Silicon Valley Bank served the venture capital and technology economy, and its balance sheet reflected that client base. By the end of 2022 the vast majority of its deposits were uninsured — sitting above the $250,000 FDIC guarantee — a share that regulators later put at around 94%. ² The California DFPI, which co-supervised the bank, called it the highest uninsured-deposit ratio of any U.S. bank with $50 billion or more in assets. ¹

The bank had also parked an unusually large amount of its money in bonds. Securities made up about 55% of total assets, roughly double the 25% typical of large peer banks. ¹ When the Federal Reserve raised interest rates sharply through 2022, the market value of those older, lower-yielding bonds fell.

The end came fast. On March 8, 2023, SVB announced it had sold securities at a $1.8 billion loss and would try to raise $2 billion in fresh capital. ¹ The next day depositors pulled tens of billions of dollars in hours. ³ On the morning of March 10, the DFPI took possession of the bank. ¹ What follows examines what was on SVB's books, how fast the money left, how that compares to past failures, what role technology played, and what management and its overseers knew.

The core vulnerability sat in SVB's investment portfolio. The Federal Reserve reported that its held-to-maturity securities — bonds the bank intended to keep rather than sell — had a weighted-average duration of 6.2 years, with most of the holdings in agency mortgage-backed securities maturing in 10 years or more. ² Long-duration bonds lose more value when rates rise, and SVB held far more of them than comparable banks: held-to-maturity paper made up 78% of its securities, against 42% at peers. ¹

The scale of the resulting losses is where the picture turns severe. SouthState Correspondent Division, an industry analyst, reported that the held-to-maturity portfolio stood at $91.3 billion at the end of 2022 and carried a mark-to-market loss over $15 billion — close to the bank's entire equity — a gap SouthState said made SVB economically insolvent. ¹ This portfolio total rests on a single source, but the loss figure does not: the Federal Reserve's own inspector general independently put the unrealized held-to-maturity loss at roughly $15 billion, so the two estimates agree. ¹ On the available-for-sale side, SouthState reported a $26.1 billion portfolio with a $2.4 billion adjustment, and the inspector general's figure of about $2.5 billion again lines up closely. ¹

The deposit concentration made this fragile. Sources report the uninsured share slightly differently — SouthState cited about 97% of $173 billion in deposits, the DFPI 93.8%, and the Federal Reserve around 94% — but the differences trace to different total-deposit denominators, ranging from about $166 billion to $173 billion depending on the snapshot date, not to any genuine dispute. ⁴ All agree the overwhelming majority of SVB's funding was uninsured and therefore free to flee at the first sign of trouble.

How to characterize the portfolio's maturity varies by metric. SouthState reported that roughly 56% of securities had repricing beyond 15 years, while the Federal Reserve described the majority as maturing in 10 years or more. ³ These are different yardsticks — repricing versus maturity, a 15-year versus a 10-year threshold — so they are not directly comparable, but both point to the same conclusion: a portfolio loaded with long-dated, rate-sensitive assets.

Open: Whether SVB's audited 10-K confirms the exact $91.3 billion held-to-maturity total and the 56%/15-year repricing figure, both of which rest on a single analyst source.

The March 8 announcement of a securities-sale loss and emergency capital raise was the spark. Within a day, SVB's uninsured, tightly networked client base moved to pull its money out at once. The California DFPI, which seized the bank, reported that within roughly eight hours on March 9 SVB received withdrawal requests of about $42 billion — nearly a quarter of its roughly $166 billion in total deposits. ³ The DFPI reported that the $42 billion represented nearly a quarter of SVB's roughly $166 billion in total deposits. ³

Not all of that money actually moved before the doors closed. The DFPI reported that about $36 billion was withdrawn on March 9, with roughly $6 billion still queued for processing when the bank was shut. ¹ Multiple independent authorities converge on the headline figure: the Federal Reserve reported an outflow of more than $40 billion that day, the Fed's inspector general and the Senate investigators both cited $42 billion. ⁴

The next morning would have been worse. SVB told its supervisors it expected over $100 billion in further outflows on March 10 — a figure the Fed and the Senate subcommittee both report. ⁴ No bank could meet that. By the close of business on March 9 the bank had a negative cash balance of $958 million and was insolvent; the DFPI took possession at 8:39 a.m. Pacific on March 10. ¹

Open: The minute-by-minute timing and digital channel of individual withdrawals during March 8-10, which would show precisely how the outflow accelerated after the March 8 announcement.

