In early 2018, the founders of Chinese artificial intelligence startup SenseTime Group flew to Tokyo to see billionaire investor Masayoshi Son. As they entered the offices, Chief Executive Officer Xu Li was hoping to persuade the head of SoftBank Group to invest $200 million in his three-year-old startup.
A third of the way into the presentation, Mr. Son interrupted to say he wanted to put in $1 billion. A few minutes later, Mr. Son suggested $2 billion. Turning to the roomful of SoftBank managers, Mr. Son said this was the kind of AI company he'd been looking for. "Why are you only telling me about them now?" he asked, according to one person in the room.
In the end, SoftBank invested $1.2 billion, helping to transform SenseTime into the world's most valuable AI startup. The young company's valuation hit $7.5 billion this year.
That investment model is now under fire after Mr. Son, 62, boosted the equity in office-sharing startup WeWork only to see it plummet as investors balked at enormous losses and troublesome governance. Indeed, SoftBank has participated, along with other investors, in scores of fundraisings that have added a total of more than $150 billion to the value of private companies, according to Bloomberg calculations. Among its deals are the world's top two startups — ByteDance Inc. valued at $75 billion and Didi Chuxing Inc. at about $56 billion. In some cases, SoftBank's involvement in multiple funding rounds helped drive up valuations that resulted in paper profits for Mr. Son's company.
The WeWork fiasco raises questions about such numbers. The co-working startup's valuation crested at $47 billion this year with SoftBank's investment, then plummeted to $7.8 billion in a bailout engineered by Mr. Son. WeWork is slashing jobs and scaling back operations.
"WeWork is not just a mistake, it is a signal of weakness in the whole model," said Aswath Damodaran, a professor of finance at New York University's Stern School of Business, who has written four books on valuing businesses. "If you screwed up that valuation so badly, what about all of the other companies in your portfolio?"
SoftBank said WeWork is an exception rather than a symptom of broader problems, and it has learned from the experience.
Since unveiling his $100 billion Vision Fund in 2016, Mr. Son has become the most active tech investor on the planet, pouring money into more than 80 companies. That helped create a bumper crop of unicorns, more than 300 startups priced at $1 billion or greater, according to the research firm CB Insights.
What's not as well understood is the incentive Mr. Son has to keep valuations rising. When SoftBank buys shares in a startup and then invests again at a higher valuation, Mr. Son said he has made a profit. That is legal under accounting standards, but SoftBank receives no money. The only change is that SoftBank has boosted the value of its original stake from, say, $1 billion to $2 billion by raising the value of the startup. In SoftBank's income statements and return calculations, at least some of the additional $1 billion can be counted as profit.
"They pump up valuations to get higher returns to look good to investors," said Eric Schiffer, CEO of Patriarch Organization, a Los Angeles-based private equity fund. "That kind of fundraising apparatus is essentially unicorn porn."
SoftBank said its accounting complies with all standards and is consistent with widely accepted practices. As for startup valuations, it said it is not determining them on its own and invests with experienced firms such as Sequoia Capital and Temasek Holdings. "Our valuations have been validated by more than 120 sophisticated investors who've invested alongside and after us," Navneet Govil, chief financial officer of SB Investment Advisers, the entity that manages the Vision Fund, said in a statement.
SoftBank said it has a rigorous internal process for setting valuations, and it books profit on any increase in valuation only after taking into account future cash flows and public market proxies, as well as private market funding prices. SoftBank's auditors at Deloitte & Touche check those calculations, and the Vision Fund's limited partners have their own auditors, including staff from Duff & Phelps and Ernst & Young, who vet the final figures. "Our valuation process is robust and reviewed quarterly by independent auditors," Mr. Govil said. "We believe our performance is strong. In just 2 1/2 years, Vision Fund 1 has already had seven IPOs, $4.7 billion of realized gains, $11.4 billion in cumulative investment gains and returned $9.9 billion to our limited partners."
Today's accounting rules may be ill-suited to an era of unprecedented speculation on unicorns. Under the International Financial Reporting Standards that SoftBank uses, companies have wide latitude to determine how much they think portfolio companies are worth — and therefore how much profit they report to investors. It's unclear whether any company has tried to determine paper profits for tech startups on the scale SoftBank is now using. "I don't believe we've ever seen an attempt to record this magnitude of income with respect to unquoted equity investments," said Robert Willens, a tax expert in New York.
Mr. Son's bookkeeping has allowed him to claim his average internal rate of return far outpaces those of other investors. This month, as SoftBank took a hit from WeWork, Mr. Son defended his investment approach. "There are 5,000 venture capitals globally and average IRR is 13%," he said. "Our return is about twice as big as this."
