How the AI Boom Found Its Buyer: Your Retirement Account
SpaceX, OpenAI, and Anthropic are heading public near a combined $4 trillion, and index rules rewritten in spring 2026 mean passive 401(k) funds will buy them automatically. A general-reader explainer on the committees, the mechanics, the fiduciary fight, and the history behind it.
Safe AI AcademyJune 7, 202641 min read11 views
On June 12, 2026, a rocket company that lost money last year is scheduled to become one of the most valuable businesses on the American stock market. SpaceX will list on the Nasdaq under the ticker SPCX at a price that values it at roughly $1.75 trillion, which would make it larger than every publicly traded American defense contractor combined and the biggest stock-market debut in history, surpassing Saudi Aramco's 2019 record.
Consider a schoolteacher who has a retirement plan at work and has never bought an individual stock in her life. Her contributions go, by default, into a fund that simply mirrors the market. She did not choose SpaceX. She may have no opinion about Elon Musk or Mars. Within a few weeks of the listing, her retirement account will own a slice of SpaceX anyway, because the rules that govern what her fund must buy were rewritten this spring, weeks before the largest run of initial public offerings the market has ever seen. This is the story of how that happened: the plumbing that moves the money, the small private committees that pulled the lever, the accounting that makes the boom look more profitable than it is, and the long history of what tends to happen next.
How a retirement account ends up buying a company it never chose
Start with the machine that moves the money. Most retirement savings in the United States do not sit with a stock-picker who studies companies. They sit in index funds.
Index fund. A fund that does not try to beat the market. It just copies a list, called an index, and holds every company on that list in proportion to its size. When a company joins the index, every fund tracking that index has to go buy it, automatically, regardless of price. The fund manager gets no vote and exercises no judgment. That is the entire point: low cost, no decisions.
The reason this reaches a schoolteacher who never picked a stock is that the United States quietly made index funds the default home for retirement money. The let employers automatically enroll workers and steer their contributions into a "default" investment, and the option that became standard is the target-date fund, a single fund built mostly from index funds that adjusts as you age. By the end of 2024, used automatic enrollment, up from 10% in 2006, and about 60% of participants held a single target-date or balanced fund. The pool this feeds is enormous. Total US retirement assets reached at the end of 2025, including about $10.1 trillion in 401(k) plans. Most of that money buys whatever the index says to buy, on a schedule, without anyone deciding.
Stay Updated
Get notified when we publish new articles and course announcements.
The most important list for technology companies is the Nasdaq-100, the 100 largest non-financial companies on the Nasdaq exchange, with more than $600 billion in investment products tracking it. For decades there were guardrails on who could get in. One of them was a rule about float.
Float. The share of a company's stock that is actually available for the public to trade. If a company has 100 shares but insiders hold 95 and only 5 trade freely, its float is 5%. A small float means a thin, easily moved market: a lot of forced buying chasing very few available shares.
Until this spring, a company needed at least a 10% public float to qualify for the Nasdaq-100, and it had to wait, sometimes up to a year, for the index's scheduled review. On May 1, 2026, Nasdaq put a new methodology into effect. The changes, laid out in the exchange's own methodology update and summarized by the law firm Ashurst, did 3 things:
A "fast entry" rule cut the waiting period from as long as a year to 15 trading days.
The 10% minimum-float requirement was removed.
A weighting mechanism was added: a company with a float below 20% has its index weight calculated as if its float were 3x larger.
That last piece matters more than it sounds. SpaceX plans to list with a float of around 4% to 5%. Under the old rule it would not have qualified at all. Under the new one, a 5% float is treated as 15%, so index funds are required to buy roughly 3x as much of the company as its freely traded shares would normally justify. FTSE Russell, which runs another widely tracked family of indexes, shortened its own inclusion window to 5 trading days.
The result is a chain that runs from a private company's decision straight into a stranger's retirement account, with no human in the middle choosing to buy.
How a company with a 5% float reaches retirement money it never had to persuade.