Placed against history, the velocity of SVB's run stands out. The DFPI reported that the largest previous U.S. bank run, at Washington Mutual during the 2008 crisis, lost $2.8 billion in deposits in a single day at its peak — about 15 times smaller than SVB's $42 billion. ² The FDIC's vice chairman independently cited the same $2.8 billion peak figure and the same roughly 15-to-1 ratio. ²

Stretch the comparison over multiple days and the gap narrows but does not close. Washington Mutual lost $16.7 billion over a nine-day run in September 2008, about 9% of the deposits it held that June — a figure corroborated across regulatory and secondary sources. ² SVB lost more than twice that amount in less than a single day, with an Oxford analysis framing it as roughly $42 billion in 24 hours against WaMu's $17 billion over 10 days. ¹ These measures are not perfectly interchangeable: WaMu's is a cumulative multi-day total, SVB's a single-day figure, and the caveat matters when reading the raw numbers side by side.

The 1984 collapse of Continental Illinois — long the textbook example of a fast, wire-driven run — offers a further benchmark. The Federal Reserve Bank of St. Louis reported that Continental lost 30% of its funding over 10 days, or seven business days, in what the FDIC described at the time as a high-speed electronic bank run. ² SVB lost about 25% of its deposits in eight hours. ⁴ Whether measured against 2008 or 1984, no earlier U.S. run approached SVB's pace.

Open: Comparable hourly withdrawal velocity and channel data for IndyMac (2008), which the brief flagged but for which no graded figures were available.

If SVB's run was historically fast, the natural question is what made it so — and here the evidence is real but thinner than elsewhere. The Federal Reserve Board concluded that the combination of social media, a highly networked and concentrated depositor base, and technology 'may have fundamentally changed the speed of bank runs'. ¹ The Fed's own hedged phrasing — 'may have' — signals that this is an assessment, not a settled measurement, and it comes from a single institutional origin.

The most specific quantitative support comes from one academic working paper. Cookson and co-authors reported that from March 8 through March 13, users posted 6,528 'run' tweets about SVB, roughly five times the volume of the next most-discussed bank. ¹ The same team found that banks in the top tier of pre-run Twitter exposure lost 6.6 percentage points more stock-market value during the run period, an effect they said was not explained by mark-to-market losses or uninsured deposits. ¹ Both findings carry a REPORTED grade: they rest on a single unpublished, un-replicated paper drawing on the authors' own dataset.

There is a competing read of depositor behavior in the record. The FDIC's staff study found that depositors with substantial uninsured funds were far more likely to run, while fully insured retail depositors generally did not — behavior consistent with rational risk-avoidance rather than undifferentiated panic. ¹ That does not contradict the technology thesis, but it shifts the emphasis: the people who ran were precisely those with money to lose, using whatever channels let them move fastest.

Open: Whether the Cookson et al. Twitter-exposure finding survives peer review and independent replication, given the technology-acceleration case leans heavily on this single working paper.; Transaction-level data separating withdrawals by digital channel, which would show whether mobile and online banking, rather than social media specifically, drove the speed.

The inconvenient part of the record is how much was known, and how early. SVB's leadership made a series of documented choices that heightened its exposure to rising rates. The DFPI reported that the bank began removing interest rate hedges in March 2022 to take profits, and by April had removed hedges that would have protected against rising rates. ¹ An academic reconstruction, corroborated at its endpoint by other sources, found SVB had held $15.3 billion in swaps in 2021, then closed out $11 billion of them in the first half of 2022, booking $517 million in gains and leaving only about $563 million in place by year-end. ¹ The Fed's inspector general later called the complete removal of hedges 'a significant error' that should have been a further red flag for supervisors. ¹

The bank also ran without a key control in place. From April 2022 to January 2023 — the most aggressive stretch of Fed tightening in decades — SVB had no chief risk officer, a roughly eight-month gap confirmed across three independent sources. ²

Regulators were not blind to any of this. As early as 2018, according to one peer-reviewed account, concerns about SVB's risk management were raised, and in 2021 Fed officials cautioned the bank's leadership about inadequate liquidity and interest rate risk management. ¹ In August 2022 examiners downgraded SVB's composite and management ratings. ¹ That November, a joint FRBSF/DFPI examination letter found the bank's interest rate risk simulations 'were not reliable,' warning they 'gave a false sense of safety in a rising rate environment' and masked the need to act earlier. ¹ By the time it failed, SVB carried 31 open supervisory findings — about triple the number at peer firms. ¹

The external gatekeeper drew scrutiny too. The Senate Permanent Subcommittee on Investigations reported that KPMG issued an unqualified audit opinion for SVB on February 24, 2023 — just 14 days before the collapse — certifying that the financial statements presented fairly in all material respects. ¹ The Fed's supervision chief, Michael Barr, summarized the official conclusion bluntly, calling the failure a 'textbook case of mismanagement', a judgment echoed by the FDIC's Travis Hill, who placed 'mismanagement of interest rate risk' at the core of the problem. ³

Open: The complete FRBSF examination history from 2018 onward, showing what deadlines, escalations and enforcement actions supervisors set against their findings — the record needed to judge whether the failure was supervisory forbearance rather than lack of awareness.; How KPMG reconciled an unqualified opinion with a held-to-maturity portfolio carrying unrealized losses near the size of the bank's equity.

