In 2000 Masayoshi Son lost more money than anyone in history. His net worth went from $78 billion to $8 billion in roughly two years. He'd bet everything on the dot com boom. He was wrong.

He did it again with the Vision Fund. $100 billion raised. WeWork imploded. Greensill bankrupt. Katerra bankrupt. OneWeb bankrupt. The Vision Fund lost $27 billion in a single year in 2022 — then lost another $32 billion the following year.

Then Vision Fund 2. Smaller. Same energy.

Now he's back. And this time he's committed $40 billion in an unsecured bridge loan to OpenAI.

A quick translation for the non-finance folks: a bridge loan is essentially a short term loan designed to keep things running until a bigger payday arrives — in this case the IPO. Think of it like borrowing against your expected tax refund to cover rent now. The refund is the repayment plan.

Unsecured means there's no collateral. No assets pledged against the loan. If you default on a mortgage the bank takes the house. There is no house here. If the IPO doesn't happen or doesn't raise enough Son owes $40 billion with nothing for the banks to seize and sell.

The loan matures March 2027. One repayment path: a successful OpenAI IPO.

The $40 billion is not Son's money alone. He distributed it across eight banks — JPMorgan, Goldman Sachs, Mizuho, SMBC, MUFG, HSBC, BNP Paribas and Intesa Sanpaolo. European banks. Japanese banks. American banks. Global institutions, all holding unsecured exposure to a company whose leadership just testified about undisclosed self-dealing in a federal courtroom.

Eight banks agreed to those terms. Because the IPO fee on the largest public offering in history is enormous. Enough to justify the risk.

In theory.

Before the Cerebras testimony. Before the CFO went to the Journal. Before two sets of numbers became sworn testimony.

But this time he's not alone. And the people holding the bag if he's wrong aren't all billionaires.

Here's who's in this, and how deep:

SoftBank — $40 billion unsecured, $64.6 billion total OpenAI exposure. S&P Global Ratings downgraded their credit outlook from stable to negative — the credit rating agency, not the stock index — specifically citing OpenAI concentration risk. S&P called OpenAI one of the investments with the "weakest credit quality" in SoftBank's entire portfolio. If the IPO stumbles Son has a $40 billion problem with nothing to sell and analysts estimate a $32 billion funding gap over the next two years beyond that.

Eight banks — JPMorgan, Goldman, Mizuho, SMBC, MUFG, HSBC, BNP Paribas, Intesa Sanpaolo distributed that $40 billion across their own balance sheets. Some of those banks hold your retail deposits. These are not separate companies from the ones managing your checking account.

Microsoft — holds a 27% stake in OpenAI Group PBC valued at approximately $135 billion following the October 2025 restructuring. Their entire AI market cap premium is priced on that relationship. Azure revenue growth story is partially built on OpenAI workloads. If the valuation craters on IPO their stock takes a hit that ripples through every index fund holding them. Which is essentially every retirement account in America.

A quick note on what that actually means: if you have a 401k or IRA in a broad market index fund — S&P 500, total market, target date fund — you don't choose which stocks it holds. The fund automatically holds everything in the index weighted by size. Microsoft is one of the largest components of most major indexes. When Microsoft's stock drops the fund automatically holds less value without you doing anything. You didn't choose Microsoft. You didn't choose OpenAI. You didn't choose any of this. But your retirement account adjusts anyway. That's how passive investing works at scale.

Nvidia — finalizing approximately $30 billion in equity investment — down from a non-binding $100 billion MOU signed last September that stalled after Jensen Huang privately criticized OpenAI's business discipline. The revised number is still their largest investment ever. Their valuation is the AI compute story. OpenAI is the loudest expression of that story. A stumbling IPO raises questions about whether AI compute demand is as sustainable as currently priced. Huang's private doubts suggest even insiders aren't entirely certain.

Amazon — $50 billion committed. AWS is the infrastructure layer — OpenAI runs significant workloads on Amazon's cloud, meaning Amazon gets paid whether OpenAI succeeds or fails as long as the compute keeps running. But Amazon also holds a $4 billion stake in Anthropic, OpenAI's main competitor. So Amazon is simultaneously exposed to OpenAI's downside and potentially positioned to benefit if enterprise AI spending redistributes away from OpenAI toward Anthropic. One of the eight institutions in the exposure chain is also holding the main alternative. That's not reassuring. That's complicated.

