Every major technology cycle eventually forces the same question onto investors, regulators, and ordinary savers alike: is the money chasing this trend real, or is it belief that has been mistaken for value? Artificial intelligence, and specifically the two companies sitting at its centre, OpenAI and Anthropic, are now far enough into their growth story that the question has stopped being theoretical.

Valuations have compounded at rates with almost no precedent in corporate history, financed by a capital base so large and so tightly interwoven with the rest of the global financial system that Norwegian pension money, American mutual funds, and ordinary 401(k) holders are all exposed to the outcome, whether they’ve ever opened ChatGPT or not.

Understanding whether this is a bubble, and what kind, requires looking past the headlines and into three things: the valuation data itself, the mechanics of how that exposure spreads through global capital markets, and the two closest historical analogues available: the dot-com crash of 2000 and the housing-driven financial crisis of 2008.

The dot-com crash and the 2008 financial crisis get lumped together in casual conversation as “market crashes”, but they were mechanically very different events, and that difference matters enormously for anyone trying to model what an AI unwind might look like.

The dot-com bubble was, at its core, a story of pure equity speculation. Companies with little or no revenue were valued almost entirely on narrative, the promise that the internet would eventually justify any price paid for exposure to it. At the peak in 2000, only around 14% of dot-com companies were actually profitable, and the Nasdaq-100’s forward price-to-earnings ratio reached roughly 60 times earnings. When investor belief cracked, valuations collapsed quickly and dramatically. But because the risk was concentrated almost entirely in equity markets and venture capital, the damage, while severe for a generation of investors and startups, never metastasised into the banking system. The economy kept functioning.

2008 was a different animal entirely. It was a story of leveraged debt and hidden, mispriced risk. Subprime mortgages were extended to borrowers who often had little realistic ability to repay them, and those loans were then repackaged into complex securities that credit rating agencies stamped as safe, despite not fully understanding, or not wanting to understand, what those instruments actually contained. Because that debt had been woven directly through the banking system, and into the pension funds and insurance portfolios that ordinary people’s retirements depended on, the failure wasn’t contained to one overheated sector. It became systemic, global, and required government intervention to stop it from taking the entire financial system down with it.

The live debate among economists right now is which of these two patterns the AI investment boom more closely resembles, and a growing number of analysts argue it may actually contain elements of both simultaneously, layered on top of each other.

The raw numbers help explain why the comparison is being made at all. OpenAI’s valuation moved from roughly $29 billion in early 2023, to $157 billion by October 2024, to $300 billion in March 2025 following a $40 billion round led by SoftBank, to $500 billion by October 2025 in a secondary sale that involved no new capital at all, simply existing shares trading hands at a higher implied price, and finally to $852 billion by March 2026, in a round backed by SoftBank, Andreessen Horowitz, Amazon, Nvidia, and Microsoft. Revenue has genuinely grown alongside this, from around $1.6 billion in 2023 to an estimated $40 billion annualised run rate by 2026, but so have losses, with 2026 GAAP losses estimated near $62 billion, several times larger than the $14 billion non-GAAP figure more commonly cited in press coverage.

Anthropic’s trajectory is, if anything, steeper. The company moved from roughly $1 billion at its 2021 seed round, to $18.4 billion in 2023, to $61.5 billion in March 2025, to $183 billion by September 2025, to $380 billion in February 2026, a $30 billion round that was, at the time, the second-largest venture financing deal in history, and then to $965 billion just three months later in May 2026, reportedly on the back of revenue that grew rapidly to a $47 billion run rate over that same short window. Anthropic has raised roughly $132 billion in total capital since its founding in 2021.

Both companies have compounded their valuations by two to three orders of magnitude in under five years, a curve steeper than almost any private company has ever produced, dot-com-era darlings included. The central question isn’t whether the growth is real; it clearly is. It’s whether revenue, still trailing losses by tens of billions of dollars annually, can plausibly catch up fast enough to justify valuations approaching a trillion dollars before investor patience runs out.