Mr. Son's confidence in his own acumen led to the creation of the Vision Fund in 2017, which at the time was more than 10 times the size of any venture capital fund. He was seeking to repeat the success of his most celebrated investment — a $20 million bet on China's Alibaba Group that turned into stock now worth more than $120 billion.
With Abu Dhabi's Mubadala and Saudi Arabia's Crown Prince Mohammed bin Salman backing the Vision Fund, Mr. Son began a blitzkrieg of deals in 2017. He invested more than $35 billion across about 100 companies, according to research firm Preqin. Among the biggest were multibillion-dollar fundings of WeWork and Didi Chuxing, the Chinese ride-hailing giant modeled after Uber Technologies Inc. In December, a SoftBank-led group invested $9 billion in Uber, including buying stock from existing shareholders.
SoftBank began including financial results for the Vision Fund during the fiscal year that ended in March 2018. Total operating profit including a related Delta Fund was ¥303 billion, or less than $3 billion. That surged to ¥1.26 trillion the following year, making it the most profitable unit at Mr. Son's company and accounting for more than half the parent company's operating income. With Mr. Son's energy directed at startups, SoftBank spun off the domestic telecom business that had made it famous and generated cash for his early investments.
"My brain and my heart, almost everything about myself is focusing on Vision Fund," Mr. Son told investors in May.
But the profits SoftBank booked were mostly on paper. In the first fiscal year, unrealized gains on investment valuations accounted for essentially all the stated income for the Vision and Delta funds. In the most recent fiscal year, unrealized gains on valuations amounted to ¥1 trillion, while realized gains — like the sale of India e-commerce giant Flipkart to Walmart Inc. — totaled less than ¥300 billion.
WeWork underscores the risks of that approach. SoftBank first took a stake in August 2017 at a valuation of $21 billion. It then invested another $3 billion in November 2018 at a $45 billion valuation and later agreed to a $1.5 billion warrant at $47 billion. As Mr. Son reported results this May, he highlighted WeWork as an example of portfolio companies heading for IPOs. When the deal fell apart, SoftBank took a ¥498 billion hit.
SoftBank Vision Fund said it never took profits from WeWork by marking it all the way up to $47 billion. It kept the shares on its books at about half that price. It still had to take that down by about 75%, which led to the loss.
Venture capital and private equity firms are mostly private so they don't need to report quarterly profits to public shareholders, and their limited partners are typically focused on returns when portfolio companies cash out through IPOs or acquisitions. SoftBank doesn't reveal specific valuation changes for each of its portfolio companies in a quarter, typically only naming a few winners or losers. Investors putting money in alongside SoftBank are at times affiliates, like Grab, Didi and Alibaba.
"If I'm an investor, I want to know how they are coming up with these numbers. Otherwise, you can't believe any of the valuations," said NYU's Mr. Damodaran. "The more they talk about accountants the less I would trust the numbers."
SoftBank concluded that under IFRS rules the Vision Fund must count valuation changes as income because its primary business is investing, and SoftBank Group must incorporate that income in its books because the Vision Fund is a consolidated subsidiary. One person close to the company said the resulting profit figures are almost meaningless, but there is no better accounting method given the current rules.
"I'm not an accountant or a lawyer, but presenting this as income doesn't make any sense," said Ilya Strebulaev, a professor of finance at Stanford University's Graduate School of Business whose research suggests the latest, post-money valuations typically overvalue startups by about 50%.
Adding to the complexity is that startup stakes are sometimes transferred between SoftBank Group and the Vision Fund, which have different shareholders. The price at which those assets are shifted has implications for profits on either side.
After WeWork, other deals are coming under scrutiny. SoftBank invested in Didi Chuxing in 2015 with the Chinese company's valuation at about $6 billion. It then put more money in about once a year as Didi's valuation climbed to $56 billion.
But Didi has run into trouble since the fundraising two years ago. Chinese regulators have cracked down on ride-hailing services for drawing migrant workers into big cities and hurting taxi drivers' incomes. Two passengers were killed after using its car-pooling service, prompting a government suspension. In addition, investors have grown skeptical about ride-hailing after the roughly 35% slump in Uber's shares since its May IPO.
Uber's drop means SoftBank should probably mark down Didi by at least the same margin, said one person who has worked on deals with SoftBank. The Japanese company would also have to look at its investments in other ride-sharing companies such as Grab Holdings Inc. in Southeast Asia, the person said. Didi declined to comment. A Grab spokeswoman said there has been no change in its valuation and it has diversified beyond ride-hailing.
SoftBank took a loss in its most recent earnings report on its Uber stake, but made no mention of the other ride-hailing firms in its portfolio. It did cut the estimated value of its stakes in Didi and other ride-hailing services in the most recent quarter, according to one person close to the company. SoftBank said it can't disclose the loss or gain on every portfolio company each quarter.