How much money? The estimates are large and, by their nature, approximate, because they depend on how aggressively each fund tracks its index. The options-analytics firm SpotGamma puts the mechanical buying at $15 billion to $30 billion across the major funds in a compressed window, and well over $100 billion in the most aggressive scenarios. Goldman Sachs analysts estimated the Nasdaq-100 change alone could force up to $60 billion of buying. The deeper problem is the mismatch: that wall of required demand would be chasing a float of only about 5%, which SpotGamma estimates could amount to 15% to 30% of all tradable shares being bought in a few weeks. When forced buyers must purchase far more of a stock than is freely available, the price tends to rise simply because of the buying, not because of anything the company did.
Wall Street has a blunt term for what the early insiders get when that wave of new buyers arrives at a high price.
Exit liquidity. The buyers who show up at the moment early investors want to sell. Insiders bought cheap years ago; to cash out, they need a large pool of money willing to buy at today's much higher price. Whoever does that buying provides the "liquidity" that lets the insiders exit.
There is a real check worth noting. S&P Dow Jones Indices, which controls the S&P 500, declined to drop its requirement that a company show positive net income, so SpaceX and the others are not being fast-tracked into the most widely held index of all, at least not yet. That single decision keeps the largest pool of passive money on the sidelines for now, which is the clearest sign that none of this is automatic in the deeper sense. Each gate was opened, or held shut, by a choice. The next question is who makes those choices, and to whom they answer.
The committees that now move trillions
The rules that decide what a passive fund must buy are not written by Congress or the Securities and Exchange Commission. They are written by the companies that publish the indexes: Nasdaq, FTSE Russell, S&P Dow Jones Indices, and MSCI. These are private firms, and the assets that move on their decisions are staggering. By S&P's own count, roughly $13 trillion was benchmarked to its indexes at the end of 2023. When one of these firms adds or drops a company, trillions of dollars in funds follow, on autopilot.
Most people picture an index as a neutral, mechanical list. The reality is more discretionary than that. In a study aptly titled "Passive in Name Only," the legal scholar Adriana Robertson found that many index methodologies, including some of the most prominent, explicitly grant a committee of individuals discretion over what goes in, rely on non-public information, and change frequently. Index investing, she argues, is really a form of delegated management, with the index provider quietly making the calls. A separate line of research by the political economists Johannes Petry, Jan Fichtner, and Eelke Heemskerk goes further, describing index providers as "becoming gatekeepers that exert de facto regulatory power" over how capital is allocated across companies and even countries.
Index committee. A group of people at a private index company who decide, under rules the company itself writes and can change, which stocks belong in an index. Because trillions in funds are contractually required to track the index, the committee's decisions move markets even though its members are not public officials and answer to no electorate.
Just 4 steps move retirement money from a private committee's decision to a saver who never sees it happen.
In most of the financial system, decisions of this consequence carry some public oversight. Stock exchanges are regulated, and the SEC oversees them. Index providers occupy a gap. They have long relied on a "publisher's exclusion" in US law that treats them like financial publishers rather than investment advisers, which keeps them outside adviser regulation. In June 2022 the SEC issued a request for comment asking whether index providers are in fact "making active decisions" that drive what funds buy, and whether that exclusion still fits. No rule followed. Europe took the harder line years ago: the EU Benchmarks Regulation, built on the voluntary IOSCO benchmark principles, requires benchmark administrators to register and submit to oversight, while the United States has no equivalent regime.
This is not a hypothetical concern, because index committees have made consequential calls before. When MSCI decided in 2017 to add mainland Chinese stocks to its emerging-markets index, it routed an estimated $22 billion of passive money into Chinese equities through the first step alone, a capital-allocation decision with clear geopolitical weight, made by a private firm. The spring 2026 rule changes drew unusually sharp reactions from inside finance journalism. As reported by InvestmentNews, The Wall Street Journal's Jason Zweig called the Nasdaq change "arbitrary, unfair and potentially risky," and the Financial Times' Robin Wigglesworth called the wave of fast-entry rules "the biggest bagholder exercise of all time." The pattern the evidence describes is that the gatekeeping function over trillions in retirement money now sits with private committees whose rules can change in the months before a large listing, with no public body required to weigh in. As long as that gap persists, the most important decision about what a default retirement fund buys will keep being made in a venue the saver cannot see and the regulator does not reach, and the pressure to close it will rise each time a committee's timing lines up this neatly with a seller's.
