What Happened to SoftBank's Vision Fund? Lessons from $100 Billion

The SoftBank Vision Fund raised $98.6 billion in 2017–18, the largest venture-style fund ever, and spent it on huge late-stage checks meant to make one company the winner in each category. It produced real wins, including DoorDash and Coupang, and some of the largest losses in venture history, including WeWork, Greensill and Katerra. After deep losses in 2022–23, SoftBank shrank the operation and, from 2025, concentrated its bets on OpenAI, whose marks turned the Vision Fund segment strongly positive in the year to March 2026.

The story starts with a single check. In January 2000, between the dot-com peak and the crash, Masayoshi Son met Jack Ma and committed $20 million to Alibaba after a short meeting. SoftBank's roughly 30% stake was worth about $70 billion or more at Alibaba's 2014 IPO, the most valuable venture investment on record. That fortune, and the conviction it gave Son, financed the Vision Fund.

What was the SoftBank Vision Fund?

The Vision Fund began in late 2016, when Son and Rajeev Misra, a former Deutsche Bank executive, persuaded Saudi Arabia's Public Investment Fund (PIF) to commit $45 billion. Abu Dhabi's Mubadala added $15 billion. Son's thesis was that superintelligent AI was coming and that one pool of capital large enough to back the leader in every AI-enabled category could create a federation of number-one companies.

The structure was unusual. External LP capital was split into preferred equity paying about a 7% annual coupon and ordinary equity. Roughly $40 billion of the fund was the fixed-coupon tranche, as reported. That made the fund behave partly like a leveraged vehicle: it had to pay distributions even when exits were scarce. SoftBank also borrowed at the fund level and against portfolio stakes. A venture fund's normal defense, the ability to wait, was weaker here from the start.

How did SoftBank invest the money?

The playbook had a name inside and outside the firm: capital as a weapon. Write checks large enough that a company can outspend its rivals, and make the rivals choose between taking SoftBank's money or facing a funded enemy. The fund often pre-empted rounds at high valuations and offered more money than founders asked for. At its 2019–21 peak, SoftBank Investment Advisers in London had about 500 staff, as reported, but Son's approval was decisive for large investments.

The scale was new to venture. Some of the biggest positions:

Many venture investors blame SoftBank for inflating late-stage valuations across 2017–21. When one buyer will pay double, every other buyer's price moves too.

What went wrong with WeWork?

WeWork is the case that defined the fund. In August 2017 SoftBank and the Vision Fund agreed a $4.4 billion investment at about a $20 billion valuation. WeWork signed long leases, redesigned buildings as shared workspaces and rented the space on short contracts. It had about $1 billion of annualized revenue, large losses, a technology-company story and a founder, Adam Neumann, with super-voting control.

The valuation framework did not match the business. WeWork was priced like a technology company but had the economics of leveraged real-estate arbitrage: leases of about fifteen years funded by monthly revenue. SoftBank encouraged expansion into adjacent businesses that widened losses. A planned investment of about $16 billion in January 2019 was cut to $2 billion, at a headline $47 billion, after resistance from SoftBank's own investors.

The IPO filing in August 2019 imposed the discipline the private rounds had avoided. It disclosed 2018 losses of about $1.9 billion on about $1.8 billion of revenue, about $47 billion of lease obligations, and a $5.9 million payment to Neumann for the "We" trademark, later returned. The IPO was withdrawn, Neumann stepped down, and SoftBank provided a $9.5 billion rescue at an implied value of about $8 billion. WeWork listed through a SPAC in 2021 at about $9 billion, filed for Chapter 11 in November 2023, and emerged in 2024 with SoftBank's equity wiped out. SoftBank's reported losses exceeded $11.5 billion.

Benchmark, which led WeWork's Series A in 2012 at a reported valuation of about $100 million, illustrates the contrast. Its remaining stake was largely wiped out in the bankruptcy, but it reportedly sold part of its holding in earlier secondary transactions, and the deal is described as profitable for the firm overall. The same company was a catastrophe for one investor and a reported gain for another. Price is a factor in its own right.

Why did Greensill and Katerra fail?

The two failures were linked. Greensill Capital, a London-based supply-chain finance firm, took about $1.5 billion from the Vision Fund in two 2019 rounds, at about $3.5 billion and then about $4 billion. Katerra, a Menlo Park construction-technology company, raised an $865 million Series E led by the fund in January 2018, at a reported valuation of $3 billion or more.