The debate over SVB is not really about which facts are true — nearly all the major figures converge across independent authorities — but about which fact was decisive. Several explanations compete, and they are best understood as complementary layers rather than rivals.

One explanation places the root cause in duration mismatch and interest rate risk. On this reading, SVB concentrated its balance sheet in long-dated, fixed-rate bonds, stripped out most of its hedges to book short-term gains, and was economically insolvent before any depositor moved — its held-to-maturity losses had swollen to near the size of its equity. ⁴ The supporting record is strong and multi-sourced, and it is the conclusion the Fed and FDIC leadership reached explicitly. ³ What this explanation cannot by itself account for is why the end came in eight hours rather than over weeks.

A second reading holds that technology and social media changed the failure mode itself. Its strongest evidence is the sheer velocity: $42 billion in eight hours, roughly 15 times the largest prior single-day run, with the Fed suggesting networked depositors and digital tools had 'fundamentally changed' run speed. ⁴ But the quantitative core of this thesis — the Twitter-exposure and 'run tweet' findings — rests on a single unreplicated working paper, and the Fed's own language is provisional. ¹ The evidence establishes that the run was fast; it does not yet firmly establish that social media, as distinct from digital banking generally or the depositor concentration, was the operative cause.

A third reading emphasizes the uninsured deposit base. With around 94% of deposits uninsured — the highest ratio among large U.S. banks — SVB's funding was inherently flight-prone, and the FDIC's transaction analysis showing that uninsured depositors ran while insured retail customers did not is direct support. ⁴ This explanation ties the others together: it explains both why the run happened once losses were revealed and why it moved so fast.

A fourth reading indicts supervision. Regulators documented unreliable risk simulations, downgraded the bank, and logged 31 open findings, yet did not compel remediation before failure. ³ This is well-supported as to what regulators knew; what the graded record cannot resolve is whether stronger enforcement would have changed the outcome, because the examination and enforcement workpapers are not in evidence.

A fifth, narrower proposition treats the March 8 loss-and-capital-raise announcement as the proximate trigger. It has clear support — the $42 billion run followed directly on that disclosure — but it is a statement about timing, not underlying cause, and rests on the single sequence of events rather than depositor-decision data. ¹ The discriminating evidence for all of these — internal deposit logs by account type and channel, full examination histories, and replication of the Twitter study — is not in the graded record.

The evidence forces a layered conclusion rather than a single verdict. What is firmly established is that SVB was gravely exposed on the asset side before the run began: a bond-heavy balance sheet, long-duration holdings, hedges removed for profit, and unrealized losses that independent official and industry sources put near the size of the bank's equity. ⁴ The unanimous judgment of the Fed, FDIC, the Fed's inspector general and Senate investigators — that this was fundamentally a management failure of interest rate and liquidity risk — is strongly supported. ⁴

Equally established is that the run itself was historically unprecedented in speed: no earlier U.S. failure approached about $42 billion in eight hours, and the margin over Washington Mutual and Continental Illinois is large. ⁴ On the weight of the evidence, the run is best characterized as the mechanism that converted a solvency problem into immediate closure, not as the independent cause of failure — a reading consistent with the FDIC's finding that it was uninsured depositors, reacting to disclosed losses, who fled. ¹

The claim that social media specifically drove the record velocity remains weaker than the rest of the case. It is plausible and endorsed in hedged terms by the Fed, but its quantitative backbone is a single unreplicated study. ² Treating that mechanism as settled would overstate what the graded evidence supports. (SPECULATIVE) It is reasonable to infer that digital banking infrastructure — the ability to move large sums instantly by wire and app — was necessary for a run of this speed, since the withdrawals plainly could not have cleared at teller windows; but the evidence does not let us separate that infrastructure effect from social-media amplification specifically, and this inference is offered only as reasoning, not finding.

The supervisory record is the genuine open wound. Regulators saw and documented nearly every fatal weakness months in advance, yet the bank operated until depositors forced the issue. ³ Whether that reflects a gap in enforcement tools or a failure to use them cannot be settled without the examination workpapers, which are outside the graded record.