What that money is being asked to buy
The dream attached to SpaceX is rockets and Mars. The business underneath the dream is 3 companies stapled together, and only 2 of them make money:
Starlink: the satellite-internet service, and the genuine success, with roughly 10 million subscribers, about $11.4 billion in revenue, and profit margins above 60%.
xAI: Musk's artificial-intelligence venture, folded in, which burns more than a billion dollars a month.
Put together, SpaceX posted a net loss of about $5 billion in 2025 and carries an accumulated deficit above $40 billion. At $135 a share, the company is priced at roughly 94x its revenue, a multiple with no precedent among the world's most valuable firms. The research firm Morningstar puts SpaceX's fair value at less than half the IPO target.
A profitable satellite business and a profitable launch business, bundled with a cash-burning AI lab, priced as the second-largest company in America.
SpaceX is only first in line. Behind it, both OpenAI and Anthropic are reported to be preparing public listings of their own. Stack the 3 together and the market is being asked to absorb something close to $4 trillion in newly issued stock within months, 3 companies that would leap near the top of the entire American market on their first days of trading. The takeaway is not that any one of these businesses is worthless; Starlink plainly is not. It is that the price being set is a bet on a future that has to arrive in full, and the buyers being lined up to fund that bet are not the ones placing it. If the listings clear at these numbers, the more telling test comes after, when the same valuations have to be defended quarter by quarter against actual revenue rather than private-round optimism.
Why the timing matters: the money that moves in a circle
To understand the urgency behind all of this, look at how the AI boom's headline profits are actually produced, because a meaningful part of them never leaves the family.
The largest technology companies, Microsoft, Amazon, Alphabet, Meta, and Oracle, are investing enormous sums in AI startups. Those startups then spend much of that money renting computing power back from the same large companies. Bloomberg has mapped the closed loop of interlocking deals among Nvidia, OpenAI, Microsoft, Oracle, AMD, and others, including Nvidia's commitment to invest up to $100 billion in OpenAI, much of which OpenAI then spends on Nvidia chips. The investment goes out, the cloud revenue comes back, and the gain is booked as profit that helps justify the next round of spending.
Circular financing. When a supplier funds its own customers so they can buy the supplier's product. The money makes a loop: out as investment, back as revenue. Each pass can look like growth on both companies' statements, even though little new cash has entered the system from outside.
Investment out, cloud revenue back, the gain booked as profit, the profit used to justify the next round.
Does the loop create real value or mostly the appearance of it? A Financial Times analysis with the research firm Panmure Liberum tried to measure it. It compared how much the 5 hyperscalers plan to spend on AI through 2030 against the revenue those investments are expected to generate, then calculated the implied return under deliberately generous assumptions: no electricity, no salaries, no overhead, revenue against capital spending alone. Even on that charitable math, only 2 of the 5 cleared zero:
Amazon: about positive 7%
Microsoft: barely above zero
Alphabet: about negative 16%
Meta: about negative 29%
Oracle: about negative 36%
These are best-case figures.
The independent short-seller Jim Chanos, who built his reputation by exposing Enron's accounting, has focused on a quieter lever: depreciation. Companies do not expense a new server all at once. They spread its cost over its expected "useful life," and they get to estimate that life. Chanos argues hyperscalers depreciate graphics chips over roughly 6 years when their real economic life, in a market where each generation is far faster than the last, may be 3 to 4. He points to rental prices for Nvidia's previous-generation Hopper chips falling about 28% year over year as evidence the hardware loses value quickly. The longer the assumed life, the smaller each year's depreciation charge, and the larger the reported profit.
Useful life. The number of years a company assumes a piece of equipment will earn its keep, used to spread the cost across time on the books. A longer assumed life means a smaller yearly expense and a bigger reported profit, even though the cash was already spent. The estimate is the company's to make.
What makes this more than theory is that the companies have been changing those estimates, and not in the same direction. The contrast is stark enough to draw.
Same hardware, 2 companies, opposite accounting choices in the same year.