Katerra expanded across too many fronts at once: factories, acquired contractors and projects in several countries. Projects ran over budget and factories sat underused. During 2020 SoftBank put in about $200 million more and arranged to absorb about $440 million of Greensill-arranged debt, as reported. In 2020 insurers declined to renew about $4.6 billion of Greensill's credit insurance, and in early March 2021 Credit Suisse froze about $10 billion of Greensill-linked funds. Greensill filed for insolvency in the UK on March 8, 2021. Katerra lost its working-capital lifeline and filed for Chapter 11 in June 2021. SoftBank's roughly $2 billion in Katerra and $1.5 billion in Greensill were largely written off.

Did anything work?

Yes, and the wins matter for the lesson. DoorDash is the cleanest. SoftBank led a $535 million Series D at about $1.4 billion in March 2018, after DoorDash's fundraising had stalled. At the December 2020 IPO, DoorDash's market value at the first-day close was about $72 billion. Coupang turned about $3 billion into a stake worth $25 billion or more at its March 2021 IPO close. Arm, bought for about $32 billion in 2016 outside the fund, became worth well over $100 billion after its 2023 IPO and the AI rally, as reported.

Capital as a weapon can work when it lands on the right operator at the right moment. The same playbook failed at WeWork, Wag, Zume and Katerra. Selection mattered more than the size of the check.

What happened to the Vision Fund after 2021?

The timing turned against it. In fiscal 2020 SoftBank posted a record ¥4.99 trillion net profit, on the back of the Coupang and DoorDash IPOs. Then came net losses of about ¥1.7 trillion in fiscal 2021 and about ¥0.97 trillion in fiscal 2022. In 2022 Son announced "defense mode." SoftBank cut staff repeatedly; Misra stepped back from the second fund in 2022 and left in November 2024.

The second fund told its own story. Vision Fund 2 was announced in 2019 with about $108 billion of letters of intent from outside investors that did not convert. It ended up at about $50 billion or more, funded almost entirely by SoftBank. No major outside LPs joined, which says a lot about how institutions viewed the first fund. Fund 1's cumulative gain peaked in 2021 and fell back toward roughly break-even to modestly positive after the 2022 losses, by reported figures, while PIF's preferred tranche earned its coupon regardless.

Then came the AI pivot. In 2025 SoftBank led OpenAI's $40 billion round at a $300 billion valuation and joined the Stargate data-center joint venture. It sold its entire Nvidia stake for about $5.8 billion in October 2025. By October 2026 it had completed a further $30 billion OpenAI investment through Vision Fund 2, as reported, bringing its cumulative commitment to about $64.6 billion for roughly 13%. The Vision Fund segment recorded a reported gain of about $46 billion for the year to March 2026, nearly all from OpenAI. Supporters see the boldest bet in technology investing; critics see a repeat of 2000 and 2019.

Which factor did the Vision Fund get wrong?

The Outlier Equation describes a venture return as four multiplying factors: creating a possible outlier, pricing it, governing it, and turning it into cash. The Vision Fund's failures mostly turned on price and governance. At WeWork the creation factor was weak because the economics were real estate's, but price was the multiplier that made the loss enormous: a technology valuation on a lease book. Governance was the other zero. Super-voting control and related-party dealings were visible and did not constrain investment, and SoftBank encouraged expansion rather than restraining it.

The fund's structure also weakened the realization factor: a fixed coupon and borrowing meant SoftBank sometimes had to sell to meet its own liquidity needs, as it did with its remaining Uber shares by 2022 and with Nvidia, which it sold early in 2019 at a reported $3.3 billion gain before the stock rose many times over. Son has called the early Nvidia sale one of his regrets.

The wider lesson is in the book's history chapter: new entrants arrive at the top. Mutual funds in 1999, SoftBank in 2017–19 and crossover funds in 2021 entered late, and prices reversed when they left. Tiger Global's 2021 run shows the same pattern with a different playbook, and our list of the biggest venture capital failures places WeWork among its peers. For how late-stage pricing is supposed to work, see the growth and late-stage study and our guide to how venture capital works.

The takeaway

The Vision Fund proved that money alone cannot fix weak unit economics and that a valuation framework must match the business being bought. Its biggest losses came where a high price met weak governance, and its rescues turned sunk costs into larger losses. Its biggest wins, DoorDash and Coupang, came where large checks backed operators who were already becoming the leaders. Whether the OpenAI concentration redeems the model is still open as of November 2026. Chapter 14 of The Outlier Equation covers growth and late-stage capital, and Chapter 25 reads the WeWork decision step by step.