Silicon Valley Bank: The Bank That Ran A Duration Mismatch Between Its Deposits And Its Bonds

Silicon Valley Bank did not die from tech. Tech is still here. Every startup that had money at SVB on March 9, 2023 is still operating. Every venture fund that told its portfolio companies to pull their money that morning is still investing. The tech industry SVB served for forty years is larger today than it was on March 10, 2023.

SVB died from the specific mistake a bank makes when it borrows short and lends long, and then hides the mismatch inside a regulatory accounting bucket that lets it pretend the mismatch does not exist. Depositors could call their money back any Wednesday. Bonds could not be sold without recognizing a loss. When one side of the mismatch moved, the other side could not respond in time. The whole thing unwound in 48 hours.

The verdict. SVB ran the Two Cancers spiral at $175 billion scale in about 24 months. Cancer 1 was the interest-rate duration obligation on $91.3 billion of held-to-maturity bonds, hidden by accounting rules from the P&L but real on the balance sheet as more than $15 billion in unrealized losses by the end of 2022. Cancer 2 was the widening Working Capital Gap between liquid assets the bank could actually deploy against withdrawals and callable liabilities that could walk out the door any morning. The trigger was the March 8 announcement that SVB had sold $21 billion of available-for-sale securities at a $1.8 billion loss and needed to raise $2.25 billion of fresh equity. The market read the announcement as a solvency signal, not a liquidity one. Twenty-four hours later, $42 billion walked out. Thirty-six hours after that, the FDIC took the keys. The doctrine reads the mechanism before the trigger arrives. The accounting layer only reads it after the trigger has fired.

The Rise

Silicon Valley Bank was founded in Santa Clara, California in 1983 by Bill Biggerstaff and Robert Medearis over a poker game. The pitch was simple. Tech startups were being underserved by traditional commercial banks that did not understand how to underwrite a business with no revenue, no assets, and a Series A term sheet. A bank that specialized in that customer, that understood the venture cycle, that could offer venture debt alongside deposit services, would own the category. For forty years, that bet worked.

By 2020 SVB was banking roughly half of all US venture-backed technology and life sciences companies. It provided commercial banking, wealth management, venture debt, and private banking services to founders, funds, and portfolio companies across the entire tech ecosystem. When a startup raised a round, the wire hit SVB. When a fund closed, the fund's operating account was at SVB. When a founder wanted a mortgage on a Palo Alto house they could not qualify for at Chase, SVB wrote the mortgage. The bank was the plumbing of the tech industry.

Then the pandemic happened. And the deposit inflow that had been steady for four decades turned into a firehose. Federal stimulus, zero interest rates, and a venture capital cycle that pushed unprecedented amounts of capital into private companies all landed in tech bank accounts. SVB deposits grew from $62 billion in March 2020 to $124 billion in March 2021 to $175.5 billion by December 2022. Nearly tripled in three years. Roughly half of every venture capital dollar deployed in the United States during that period was sitting in an SVB account within days of leaving the fund.

A regional commercial bank grew into the sixteenth-largest bank in the United States on the back of a three-year deposit surge that had no historical precedent. Total assets reached $209 billion by the end of 2022. The bank was not just growing. It was ballooning. And the balance sheet management decisions that got made during that ballooning are the whole story of what came next.

SVB took the deposit inflow and did what banks do. It invested it. Loans were the natural first destination, but SVB's customer base was startups that were not asking for traditional commercial loans at the volume that could absorb the deposit growth. So SVB deployed a huge portion of the inflow into securities. Government bonds. Agency mortgage-backed securities. Long-dated. Held-to-maturity. Yields around 1.5 to 2 percent, which looked attractive against the near-zero rates the deposits themselves were paying. By the end of 2022 the bank held $117 billion in bonds, $91.3 billion of which sat in the held-to-maturity bucket and $26 billion of which was classified as available-for-sale.

