The venture power law: why Sequoia's $150 million FTX loss was a win and SoftBank's $14 billion WeWork loss nearly broke the firm, 2026 edition

Two of the highest-profile venture losses of the last decade look, on the surface, almost identical: a legendary investor makes a large, high-conviction bet on a category-defining founder, and the company implodes. Sequoia backed Sam Bankman-Fried's FTX. SoftBank backed Adam Neumann's WeWork. Both investors were embarrassed publicly. Only one of them nearly took the entire firm down with it. The difference was not conviction, temperament, or diligence, all three were arguably comparable across both firms. The difference was arithmetic: whether the losing position was sized as a fraction of a genuinely diversified portfolio built to survive most of its bets failing, or sized and leveraged as though it were a bet the fund could not afford to lose.

The power law is not a metaphor, it is a measured distribution

Venture capital returns do not cluster around an average the way most financial assets do. They follow what statisticians call a power law: a small number of extreme outliers account for nearly all the value, while the bulk of outcomes cluster near zero. Correlation Ventures, analyzing more than 21,000 US venture financings between 2004 and 2018, found that roughly 65 percent of investments fail to return the capital invested at all, while fewer than 4 percent return more than 10 times the money, and the top 0.4 percent of deals return more than 50 times. A separate long-running study by Horsley Bridge, a limited partner across dozens of venture funds since the 1970s, found that just 6 percent of investments generated roughly 60 percent of total industry returns. Academic work fitting this data to a formal power law distribution estimates a shape parameter, alpha, near 2.05, mathematically confirming what practitioners have described qualitatively for decades: there is no meaningful average outcome in venture investing, only a small number of enormous wins carrying an overwhelming majority of failures.

This is the single fact every strategy on this page is a response to. A power law return distribution rewards a specific, counterintuitive portfolio construction: make enough individually-sized bets that at least one can become the outlier, and size each bet small enough relative to the total fund that the other 65 percent failing outright does not threaten the fund's survival. Diversification in venture capital is not primarily about smoothing returns, the way it is in Ray Dalio's framework covered elsewhere on this site. It is about buying enough lottery tickets that the one or two winning numbers are statistically likely to be somewhere in the portfolio.

Sequoia and the discipline of sizing for the power law

Sequoia Capital's handling of its FTX position is, in retrospect, closer to a case study in correct power law discipline than a cautionary tale, however the headlines read in November 2022. The firm invested a combined $150 to $214 million across FTX's international and US entities, and when FTX collapsed amid Sam Bankman-Fried's fraud, Sequoia wrote the entire position down to zero within days, apologizing to investors and acknowledging it had been misled. The detail that matters is what Sequoia disclosed alongside the write-down: that FTX represented under 3 percent of the committed capital of the fund it sat in, a fund Sequoia said had roughly $7.5 billion in realized and unrealized gains at the time. Losing $150 million on a single position sounds catastrophic in isolation. Inside a fund built to expect the majority of its positions to fail, and sized so that no single failure could threaten the whole, it is close to exactly what the model predicts should happen regularly.

Founders Fund, built by Peter Thiel, Ken Howery and Luke Nosek since 2005, runs a more concentrated version of the same underlying logic, deliberately making fewer, larger bets than a broadly diversified fund like Sequoia, reasoning articulated in Thiel's own writing that venture returns are so extremely concentrated in the biggest outliers that spreading capital too thin actually dilutes a fund's exposure to the handful of outcomes that will determine its performance. The firm was the first institutional investor in both SpaceX and Palantir, two of the largest venture outcomes in history, and closed a $4.6 billion AI and defense-focused growth fund in mid-2026, pushing total assets under management past $20 billion on a thesis concentrated even more narrowly than its historical average. Founders Fund's approach demonstrates that power law investing does not require Sequoia's degree of diversification to work, but it does require the same underlying discipline: every position, however large, has to be sized against a fund structure built to absorb it failing, not against the fund's ability to survive that specific bet going to zero.

