SVB Didn't Die From Bad Loans. It Died Because All Its Depositors Knew Each Other.

On March 8, 2023, Silicon Valley Bank told the market two ordinary-sounding things: it had taken a $1.8 billion loss selling government bonds, and it would raise $2.25 billion in fresh capital to fill the hole.⁸ By the next afternoon, more than $40 billion had walked out the door.³ On the morning of March 10, California regulators closed the bank and handed it to the FDIC.² Forty-eight hours, start to finish. The most famous bank in technology - the one that banked the people who built the future - died at the speed of a group chat.

The story everyone tells is that SVB made a reckless bet on startups and the bet went bad. Almost none of that is right. SVB did not fail because startups stopped paying it back; its loan book was fine. It failed because the very thing that made it special - a single, tight, interconnected tribe of customers - turned out to be the thing that could kill it in an afternoon.

The wound was on the asset side. The lethality was on the deposit side.

Start with what actually broke. During the 2020-21 funding boom, SVB's deposits exploded - from $50 billion in 2018 to roughly $191 billion in 2021, growth of around 280% while the entire insured banking system grew about 30%.⁷ Flooded with cash, the bank parked an enormous share of it in long-dated bonds: $91.3 billion classified as held-to-maturity at year-end 2022, a portfolio that as a share of total securities ran nearly double the peer average.³ Then management did the thing the Federal Reserve later singled out: it removed its interest-rate hedges to chase short-run profit, leaving the bank naked to exactly the risk that arrived - rising rates.⁴ When rates climbed, those long bonds lost market value. Selling $21 billion of the available-for-sale slice crystallized a roughly $2 billion loss.⁷ That was the wound.

But a wound is not a death. Plenty of banks carried underwater bond books in 2023 and are still standing. What made SVB's wound fatal was who held its money. VC-backed companies accounted for more than half of SVB's deposits, and a stunning 94% of its domestic deposits sat above the $250,000 FDIC insurance line - against roughly 47% at most large banks.³⁶ Translate that: nearly all of SVB's depositors had real money to lose and a powerful reason to run at the first sign of trouble. They did not run because they ran the numbers on the bond portfolio. They ran because the person next to them in the same Slack channel did.

A bank run needs a crowd. Concentration handed SVB a coordinated one.

Here is the mechanism, worked all the way down. A classic bank run is a coordination problem: everyone is safe if nobody panics, and everyone loses if too many do, so each depositor watches the others. In a normal bank, the depositors are strangers - a retiree in Ohio, a contractor in Texas, a dentist in Oregon - who don't know each other, can't see each other's moves, and react at the lazy speed of human rumor. The crowd is slow because the crowd is diffuse. SVB had the opposite. Its customers were one tribe: founders funded by the same handful of venture firms, advised by the same partners, talking on the same group threads. When a few prominent VCs reportedly told their portfolio companies to pull cash, that instruction didn't trickle - it broadcast.⁸ The depositor base was pre-wired for a run because it was a network, and a network transmits panic the way a power grid transmits a surge: instantly, everywhere, at once.

There's a deeper layer the 'startup bank' label hides. SVB wasn't even primarily lending to fragile startups - by end of 2022, 56% of its loans went to VC and private equity firms, secured by their limited partners' capital commitments.⁶ It custodied assets for around 1,074 PE and VC funds.⁶ So the same small set of venture firms sat on both sides of the balance sheet: they were the bank's borrowers and, through their portfolio companies, the source of its deposits. The UK regulator had flagged exactly this overlap of clients on both the asset and liability sides as a concentration risk and an area of supervisory focus for 18 to 24 months before the failure.⁷ The tribe wasn't a customer segment. It was a single point of failure wearing thousands of logos.

Wasn't this just bad rate management - and wouldn't the bonds have sunk it anyway?

The honest objection is that the bonds and the missing hedges were the real killers, and concentration is just where the camera happened to point. There's truth in it. The Federal Reserve was blunt: SVB managed interest-rate risk for short-run profit, pulled its hedges, and ran a business so concentrated and so reliant on uninsured deposits that it was a genuine outlier - with 31 open supervisory findings at the time it collapsed.⁴ Regulators saw the rate problem and failed to force a fix. So yes, the asset side built the bomb. But a bomb still needs a detonator, and here the counterfactual is the whole point: a bank with the same underwater bond book and a diffuse, mostly-insured retail depositor base would have had months to recapitalize, shrink, or be quietly acquired. It would have bled, not hemorrhaged. SVB got hours. The concentration didn't cause the loss - it removed the bank's time to survive the loss. Time is the asset a slow run leaves you and a coordinated one takes away.

When the dust settled, the FDIC moved most of what was left - roughly $72 billion in loans and $56 billion in deposits - to First-Citizens, and pegged the loss to its insurance fund at $16.1 billion, among the largest it has ever absorbed.⁵ Note the proportions: the loans, the thing everyone blamed, sold for tens of billions. The damage came from the run, not the book. SVB spent two decades building the most enviable customer relationship in finance - the bank the whole venture economy trusted by default. That trust was its moat, and it was its undoing, because a moat full of people who all think alike empties in a single tide. The bank that knew everyone died of being known by everyone at once.