The Federal Reserve was signaling the entire time that rates were going up. Then, starting in March 2022, they went up faster than at any point in forty years. Five hundred basis points across ten months. The bond portfolio SVB had built at near-zero yields lost enormous value the moment rates moved. And that is where the accounting layer took over from the operating layer.

The Fracture

Under US bank accounting rules, securities classified as held-to-maturity do not get marked to market on the P&L. If a bank commits to hold a bond until it matures, the bond stays on the balance sheet at book value even if rising rates have driven the market price down. The unrealized loss exists. It is reported in the footnotes. It does not touch reported earnings or equity in the same way an available-for-sale loss would.

This accounting treatment exists for a defensible reason. A bank that genuinely holds a bond to maturity will receive par at redemption regardless of what happened to the market price in between. Mark-to-market accounting for a truly locked-in portfolio would create earnings volatility that does not reflect economic reality. The rule works when the assumption underneath the classification is honored: that the bank has enough other liquid assets and stable funding to actually hold the HTM bonds to maturity without ever needing to sell them.

By the end of 2022 SVB's held-to-maturity portfolio had more than $15 billion in unrealized losses against a total bank equity base that regulatory filings placed around $16 billion. In economic terms, the bank was arguably underwater. In reported terms, the bank was well capitalized. Both statements were true simultaneously under the accounting framework that governed the disclosure. The regulators reading the call reports could see the HTM losses in the footnotes. The market reading the earnings releases could see them if they knew where to look. The depositors making a decision about where to keep their operating cash could not see them at all.

Meanwhile, on the funding side, the venture capital cycle that had produced the deposit surge went into reverse. 2022 became the year private company valuations reset, growth-stage rounds became difficult, and portfolio companies started drawing down their SVB balances to fund operations rather than adding new capital to them. Deposit inflow slowed. Deposit outflow accelerated. The deposit base at SVB, which had grown roughly $50 billion in 2021 alone, plateaued in the first half of 2022 and started falling by the second half.

The bank now had a problem it could see and could not solve. Deposits were leaving. Bonds could not be sold at book value because rates had moved. The only way to fund the withdrawals was either to sell available-for-sale bonds and recognize the loss on the P&L, borrow at rates that would compress the net interest margin, or raise fresh equity. Every option was expensive. The bank chose to keep the machine running as long as possible and hope the Fed pivoted before it broke.

The 48-Hour Run

On Wednesday, March 8, 2023 SVB released a press statement disclosing three things at once. It had sold $21 billion of available-for-sale securities and taken a $1.8 billion loss. It had arranged a $15 billion term borrowing to fund additional liquidity needs. And it was launching an emergency $2.25 billion equity raise, including a $500 million commitment from General Atlantic.

The market read the announcement in real time. A bank that is genuinely fine does not sell $21 billion of bonds at a $1.8 billion loss and simultaneously raise $2.25 billion of dilutive equity on 24 hours' notice. The disclosure was intended to be reassuring. What it actually signaled was that management had concluded the deposit base was leaving and there was no other way to fund the outflow.

The venture capital ecosystem responded within hours. Peter Thiel's Founders Fund reportedly advised its portfolio companies to withdraw all their SVB deposits on Thursday morning. Y Combinator sent similar guidance to its portfolio. General partners at other funds started calling their portfolio companies. Group chats among founders on Slack and Signal filled with screenshots of wire transfer confirmations. The venture capital industry, which had spent forty years concentrating its operating deposits at one bank, now spent 24 hours coordinating a withdrawal of those deposits.

On Thursday, March 9, depositors pulled $42 billion in a single trading session. That is not a typo. Forty-two billion dollars. In one day. Roughly a quarter of the entire deposit base. By the close of business on March 9, SVB was reporting a negative cash balance of approximately $958 million. The bank could not meet the withdrawal requests already queued for the following morning. An additional $100 billion in withdrawal requests was expected to hit on Friday, March 10.