Retail investors — ARK Investment Management — the firm that runs actively managed ETFs betting on disruptive technology, popular with individual investors — holds OpenAI across three of its publicly traded funds. Fundrise — a platform that lets regular people invest in private assets — listed a closed end fund on the NYSE in March 2026 holding OpenAI, Anthropic and SpaceX. Within days shares hit $575 against a net asset value of $18.26 — a 31 times premium. Net asset value is what the underlying shares are actually worth. People were paying $575 for $18 of actual value. Citron Research called it "SIMPLE MATH" and shorted it. The stock dropped nearly 50% in a single session. Trading halts fired during the collapse — the same mechanism as GameStop. 100,000 retail investors are now locked up until September, all looking at the same exit at the same time. Fundrise had previously settled with the SEC in 2023 for paying over 200 social media influencers to promote its products without legally required disclosure. Those were regular people chasing AI exposure through a product they probably didn't fully understand. More retail products like this exist. More are coming.

Delay is the best case scenario if things go wrong. Not failure. Delay.

The CFO already wants to push to 2027. The trial is generating new risk factors daily. Every institution holding exposure has to reprice for a delayed timeline. Repricing simply means adjusting what something is worth on paper to reflect new information — when the IPO timeline slips, the implied value of everything connected to it gets revised downward accordingly.

SoftBank's $40 billion matures March 2027. A delayed IPO means Son needs to either refinance — essentially take out a new loan to pay off the old one, almost certainly at worse terms given the current risk picture — or find another path. Both options are expensive. Both options require the eight banks to agree.

Microsoft's 27% stake has a lockup period — meaning they can't just sell it whenever they want. They're committed to holding through the IPO process. Every month of delay is a month their AI premium sits unrealized on the balance sheet. That's not a crisis. That's a drag. But drags have a way of becoming crises when earnings calls start and analysts want answers.

For retail investors holding OpenAI exposure through ETFs and closed end funds: delay means the underlying asset stays illiquid longer. Illiquid means you can't easily sell it — unlike a stock you can dump on a Tuesday afternoon, a stake in a private company waiting for an IPO is essentially frozen until the IPO happens. The gap between the fund's net asset value — what the underlying assets are actually worth — and what investors paid for it doesn't close until there's a liquidity event. The Fundrise situation — 31x NAV before the correction — is a preview of what happens when the liquidity event gets pushed and reality reasserts itself all at once.

Nobody is saying the IPO fails. This section is the speculation clearly labeled as such.

But if it does:

Son's $40 billion unsecured loan has no collateral to liquidate. The eight banks take a distributed loss on exposure they underwrote expecting IPO fee income to offset the risk. JPMorgan is both the lead underwriter and one of the largest mortgage servicers in America. These aren't separate companies. They're different floors of the same building.

Microsoft's AI premium evaporates. Their stock reprices. Every index fund holding Microsoft — which is most of them — adjusts. Your 401k holds index funds. The math is uncomfortable.

Nvidia's demand story wobbles. The $30 billion commitment was a bet on OpenAI's growth trajectory. A failed IPO raises questions about every AI infrastructure commitment across the industry. AMD, Oracle, CoreWeave all have exposure to the same narrative.

The Cerebras IPO — filed at $26.6 billion, valuation built largely on the OpenAI deal — gets repriced by association. The two executives whose undisclosed stakes fueled that valuation are now subjects of potential regulatory inquiry. Cerebras investors didn't sign up for that.

And then there's the question the series has been asking since the fuse post: the burn rate is $25 billion this year and climbing. The IPO was supposed to be the answer to that math. Without it, what's the plan?

Most people remember GameStop. January 2021. Reddit traders squeezed the short sellers — investors who had bet the stock would go down by borrowing shares, selling them, and planning to buy them back cheaper later. When the stock shot up instead those short sellers had to buy at any price to cover their losses, which pushed the price even higher. The stock went from $20 to nearly $500. Circuit breakers fired on the way up AND the way down — halting trading when the moves got too fast in either direction. Robinhood halted buying entirely to protect institutional investors. The app froze when people needed it most. The broader market got dragged into correlated selling as hedge funds liquidated unrelated positions to cover losses. Regular people caught in the middle of institutional decisions they had no visibility into. Staring at a loading screen while the number moved.

That was one meme stock. One pressure point. The system strained for a week and held. The Fundrise collapse six weeks ago was a preview in miniature — trading halts, retail investors locked up with no exit, a 50% single session drop. That was one fund. One private company proxy vehicle.

This exposure chain is Microsoft, Nvidia, Amazon, eight global banks, SoftBank, retail ETFs, and a $40 billion unsecured loan maturing in March 2027 with one repayment path.

