Every collapse covered on this channel so far involved somebody doing something.
Silicon Valley Bank did not. Nobody invented a customer. Nobody moved money offshore. Nobody hid a loss in a shell company. The bank took deposits and bought United States Treasuries and government-backed mortgage securities, which are the safest instruments in the world.
It was the third-largest bank failure in American history, and it took less than 48 hours.
On Thursday 9 March 2023, customers attempted to withdraw $42 billion. Chief executive Greg Becker later told the Senate that worked out to roughly a million dollars a second. By the close of business the bank held a negative cash balance of about $958 million. The following morning customers asked for another $100 billion, and the bank never opened.
For context, the largest bank run in American history before this one was $19 billion spread across sixteen days.
THIS WEEK'S AUTOPSY
Silicon Valley Bank was not a normal bank, and the difference was on the liability side rather than the asset side.
Ordinary banks take small deposits from thousands of unconnected people, almost all of it federally insured to $250,000 per account. That insurance is the reason ordinary depositors do not run. There is a floor beneath them.
More than ninety percent of the money at Silicon Valley Bank was uninsured. Its depositors were startups and venture capital funds, which meant they had no floor, and they were connected to each other through the same boards, the same investors, and the same group chats. That network was the bank's greatest asset for thirty years. It was also a mechanism for everyone deciding to leave on the same morning.
When deposits tripled from $62 billion to $189 billion across 2020 and 2021, the bank had to do something with the money. It could not lend it fast enough, so it bought bonds, and it bought long ones maturing into the 2030s because they paid slightly more. Most of that portfolio was filed as held to maturity, an accounting category that lets a bank report what it paid rather than what the bonds are currently worth.
That treatment is entirely legitimate, and it holds on one condition. The bank must never need to sell.
When the Federal Reserve raised rates at the fastest pace in forty years, the bonds fell in value. By the end of 2022, the portfolio carried roughly $15 billion in unrealized losses against a bank worth about $16 billion, and none of it appeared on the income statement. Selling to meet withdrawals would have converted invisible losses into real ones. The only exit from the room was the door that proved the room was on fire.
On 8 March 2023, the bank sold $21 billion of bonds at a $1.8 billion loss and announced it needed to raise $2.25 billion. Management described it as balance sheet repositioning. The market read it as a bank that needed money.
There is a regulatory half to this that deserves stating fairly. Federal Reserve supervisors had cited the bank repeatedly for deficient liquidity risk management and inadequate interest rate modeling, and it spent roughly eight months of 2022 with no chief risk officer at all. In 2015 Becker told the Senate that banks like his did not present systemic risks and urged lawmakers to raise the threshold for stricter supervision from $50 billion to $250 billion. Congress did so in 2018. The bank failed at $209 billion, inside the gap. The rollback did not blind the regulators, who saw the problem and wrote it down. What it removed was the machinery that would have forced the bank to act on what they found.
Regulators invoked the systemic risk exception on 12 March and guaranteed every deposit. Depositors were made whole. Shareholders were not. The bank was sold to First Citizens.
Greg Becker has never been charged with a crime. Investigators examined the $3.6 million of stock he sold eleven days before the collapse under a plan filed that January, and no charges followed. The FDIC is suing him and sixteen others for gross negligence and breach of fiduciary duty. A judge has ruled the case can proceed, and it has not been decided. Becker denies wrongdoing and told the Senate he did not believe any bank could have survived a run of that velocity. Nothing here is a finding against anyone.
The lesson is the uncomfortable one. Everything that killed this bank was public. The bond portfolio was in the filings. The unrealized losses were disclosed. The concentration of uninsured deposits was published every quarter. The information was not hidden. It was boring, and boring does not trend.
THE WATCHLIST
Three companies where a small number of counterparties can move at once.
BigBear.ai illustrates how quickly concentration converts into a hole. A customer that accounted for 19% of revenue in the first quarter of 2025 accounted for zero in the second quarter of 2026. The company continues to report several customers individually above 10% of revenue. In government and defense contracting, the customer list is short by design, which is not itself a flaw, but it does mean a single non-renewal removes a visible fraction of the business in one step.
Exzeo Group carries a concentration of an unusually pure kind. Two related-party customers, affiliated with the company's controlling shareholder, accounted for approximately 85.6% of total revenue in the second quarter of 2026, down from 93.5% a year earlier. The company discloses this plainly as customer, related-party, and cash concentration risk. Where the customer and the controlling shareholder are connected, the revenue and the ownership are not independent of each other.
Terra Property Trust is the closest structural echo of this week's mechanism. The company disclosed approximately $57.9 million of debt maturing within twelve months against $9.7 million of cash and stated substantial doubt about its ability to continue as a going concern. Its assets are loans and property, which take time to convert. Its obligations have dates. That gap between how quickly assets can be sold and how quickly liabilities come due is the same gap that ended Silicon Valley Bank, at a very different scale.
None of these three is accused of any wrongdoing. They appear because their concentration and maturity disclosures are sitting in current filings, which is where this kind of risk always sits before it becomes news.
THE VULTURE'S PICK
Every issue of all until The Vulture shares one tool or resource actually used in the research. Coming in the next issue.
NEW ON YOUTUBE
The full Silicon Valley Bank autopsy is live on the channel. It covers the uninsured deposit base, the held-to-maturity portfolio, the 2015 testimony and the 2018 threshold change, the supervisory citations, and the four days in March.
Watch it here: https://youtu.be/5WVwo4gAdV4
NEXT ISSUE
A company with seven thousand stores and no debt at all until a private equity firm arrived, loaded it with $1.4 billion, and took the cash out the front door. RadioShack did not die because it missed the internet. It died because somebody was paid to make it fragile.
— The Vulture