The FDIC did not wait for Friday's opening bell. Regulators from the California Department of Financial Protection and Innovation closed the bank Friday morning and appointed the FDIC as receiver. The largest bank failure since Washington Mutual in 2008. The third-largest bank failure in the history of the United States. Forty years of franchise value collapsed in 48 hours because the mismatch between the maturity of the liabilities and the maturity of the assets became visible at the same moment the depositor base had the network coordination to act on it.

The Doctrine Overlay

Read the SVB collapse against the doctrine and the mechanism is the exact configuration the Two Cancers doctrine predicts for a business that treats a temporary capital surplus as if it were durable working capital.

Cancer 1 was unmeasured. Debt service in a traditional operating business is interest and principal on notes and lines of credit, a fixed monthly cash outflow that the business either meets from operating profit or covers by drawing on working capital. In a bank, Cancer 1 takes a different shape. It is the interest-rate duration obligation on the asset side of the balance sheet against the callability of the liability side. SVB's Cancer 1 was the certainty that a $91.3 billion portfolio of long-dated bonds locked in at 1.5 to 2 percent yields would develop enormous unrealized losses if rates rose meaningfully, and the further certainty that those losses would be economically real even if the accounting rules let the bank hide them from reported earnings. The bank was not sizing Cancer 1 against its actual funding stability. It was sizing Cancer 1 against a regulatory disclosure requirement, which is not the same thing. In a normal SMB, Cancer 1 is $6,900 a month in debt service on a $340,000 note that shows up on the P&L every month. In SVB, Cancer 1 was $15 billion in unrealized losses that never showed up on the P&L until the bonds actually had to be sold. Same mechanism. Different scale. Different disclosure regime.

Cancer 2 accelerated when the funding base concentration turned into a coordination problem. This is the counterintuitive piece for a bank whose competitive advantage was its concentration in one industry. Forty years of building the tech industry's bank meant SVB had roughly 94 percent of its deposits above the FDIC insurance limit and roughly all of them tied to a single industry's capital cycle. When that industry's capital cycle turned in 2022, deposits started leaving. When the industry's key network operators, the venture capitalists themselves, coordinated on Slack and Signal in March 2023, deposits left simultaneously. A $4 million contractor whose top three customers are all in the same industry has the same concentration problem at SMB scale. When that industry has a rough quarter, all three receivable accounts age at once. The Working Capital Gap opens instantly. Same mechanism. Different scale.

The trigger looked like a solvency event because the liquidity event exposed the solvency event underneath. Banks fail from liquidity, not solvency, in the technical sense. But a liquidity crisis at a bank with $15 billion in hidden bond losses on a $16 billion equity base is functionally a solvency crisis dressed up as a run. The March 8 announcement did not create the solvency problem. It disclosed it. The $42 billion run on March 9 was not the cause of the failure. It was the mechanism by which the failure was executed. The doctrine reads a bank the same way it reads an SMB. If the liquidity that funds the operating cycle is not durable, and the balance sheet has unrealized losses that would surface if the liquidity had to be replaced, the business is closer to failure than the P&L suggests. SVB proved that at $175 billion scale in 48 hours.

The retreat sequence was too late by the time it started. Emergency asset sales at a loss to fund withdrawals plus a dilutive equity raise to shore up capital plus a hoped-for market rally to bail the balance sheet out is not a plan. It is the last stop before the receiver gets called. The intervention that would have worked was one Greg Becker and the SVB board did not make in 2021 or 2022, when there was still time to make it. The doctrine has a name for the point at which reversal is no longer possible. It is when the working capital reserve is smaller than the cost of stopping the bleeding. SVB reached that point somewhere in late 2022 and did not know it had crossed the line until March 8.

The Intervention That Would Have Worked

Every autopsy in the archive is a business where an earlier intervention would have changed the outcome. SVB is no different. The intervention that would have worked was a doctrine question the board and the ALCO committee did not ask in 2021.

How much of the deposit surge was durable working capital that would stay with the bank through a normal downturn, and how much was temporary capital surplus that would leave the moment the venture cycle turned.