In 2025 Meta extended the assumed life of certain servers and networking gear to about 5.5 years, a change it expected to reduce depreciation and lift profit by roughly $2.9 billion. The same year, Amazon went the other way, shortening the assumed life of some equipment to 5 years and explicitly citing the fast pace of AI development, a change that cut its operating income by about $0.7 billion. These are legitimate, disclosed accounting estimates, not fraud, and reasonable people can defend either choice. But when 2 of the most sophisticated companies in the world look at the same hardware and reach opposite conclusions about how long it will last, it underlines how much of the AI boom's reported profit rests on a judgment call. Chanos's broader warning is that this inflates earnings across the market, because revenue is booked now while much of the cost is spread into the future.
This is what makes the IPO timing legible. As long as the AI companies stay private, their valuations are whatever a handful of investors agreed to in the last funding round, and the paper gains supporting everyone's earnings stay unchallenged. The moment they go public, the open market sets the price. The pattern the evidence describes is an industry whose reported profitability depends, to an uncomfortable degree, on circular revenue and on estimates the companies themselves control. If those foundations are sound, public listings will confirm them in the open; if they are not, the same listings are the most efficient way to transfer the risk before the question is forced, which is why the next 2 earnings seasons, when public prices finally meet private assumptions, are the venue where this resolves.
The case that the spending is justified
A fair account has to take the other side seriously, because the bull case is not flimsy. The strongest version of it rests on contracted demand. When a company signs a multiyear cloud deal, the future revenue shows up on the seller's books as a backlog.
Remaining performance obligations. The dollar value of contracts a company has signed but not yet delivered, in other words, revenue it is legally owed in future years. Unlike a forecast, a backlog reflects deals already struck with named, paying customers.
Those backlogs are now immense. Microsoft reported about $625 billion in remaining performance obligations, roughly double a year earlier; across Microsoft, Amazon, Google, and Oracle the combined figure has been estimated at more than $1.6 trillion. Bulls argue this is real evidence the demand is not imaginary. Adoption supports them: more than 90% of businesses report using AI in at least one function, and about two-thirds report measurable efficiency gains. Goldman Sachs, which expects AI companies to invest more than $500 billion in 2026, notes that analysts have underestimated this demand 2 years running. The investor Dan Ives of Wedbush calls 2026 "the year of AI monetization," the third inning of a long game.
There is also a balance-sheet argument that this is not the late-1990s telecom bust in disguise. That buildout was financed largely with junk debt by speculative startups. Today's lead spenders are profitable incumbents, and Morgan Stanley has characterized their balance sheets as strong, with low leverage and cash flows that can support the spending. A backlog is harder to fake than a forecast, and a profitable company can absorb a bad bet that a leveraged startup cannot.
There are 2 caveats that keep the bull case honest:
The funding model is shifting toward debt: Morgan Stanley and others estimate the sector may need roughly $1.5 trillion in new borrowing over 3 years as free cash flow tightens, which erodes the "funded from strength" claim over time.
The comparison to past infrastructure cuts against the bulls in one specific way. Railroad track and buried fiber had useful lives measured in decades, so the assets kept earning long after the bust. The chips at the heart of the AI buildout have useful lives measured in a handful of years, which means the "the infrastructure survives even if investors don't" reassurance applies far more weakly to a warehouse of 2-year-old GPUs than it did to a rail line.
The evidence here genuinely cuts both ways: the demand is real and the incumbents are strong, but the margin for error narrows the more the spending leans on debt and on hardware that depreciates fast. Which way it resolves will show up first in whether backlog converts into free cash flow rather than into more borrowing, over the next several quarters.
Who is holding it when the music slows
There is one more reason the buyer pool matters, and it is about how concentrated the market has already become. AI-related companies now make up roughly 45% of the S&P 500's total value, an all-time high for any single theme, up from about 25% when ChatGPT launched. Nvidia alone is the largest company in the index, weighing more than entire sectors like energy or utilities. For the schoolteacher in the index fund, this means her "diversified" retirement account is, to an unusual degree, a single concentrated bet on artificial intelligence, and it is about to add more names to that bet. The funds doing the buying are dominated by 3 firms: BlackRock, Vanguard, and State Street together held an average stake of about 20% of S&P 500 companies as of 2017, a share that has grown since, which concentrates the passive bid in remarkably few hands.
The concentration matters most for people near the end of their careers, because of a risk that has a name.
