means that these companies will need to spend trillions to get anywhere near the computing power they’re looking for, and for negligible advancement.
Microsoft’s GitHub Copilot, OpenAI’s ChatGPT Plus and similar products currently lose money on their heaviest users because the compute costs of serving complex queries exceed the fees customers pay in monthly subscriptions. When your most engaged customers, the ones proving the technology works at scale, are unprofitable, your company faces a huge dilemma. You can’t charge more without destroying adoption, but you can’t keep current pricing without bleeding cash on every power user.
This is what the end of a bubble looks like: Costs increasing 250 times while results plateau, but spending continues anyway because stopping would mean admitting the entire premise was flawed. This all happens because these companies are not being held accountable by external, public powers.
The Magnificent 7 Amplify Harm at Computational Speed
Economic Harm
The Magnificent Seven will collectively spend more than $370 billion on AI infrastructure alone this year, up 58 percent from $228 billion in 2024. The pattern of repeated increases exposes a competitive arms race where each upward revision by one company forces the others to match or exceed. Falling behind by even one quarter means surrendering billions in market positioning. The escalation reflects competitive pressure rather than planned investment based on ROI projections.
Seven companies controlling more than one-third of the entire S&P 500 violates every diversification principle we learned from 1929. When the Four Horsemen (Cisco, EMC, Oracle and Sun Microsystems) dominated during the dot-com boom, they didn’t come close to this level of market control.
The Magnificent Seven achieved a 697.6 percent combined return from 2015 through 2024 while the broader S&P 500 gained 178.3 percent, outperforming eight out of nine years. The only year the broader market did better was in 2022, when everything crashed and the Magnificent Seven lost 41.3 percent while the S&P 500 lost 20.4 percent.
When these seven stocks fall, they fall hard, and they drag everything down with them. And this dynamic is raising serious questions about whether this massive spending will generate returns that justify current valuations.
Because the economic bubble is so astounding due to this unprecedented market concentration, when the Magnificent Seven correct from these valuation levels, we will get all the following negative outcomes simultaneously: a market collapse, legal liability for the crash and a trust crisis in the AI systems that now touch every sector of the economy.
Human and Social Harm
The Magnificent Seven aren’t just building a financial bubble; they’re also encoding systemic bias into systems deployed at cloud scale, then hiding the damage behind private capital structures that require zero public disclosure. The evidence is alarming.
The University of Washington found AI models preferred white names in 85 percent of tests, while Denver Public Schools’ AI consistently shows white men as doctors and Black men as janitors. Likewise, Delta Air Lines uses AI pricing that delivers the best deals to wealthy customers and the worst deals to poor customers. AI pricing systems increasingly use personal data to determine the maximum price each customer will pay based on their browsing behavior, purchase history and demographic information.
When bias exists in one company, the market can route around it. Customers can switch, their competitors gain share and the problematic company loses value. But with such a small concentration of companies providing the foundational AI infrastructure that thousands of other companies build on top of, bias doesn’t stay contained. It propagates.
I’ve audited enough systems to know 85 percent of bias can be eliminated with proper standards and real commitment. The Magnificent Seven are choosing the opposite path. They’re making a calculated, operational choice to ship fast and settle discrimination lawsuits later rather than invest in prevention that would delay market capture.
The Dot-Com Comparison Should Worry Us
The Four Horsemen were seen as essential infrastructure providers during the dot-com era. When the bubble burst, NASDAQ dropped 78 percent from its March 2000 peak to October 2002. Cisco’s stock fell by more than 80 percent and took 24 years to recover its 2000 peak price. EMC and Sun were eventually acquired after years of decline.
This time, the stakes are much higher. The Four Horsemen never exceeded 20 percent of the market, but the Magnificent Seven control nearly twice that. A 40 percent correction in these seven stocks, matching their 2022 decline, would automatically trigger a 15 percent S&P 500 drop before broader selling starts, impacting every index fund, retirement account and pension.
The crash will hit at the same moment regulatory intervention starts to accelerate. European regulators are already implementing mandatory bias audits under the EU AI Act. Class action lawsuits are being prepared alleging systematic harm from biased hiring algorithms with potentially millions of affected job applicants. The Department of Justice and the Equal Employment Opportunity Commission have issued guidance on how anti-discrimination laws apply to algorithmic hiring tools.
When stock crashes from bubble valuations, that’s when billion-dollar settlements hit, regulators force system rebuilds and operational restrictions eliminate the competitive advantages that justified those valuations.
The Choice Ahead
Post-1929 reforms worked because transparency let markets price risk accurately. Companies disclosed problems before they became disasters. That framework turned speculation into stable economic activism that empowered nation-building.
The AI boom has led companies to choose the opposite direction, preferring opacity to transparency. The bubble will pop. The math guarantees it. But two paths lie ahead.
Either we adopt the framework that built the American Century and demand transparency now to stop speculation in its tracks and replace it with accountability. We can demand insight into the biases AI may exacerbate and disclosure of product failures.
Or we can do nothing, allowing history to repeat itself, and watch a crash erase millions of jobs, trillions in wealth and decades of trust.
We still have the opportunity to make the right choice.
