SoftBank

SoftBank Group Corp.
Type	Public
Traded as	TYO: 9984; Nikkei 225 component
Industry	Conglomerate; Investment; Telecommunications
Founded	3 September 1981
Founder	Masayoshi Son
Headquarters	Minato, Tokyo, Japan
Key people	Masayoshi Son (Chairman and chief executive officer); Yoshimitsu Goto (Chief financial officer)
Products	Investment holdings, telecommunications, semiconductor design, robotics
Employees	Approximately 65,000
Subsidiaries	Arm Holdings, SoftBank Corp., SoftBank Vision Fund, PayPay, Ampere Computing
Website	group.softbank

SoftBank Group Corp. is a Japanese multinational holding company headquartered in Minato, Tokyo. It began as a software distributor, became a telecommunications operator, and is now principally an investment company holding stakes in technology businesses worldwide, most significantly Arm Holdings, whose processor designs are in almost every mobile phone.

The company was founded on 3 September 1981 by Masayoshi Son, who remains chairman and chief executive and whose personality and judgement have determined almost everything about it. Its logo is based on the flag of the Kaientai, a naval trading company founded in 1865 by Sakamoto Ryōma, a figure of the late Tokugawa period whom Son has cited as an influence.

SoftBank was ranked 130th in the 2025 Forbes Global 2000. Its principal assets are its holding in Arm Holdings, which it took public in 2023 while retaining the great majority; SoftBank Corp., the Japanese telecommunications operator; the Vision Funds, which invested approximately US$100 billion and more in private technology companies; and, since 2024, a very large commitment to artificial intelligence infrastructure including a US$30 billion investment in OpenAI and participation in the Stargate project.

The company is among the most volatile large businesses in the world, having recorded both the largest annual loss ever reported by a Japanese company and, in other years, extraordinary gains. Its history is a sequence of very large concentrated bets, of which several have been among the best investments ever made and several among the worst.

History

Software distribution and the early years, 1981 to 1995

Masayoshi Son was born in 1957 in Tosu, Saga Prefecture, to a Korean family who had come to Japan in his grandparents' generation and who lived in difficult circumstances. He used the Japanese surname Yasumoto in childhood, a common practice among Zainichi Koreans, and adopted Son publicly as a student, which required a legal process because Japanese law at the time did not permit a naturalised citizen to use a name not already in use.

He studied at the University of California, Berkeley, where he invented a pocket translator that he sold to Sharp for a reported ¥100 million, and imported used arcade machines to American universities.

He founded SoftBank in Tokyo on 3 September 1981 to distribute packaged software to Japanese retailers, at a moment when the personal computer was arriving in Japan and when publishers had no distribution and retailers had no idea what to stock. The name means the bank of software.

He was diagnosed with hepatitis shortly afterwards and spent much of the next three years in hospital, reading, an episode he has described as formative.

The business expanded into technology publishing, producing magazines, and into trade exhibitions, and it acquired the American publisher Ziff-Davis in 1996 and the Comdex trade show in 1995, which gave Son a position in the American technology industry and an unusual volume of information about what was being built.

Yahoo, Alibaba and the internet, 1995 to 2005

Yahoo!. Son invested US$2 million in Yahoo in 1995 and then approximately US$100 million for around a third of the company, an investment that returned many billions. He also established Yahoo! Japan as a joint venture in 1996, which became the dominant internet portal in Japan and which SoftBank controlled, a position far more durable than the American parent's.

Alibaba. In 2000 Son met Jack Ma in Beijing and, by both men's accounts, decided within minutes to invest. He put in US$20 million for approximately a third of the company. When Alibaba listed in 2014 the stake was worth approximately US$60 billion, a return of roughly 3,000 times and among the most profitable venture investments ever made. SoftBank has since sold the great majority of it, principally to fund other commitments and to meet obligations, and the disposal of an asset that had underpinned its balance sheet for two decades was completed in 2024.

The dot-com collapse in 2000 and 2001 destroyed approximately 99 per cent of SoftBank's market value, a fall that would have ended most companies and from which Son recovered.

Telecommunications, 2001 to 2013

Son's response to the collapse was to enter Japanese telecommunications, a market dominated by NTT and regarded as closed.