If that goes sideways this might be more like Alderaan.

The comparison that gets made most often is dot com. And it's not wrong. 2000 had irrational exuberance about unproven technology. Overvalued companies burning cash with no path to profitability. A correction that was painful but relatively contained — investors who chose to be in it absorbed most of the loss. The companies were largely fake. The writedowns hurt but they were writing down things that never had real value.

OpenAI's technology genuinely works. ChatGPT has 900 million weekly active users. The product is real. The demand is real. The infrastructure is real. So the dot com parallel covers the valuation bubble — overpriced, burn rate unsustainable, market correction possible — but it doesn't cover what happens to people who never made a single investment decision related to any of this.

For that the more uncomfortable analogy is 2008.

This is not a prediction of another financial crisis. That needs to be said clearly and explicitly. What follows is a structural comparison not a forecast.

In 2008 the mechanism was: banks issued mortgages, packaged them as securities, those securities got AAA ratings, institutional investors bought them, those institutions borrowed against them, the whole stack depended on housing prices continuing to rise, and when they didn't the circularity unwound in ways nobody had fully mapped. The underlying asset — the mortgage — was real. The financial engineering around it created systemic exposure that reached people who never made a single decision related to mortgage backed securities. They just had a savings account. A 401k. A mortgage with a servicer that needed a government bailout.

The circular funding structure here has structural similarities worth naming carefully. OpenAI raises at an $852 billion valuation. Uses capital to sign compute deals with companies — Cerebras, Helion — whose leadership holds undisclosed stakes. Those deals inflate vendor valuations. Vendors file IPOs. OpenAI's own IPO is supposed to validate the whole stack. SoftBank's $40 billion unsecured loan gets repaid. Eight banks collect their fees. Retail investors get liquidity.

Every step depends on the next step working.

That's not dot com energy. Dot com companies were mostly standalone failures. This is interconnected institutional exposure that, if it unwinds, unwinds through the same channels that reach your index fund, your savings bank, your mortgage servicer.

Elizabeth Holmes didn't go to prison for building a bad product. She went for telling investors the product worked when it didn't. OpenAI's product works. The question the trial is now forcing into the public record is whether the financial disclosures around it have been telling investors the whole story.

That's TheranasAI. Not because the product is fake. Because the investment case might be built on something that looks very different under oath than it does in a press release. And because the structure around it, if it goes wrong, doesn't stay contained to the people who chose to be in it.

If you have a 401k invested in index funds you almost certainly hold Microsoft. Possibly Nvidia. Possibly funds with OpenAI exposure you didn't choose.

If you have savings in any of the eight banks that distributed SoftBank's loan you are technically downstream of this exposure. Not directly. Not obviously. FDIC insurance covers deposits up to $250,000 per depositor per bank — your retail deposits are protected to that limit. But the FDIC's entire insurance fund holds roughly $120-130 billion. It exists to handle regional bank failures. Not correlated stress across multiple global institutions simultaneously — meaning when several large institutions take losses at the same time for the same underlying reason, which is a very different problem than one bank failing on its own. The real backstop for that scenario is the Federal Reserve and Congress. Which is what happened in 2008. Which required taxpayers. It's worth noting that the current administration doesn't always respond to institutional stress in predictable ways. That's not a political statement. It's an observable fact worth factoring in.

If you have a mortgage with JPMorgan — which is both a lead IPO underwriter and the largest mortgage servicer in America — you are doing business with an institution that has material exposure to this outcome.

None of this is certain. All of it is possible. And none of the people it would affect were in that Oakland courtroom this week watching Musk admit he doesn't know what TL;DR means while OpenAI's own president testified about undisclosed self-dealing under oath.

The trial started as a billionaire grudge match. A bitter founder suing the company he walked away from because it became worth $850 billion without him.

It became something else when Brockman's journals hit the screen. When the Cerebras conflict landed under oath. When the CFO went to the Journal. When the SEC settled for one cent on the dollar on the same Monday.

The judge in Oakland is presiding over what started as a contract dispute. The scope of what she's holding may extend further than the immediate case suggests — a thread that, pulled far enough, connects to the retirement accounts, savings deposits, and mortgage servicers of people who have never heard of OpenAI and have no idea any of this is happening.

The series has been saying from the beginning: read the terms. They're more honest than the marketing.

The terms right now are: $40 billion unsecured, no collateral, one repayment path, undisclosed self-dealing in the sworn record, two sets of numbers, a CFO who went public, and a trial that's adding new risk factors daily.

The marketing says the IPO is on track.