SoftBank and what happens when the same logic loses its discipline

Masayoshi Son's SoftBank Vision Fund pursued outcomes on the same scale Founders Fund targets, but without the sizing discipline that made both Sequoia's and Founders Fund's models survivable. Son deployed capital in checks often exceeding $1 billion into individual companies, reportedly telling founders to think bigger and take more money than they had asked for, a strategy that produced genuine successes, including an early stake in Alibaba that generated tens of billions in returns, alongside catastrophic ones. WeWork alone cost the Vision Fund more than $14 billion in cumulative losses before SoftBank wrote the position to zero following WeWork's 2023 bankruptcy, a single company's failure representing a loss larger than Sequoia's entire disclosed FTX exposure by a factor of nearly 100. The Vision Fund posted a record $17.7 billion annual loss in fiscal 2019 driven substantially by WeWork and Uber write-downs, and a further $32 billion loss in fiscal 2022 as a broader downturn hit portfolio valuations across the board, the worst annual performance in SoftBank's 39-year history.

The structural difference from Sequoia's model is leverage, not just position size. SoftBank has historically carried more than $100 billion in debt at the corporate and fund level, secured against telecom operations and equity stakes, financing that amplifies gains in favorable periods and, as documented in the Vision Fund's repeated multibillion-dollar quarterly losses, forces valuation write-downs and asset sales at exactly the moments a diversified, unlevered venture portfolio would simply absorb a bad quarter and continue. SoftBank's own stock has traded at a persistent discount to its disclosed net asset value, at times exceeding 40 to 50 percent, reflecting market skepticism about both the leverage and the concentration risk embedded in a portfolio built around conviction-sized bets rather than power law-sized ones. The Vision Fund's eventual recovery, driven substantially by Arm Holdings' 2023 IPO and a windfall T-Mobile stake received without additional investment, demonstrates the strategy can still work. It also demonstrates how much closer to the edge it operates than a fund sized the way Sequoia sizes its own outlier bets.

The DN synthesis: conviction sizing and power law sizing are not the same discipline

Read against the conviction-investing framework covered elsewhere on this site, the venture power law looks superficially similar to Druckenmiller and Soros's philosophy of sizing hard when conviction is high, and it is often described that way by venture investors themselves. It is actually closer to the opposite. Druckenmiller's sterling trade was one enormous, liquid, exitable position sized against a specific, time-bound catalyst. A venture power law bet is one of dozens of small, illiquid, multi-year positions, each individually unable to be exited if wrong, where the entire strategy depends on not needing any single one of them to work. Sequoia's FTX position behaved exactly like this: a loss that could be absorbed without altering the fund's trajectory. SoftBank's WeWork position behaved like a Druckenmiller-style concentrated bet without Druckenmiller's liquidity or exit discipline, a venture-scale illiquid holding sized and leveraged as though it were a single macro trade that had to work.

The number that actually separates these two outcomes is rarely disclosed alongside a venture fund's headline returns: how many winners, and at what size, does the fund's structure actually require to hit its target, and is the fund sized to survive the 65 percent of positions that, per the industry's own data, will not become one. That is the calculation the tool below makes explicit.

What this means for DN's readers

The same power law logic increasingly governs how sophisticated allocators approach early-stage crypto and DePIN token investing, where the base rate of project failure is at least as extreme as traditional venture's 65 percent, and the return distribution on the winners that do work is frequently more extreme still. The discipline that separates Sequoia's FTX outcome from SoftBank's WeWork outcome transfers directly: size each individual token or protocol bet against a portfolio built to survive most of them failing, rather than against conviction in any single project, and be explicit about how many genuine winners, and at what multiple, the portfolio actually needs.

For readers looking to build diversified exposure across the AI infrastructure and DePIN theme discussed throughout this publication, spot and derivatives markets are available through most major exchanges, including Bybit, OKX and MEXC. As always, this is not financial advice. A power law portfolio and a concentrated conviction bet can look identical from the outside. The math behind them, and what each can survive being wrong about, is not.

Frequently asked questions