What makes this different from a typical venture-capital story is how deeply that exposure has already spread into ordinary people’s savings. At least 66 mutual fund and interval fund vehicles report direct exposure to OpenAI shares, totalling roughly $3.79 billion as of early 2026, spanning household names like Capital Group, T. Rowe Price (through 19 separate fund vehicles), Fidelity (across 33 funds), BlackRock, and JPMorgan. Anthropic’s investor base includes T. Rowe Price, Goldman Sachs, Fidelity, and Alphabet, among 129 total investors. Crucially, both companies sit inside target-date retirement funds, the default investment option baked into most American employer-sponsored 401(k) plans, meaning millions of people hold indirect exposure to these valuations without ever having made an active decision to do so.

This pattern, private, high-flying valuations quietly filtering into diversified retirement vehicles held by people with no direct interest in the sector, is exactly how 2008’s mortgage risk ended up embedded inside pension funds that had never originated a single mortgage. It’s worth understanding not as a coincidence but as how modern capital markets structurally work: money doesn’t stay siloed in the place it was created.

The scale of that interconnection becomes vivid when looking at Norway’s sovereign wealth fund, the Government Pension Fund Global, the largest in the world at roughly $2 trillion, which owns approximately 1.5% of every publicly listed share on the planet. In 2025 and 2026, the fund divested from 11 of its 61 Israeli company holdings, and later exited five Israeli banks and Caterpillar, following its ethics council’s findings on those companies’ roles in the Gaza and West Bank conflict. The point isn’t the specifics of that particular divestment; it’s what the episode demonstrates about how thoroughly woven global capital already is. A single pool of Norwegian oil revenue simultaneously holds positions in Israeli banks, American industrial companies, and, by virtue of its globally indexed mandate, almost certainly a meaningful slice of the AI sector as well. A valuation shock in one concentrated sector doesn’t stay contained to the venture capitalists who funded it; it ripples outward through the same index funds, sovereign funds, and pension allocations that touch nearly every other sector and country at once.

Economists at Oliver Wyman, Goldman Sachs, and Fidelity, along with independent analysts, increasingly argue that “the AI bubble” is really two or three distinct risks stacked on top of each other, each resembling a different historical crisis.

The first is an equity bubble that closely resembles the dot-com pattern. The so-called Magnificent Seven tech companies have seen their combined valuation rise nearly eightfold since January 2020, and now represent roughly 35% of the entire S&P 500, matching the same degree of market concentration seen at the exact peak of the dot-com bubble in 2000. If AI monetisation disappoints investors, this layer behaves like a stock market correction: painful, but largely self-contained, the way the 2000–2002 downturn was.

The second is a debt bubble that more closely echoes 2008. Analysts estimate that more than $1.65 trillion in AI data-centre debt now sits off the balance sheets of major tech companies, financed through Special Purpose Vehicles that function structurally similar to the CDO instruments of the housing crisis, isolating individual loans so losses can, in theory, be absorbed piecemeal rather than threatening a parent company directly.

Roughly $500 billion in data-centre debt is already outstanding, with about $200 billion of it held by private credit funds, the same private-credit market that pension funds and insurers turned to for yield in the years after 2008. Some estimates suggest planned AI data-centre capacity could exceed real compute demand by as much as fifteen times.

A third and more novel risk is circular financing, an arrangement where companies like Nvidia, OpenAI, and Oracle invest in each other and lease compute capacity back and forth, inflating reported revenue in ways that don’t necessarily reflect genuine external customer demand. Oracle’s reported debt-to-equity ratio, near 500%, and its rising credit-default-swap spreads are being watched as an early stress signal in this loop.

Most serious analysts stop short of predicting a full 2008-style systemic collapse. AI’s underlying assets, data centres, chips, are productive and cash-generating, unlike the empty houses at the centre of the housing crisis, and the borrowers are large, profitable corporations rather than individually overleveraged subprime homeowners.

The scenario floated most often across current research is a dot-com-style equity correction layered on top of a more contained, data-centre-specific credit event, serious and potentially painful for the tech sector and private credit markets specifically, but not automatically a repeat of a global systemic banking collapse. Whether that assessment holds may depend less on the technology itself than on how far the debt underneath it has already spread, and how many people’s retirement accounts are quietly along for the ride.