If SVB had assumed that 40 percent of the $60 billion deposit surge from 2020 to 2021 was durable and 60 percent was temporary, the bank would have deployed a much smaller share of the inflow into long-duration held-to-maturity securities and would have held a much larger share in short-duration Treasury bills and cash equivalents. The yield giveaway on the temporary portion would have been meaningful in 2021, maybe 100 to 150 basis points on tens of billions of dollars in deposits. That is real money in the short run and it would have compressed reported earnings. It also would have kept the bank alive through 2023.

The doctrine question that no one asked was the same question a $6 million contractor should ask when a big customer pays 30 days early and the operating account looks flush. Is this durable working capital or is this a timing benefit that will reverse next quarter. If the answer is that the surge is timing, the operating decision is to hold the cash short and give up the yield. If the answer is that it is durable, the operating decision is to deploy it into a longer-duration use with a better return. The bank version of this question is called asset-liability management, and it is exactly what the bank's ALCO committee was chartered to do. SVB's ALCO committee had no chief risk officer for eight months in 2022, during the exact period when the interest rate environment was changing under them. The doctrine question did not get asked because the person whose job it was to ask it was not in the seat.

In the post-collapse period, First Citizens BancShares acquired SVB's commercial banking business from the FDIC receivership on March 27, 2023. First Citizens took about $119 billion in deposits and $72 billion in loans, purchased at a $16.5 billion discount. Approximately $90 billion in securities were left in the receivership because those were the bonds that had caused the problem in the first place. First Citizens continues to operate the acquired business today under the Silicon Valley Bank name. The commercial banking franchise, stripped of the bond portfolio, was viable. The bond portfolio, stripped of the deposit funding, was worked down through the receivership. Same lesson at $175 billion as the lesson at $4 million. The operating business was fine. The balance sheet mismatch killed the operating business.

The Lesson For SMB Owners

Most SMB owners reading this autopsy are not running a $175 billion regional bank. Most are running a $2 million to $8 million business with a bank line of credit, a customer concentration they know about, and a working capital cycle they read every Friday morning by looking at the operating account. The SVB failure looks unlike an SMB failure at first glance because the numbers are so much bigger. Read it again with the doctrine lens on and the mechanism is exactly the same.

Lesson one: Do not treat temporary capital surplus as durable working capital. A big customer pays 30 days early and the operating account looks flush. An insurance settlement lands and the balance jumps by six figures. A rebate check arrives in Q4 and adds to year-end cash. Each of these is temporary capital. The doctrine question is whether the operating decisions made against that temporary capital will still work when the cash normalizes back to trend. SVB deployed a three-year deposit surge that was 60 percent driven by pandemic-era venture activity into a ten-year bond portfolio. When the deposit surge normalized, the bond portfolio was stranded. An SMB owner who takes a temporary cash surge and uses it to hire two additional employees at $85,000 a year each is running the same mistake at SMB scale. The cash normalizes. The salaries do not.

Lesson two: Concentration risk is invisible until it becomes coordination risk. SVB banked half the venture capital industry for forty years and treated the concentration as a competitive advantage. It was, until the day the concentration became a coordination problem and every VC in the country told their portfolio companies to pull money on the same Thursday morning. A $6 million contractor whose top three customers all work in the same industrial sector has the same profile. The concentration produces higher gross margins in good times because the contractor knows the industry, the pricing, and the delivery cycle intimately. When the industry has a bad quarter, the concentration produces simultaneous receivable aging across every account, and the Working Capital Gap opens instantly. The doctrine reads customer concentration in the biomarkers. The accounting layer treats concentration as either a strength or a footnote, never as the sizing problem it actually is.