Sequence-of-returns risk. The danger that a market drop early in retirement does lasting damage. A retiree who must sell shares during a downturn to cover living costs locks in losses and has fewer assets left to recover when the market rebounds. The same average return, in a worse order, can mean running out of money years sooner.
A 30-year-old can wait out a drawdown in an AI-heavy index fund. A 64-year-old who is about to start drawing on the same fund cannot, which is precisely the population that automatic enrollment has swept into these funds by default.
The exposure is being deliberately widened on a second front. In August 2025, Executive Order 14330 directed federal agencies to clear the path for private-market assets, including private equity and private company stakes, to enter ordinary 401(k) plans, and gave the Department of Labor 180 days to revisit its guidance. The department issued a proposed rule on March 30, 2026, creating a legal "safe harbor" meant to protect employers from lawsuits when they add such assets, a response to the more than 500 retirement-plan suits filed since 2016. BlackRock's chief executive Larry Fink, whose firm manages around $14 trillion, has argued for routing retirement and pension savings toward private markets and tokenized assets, and BlackRock has set a target of $400 billion in private-markets fundraising by 2030.
The reason watchdogs object is that private assets carry exactly the features a default retirement fund is supposed to avoid. They charge far higher fees, they lock money up for years, and their values are marked infrequently and subjectively, which can create what SEC Commissioner Mark Uyeda, who supports wider access, nonetheless called "the illusion of stability." The advocacy group Better Markets argues the push "benefits Wall Street rather than Main Street," noting that private equity returned about 5.8% annually from 2022 through September 2025 against the S&P 500's 11.6%, and the CFA Institute's research concluded plainly that retail investors should "stay away." The illiquidity is not hypothetical: when investors rushed to pull money from Blackstone's retail real-estate fund in late 2022, the fund capped withdrawals and honored only a fraction of requests for more than a year.
Every gate loosened in the year before the listings it now admits.
The pattern across these moves is consistent: the same months that opened the index door to low-float mega-listings also opened the retirement door to illiquid private assets, both justified as expanding access, both shifting risk toward the saver who chose neither. Considered together, the concentration data, the fiduciary changes, and the documented behavior of illiquid funds under stress suggest that the next stage of this story will be decided less by the IPOs themselves than by whether regulators and plan sponsors treat default retirement money as a pool to be tapped or a trust to be protected, a choice the public-comment period on the Labor Department's rule will begin to settle.
What history says happens next
This kind of episode has happened before, and the record is clear enough to be useful, because the lesson is not that the technology fails. The economic historian Carlota Perez described a recurring shape to technological revolutions: an installation phase in which financial capital races ahead of what the technology can yet deliver, a crash, and only then a long deployment phase in which the technology's real benefits spread. The crash is not the end of the technology. It is the end of the first set of investors.
Two episodes make that pattern concrete:
British railway mania (1840s): the starkest example. According to the economic historian Andrew Odlyzko, British investors poured roughly £250 million into railways by 1850, close to half of the country's entire annual output, a sum he likens to several trillion dollars in a modern US economy. Parliament authorized hundreds of new railways; about a third were never built, the share index fell by more than half, and ordinary investors were ruined. The rail network, of course, survived and carried the Victorian economy for a century.
The 1990s fiber-optic build: Telecom companies spent more than $500 billion, much of it borrowed, on a buildout justified by a now-famous myth that internet traffic was doubling every 100 days, when it was really doubling about once a year. By the mid-2000s the great majority of the fiber sat unused, or "dark," WorldCom and Global Crossing collapsed in 2 of the largest bankruptcies of their era, and the investors were wiped out. A decade later that same fiber became the backbone of the modern internet, bought cheaply by those who came after.
The academic record on individual IPOs points the same way. The finance scholar Jay Ritter has documented for decades that newly public companies tend to underperform the broader market over the years following their debuts, and that the companies going public during periods of peak enthusiasm underperform the most, precisely because that is when sellers find the most eager buyers. This is not a fringe view in 2026. The Pulitzer Prize-winning financial historian Liaquat Ahamed has explicitly compared the AI buildout to the 1870s railroad boom and the fiber era, noting that the railroads survived their crash to drive American growth for half a century even as their early bondholders were ruined. The takeaway from 2 centuries of these episodes is consistent: the useful question is rarely whether a transformative technology is real, but who is positioned to absorb the distance between what it costs to build now and what it earns later. The current arrangement answers that question in an unusually specific way, by routing the cost toward the most passive, least consulted money in the system, which is why the coming listings are better understood as a test of who bears the risk than as a verdict on whether AI works.
The buyer's paycheck is the thing being automated
There is a quiet contradiction at the center of this arrangement, and it ties the financial story back to the rest of the AI economy. The passive money these listings count on exists only because working Americans have paychecks to invest every 2 weeks. That foundation is already thin. The personal saving rate fell to 2.6% in April 2026, less than half its long-run average and a level previously seen mainly in the runups to recessions. Corporate earnings are rising while worker incomes stay flat, and part of that gap is the same automation these companies are selling.
That is the loop that does not close. The AI buildout is being financed, increasingly, by the retirement savings of the workers whose paychecks the same technology is built to reduce. If the displacement arrives faster than the returns, the pool of fresh money flowing into index funds every 2 weeks gets shallower at exactly the moment the system needs it deepest.
None of this requires a crash to matter, and a crash is not the only possible ending. The listings may clear, sentiment may hold, and the returns may eventually justify the spending; the historical pattern rhymes rather than repeats. What has already changed is structural: the machinery of ordinary retirement saving has been quietly rewired to serve as the buyer of last resort for the largest private valuations ever assembled, and the people providing the money are, by design, the ones with no vote on the price. For the teacher whose index fund buys SpaceX on June 12, the practical task is to understand what she now owns and why. The larger question, for everyone, is what it means that the most consequential financial decision of the AI era is being made, one rule change and one default setting at a time, by people the saver will never meet. The 2 earnings seasons that follow these listings, when public prices finally meet private assumptions, are where the answer starts to come due.
Is it even worth buying these new stocks?
Here are the main factors to consider:
What you would actually be buying: SpaceX plans to list only about 4% to 5% of its shares, so the publicly traded slice is small and insiders hold the rest. Musk has agreed not to sell for 366 days, but other insiders can begin selling much sooner, after the first quarterly earnings report.
Price against fundamentals: At $135 a share, SpaceX is valued at roughly 94x its revenue and posted a net loss of about $5 billion in 2025, and Morningstar's independent estimate of fair value is less than half the listing price. OpenAI and Anthropic are not yet profitable either.
What is inside the box: Each company bundles very different parts. SpaceX pairs the profitable Starlink with the cash-burning xAI, and OpenAI and Anthropic depend heavily on circular cloud-and-chip commitments with the same large tech firms that invest in them. Reading the financials means separating the parts.
How solid the reported profits are: Sector earnings lean on accounting estimates the companies set themselves, such as how many years a chip is assumed to last; Meta extended that assumed life to lift profit while Amazon shortened it. A Financial Times analysis found the implied return on AI spending is negative for most large tech firms even under generous assumptions.
The effect of forced index buying: Index inclusion triggers mechanical buying that can lift a price near the listing regardless of the business, though research on the "disappearing index effect" finds the once-reliable inclusion pop has faded as markets learn to anticipate it. A price moved by forced flows is not the same as a price set by demand for the company.
What you may already hold: AI-related companies are already about 45% of the S&P 500, so an ordinary index fund is already a large AI position, and adding these listings raises that concentration rather than reducing it.
The base rate from history: Companies that go public at peak enthusiasm have tended to underperform afterward, and in past technology booms the technology often succeeded while the earliest investors did not. That is a tendency rather than a rule, and it sits against the genuine demand signals on the other side: a contracted backlog above $1.6 trillion and balance sheets strong enough to absorb a bad year.
Your own time horizon: Because a drop early in retirement does lasting damage (sequence-of-returns risk), the same holding can carry very different risk for a 30-year-old saver and a 64-year-old one. The horizon matters as much as the company.
None of these facts settles the question on its own. Together, they are the ground the decision actually stands on, and each one is easy to verify on your own before you buy.
Comments
0 commentsBe the first to leave a comment.
Leave a comment
Posted a comment before?