Lesson three: If the disclosure regime lets you hide the losses, you will hide them from yourself first. Held-to-maturity accounting let SVB avoid recognizing $15 billion of unrealized bond losses in reported earnings. The rule was not the cause of the failure. Management's willingness to accept the reporting relief without stress-testing what happened if the losses actually had to be crystallized was the cause. The SMB equivalent is the owner whose balance sheet carries $180,000 of obsolete inventory at cost because nobody has run the writedown analysis in three years. The loss is real. The balance sheet does not show it. When the line of credit gets called and the inventory has to be liquidated, the loss surfaces at exactly the moment the working capital reserve has to absorb it. Same mechanism. Different accounting rule.

Lesson four: A run happens in the time it takes trust to break, not in the time it takes the P&L to catch up. SVB's actual solvency did not change between March 7 and March 10. The bonds were worth what they were worth. The loans were what they were. What changed was the depositors' willingness to leave their money in the bank. That change happened in 24 hours because the venture capital ecosystem had the coordination infrastructure to move together. An SMB with a strong balance sheet and one lawsuit that goes viral on LinkedIn can lose its top ten customers in a week. Trust is not on the balance sheet. When trust breaks, the balance sheet stops mattering. The doctrine reads trust as a leading indicator. The accounting layer does not read it at all.

Postscript

Greg Becker was the CEO who signed the March 8 disclosure and watched the bank fail 48 hours later. On February 27, 2023, thirteen days before the collapse, he sold 12,451 shares of SVB Financial Group stock for approximately $3.6 million under a 10b5-1 trading plan he had filed with the SEC on January 26 of the same year. Twenty-nine days between the filing of the trading plan and the execution of the sale. Forty-one days between the filing of the trading plan and the failure of the bank. Over the prior two years, SVB executives and directors had cashed out approximately $84 million in company stock. Becker was terminated as CEO on the day the bank was seized. He has not returned to a financial services executive role.

The FDIC estimated the total cost to the Deposit Insurance Fund at $20 billion, of which approximately $18 billion was to cover the uninsured deposits that a March 12 Treasury, Federal Reserve, and FDIC joint statement decided to make whole under the systemic risk exception. Depositors did not lose money. The DIF, which is funded by assessments on the broader banking industry, absorbed the loss. Every US bank paid a special assessment in 2023 and 2024 to replenish the DIF. The tech industry that had banked at SVB continued operating without missing payroll. First Citizens continues to operate the commercial banking business under the SVB name.

The Federal Reserve's own review, published in April 2023, identified four proximate causes: SVB failed to manage its interest rate and liquidity risk, its business model concentrated the customer base and deposit funding, it grew rapidly and its risk management did not keep pace, and Federal Reserve supervisors identified deficiencies but did not sufficiently act on them. Each of those is a defensible causal reading. The doctrine reads it more simply. A bank borrowed short and lent long, hid the resulting duration mismatch inside an accounting bucket, and got exposed the moment the funding base concentration turned into a coordinated run. The failure was not surprising in retrospect. It was surprising only to people who had been trained to read the P&L and not the balance sheet.

Bill Biggerstaff and Robert Medearis, the two co-founders who launched SVB over a poker game in 1983, were both retired long before the collapse. Neither survived to see it. Biggerstaff died in 2020, Medearis in 2015. Both spent their careers arguing that a bank could serve one industry very well and grow into a durable franchise. They were right. What killed the bank was not the concentration in tech. It was the failure to size the balance sheet for the specific coordination risk that the concentration created.

The tech industry SVB served is larger today than it was on March 10, 2023. The venture capital funds that pulled deposits that Thursday are still deploying capital. The founders whose payroll almost missed on Monday morning are still shipping product. The system that SVB was built to serve absorbed the failure and moved on. The bank did not.

The next SMB owner reading this autopsy who is running with a temporary cash surge deployed into a durable operating commitment, or a customer concentration that has never been stress-tested, or an accounting treatment that lets them hide a real loss until the day it has to be recognized, has the same question in front of them that Greg Becker and the SVB board did not answer in 2021. Is this working capital durable, or is it borrowed time. The doctrine reads the question. The accounting layer will not surface it. Every quarter the question goes unanswered is a quarter of the reserve gone. When the run comes, it comes in hours.