Yahoo! BB, launched in 2001, offered broadband over the incumbent's copper lines at a price far below what NTT charged, and SoftBank distributed modems free on street corners and outside railway stations. The campaign was chaotic and effective, Japanese broadband prices fell to among the lowest in the world, and Son had to fight the regulator and the incumbent continuously to obtain access to the lines.

Japan Telecom was acquired in 2004 for approximately US$3 billion, giving SoftBank a fixed-line business.

Vodafone Japan was acquired in 2006 for approximately ¥1.75 trillion, then the largest leveraged buyout in Asian history, and became SoftBank Mobile. The business was the weakest of the three Japanese mobile operators and Son turned it around, principally by securing the exclusive Japanese rights to the iPhone in 2008, which he obtained by approaching Steve Jobs before the device existed and which transformed the company's position.

Sprint was acquired in 2013 for approximately US$21.6 billion for 78 per cent of the American operator, the third largest carrier and a distant third. Son's intention was to combine it with T-Mobile US and to repeat in the United States what he had done in Japan. The regulators refused in 2014, and Sprint then absorbed very large sums for six years before a merger with T-Mobile was finally permitted and completed in April 2020, leaving SoftBank with a substantial stake in the combined company which it subsequently sold. The episode cost a great deal and is generally assessed as one of Son's poorer judgements.

Arm and the Vision Fund, 2016 to 2020

Arm Holdings was acquired in July 2016 for approximately £24.3 billion, a 43 per cent premium and the largest acquisition of a European technology company. Arm does not make chips; it designs processor architectures and licenses them, and its designs are in essentially every mobile phone and in a rapidly growing share of servers and vehicles.

Son's stated reasoning was that the number of connected devices would grow enormously and that Arm would collect a small royalty on almost all of them. The purchase was funded partly by selling Alibaba shares and was criticised at the time as expensive. It is now generally regarded as the best investment SoftBank has made after Alibaba.

The SoftBank Vision Fund, announced in October 2016 and closed in 2017 at approximately US$100 billion, was the largest technology investment fund ever assembled by an order of magnitude. Approximately US$45 billion came from Saudi Arabia's Public Investment Fund, US$15 billion from Abu Dhabi's Mubadala, and the remainder from SoftBank, Apple, Qualcomm, Foxconn and others. Much of the outside money was structured as preferred equity paying a seven per cent annual coupon, which meant SoftBank owed returns regardless of performance, a structure that magnified both outcomes.

The fund's strategy was to write cheques far larger than venture capital had previously written, on the argument that in winner-take-all markets the company with the most capital wins, and that being the largest shareholder in the eventual winner is worth paying for. It invested in Uber, DoorDash, Didi, Grab, Coupang, ByteDance, Oyo, Slack, Compass, Wag, Zume and approximately ninety others.

The results were mixed in a way that was masked for several years by private valuations that the fund itself was setting. Coupang, DoorDash and ByteDance were very successful. A great many others were not.

WeWork was the failure that defined the period. SoftBank invested approximately US$10 billion at valuations reaching US$47 billion in a business that leased office space long and sublet it short, which is a structurally hazardous arrangement, and whose founder Adam Neumann had extracted money from the company in ways that the listing prospectus of August 2019 disclosed. The offering was withdrawn, the valuation collapsed, SoftBank took control in a rescue that included a payment to Neumann widely criticised as rewarding failure, and WeWork eventually filed for bankruptcy in November 2023. The episode produced substantial reputational damage and a reassessment of whether very large cheques improve businesses or merely allow them to lose money faster.

The SoftBank Vision Fund reported a loss of approximately US$18 billion for the year to March 2020, and the group's annual loss of approximately ¥1.36 trillion was the largest in its history.

Son's public response in 2020, showing investors a slide depicting himself as a unicorn falling into a valley and climbing out, has become among the better-known moments in corporate presentation.

Retrenchment and return, 2020 to 2026

SoftBank sold approximately US$80 billion of assets in 2020 to repay debt and repurchase shares, including its Sprint stake, most of its remaining Alibaba holding, its stake in T-Mobile and a large part of Arm, the last through a proposed sale to NVIDIA for approximately US$40 billion that regulators blocked in February 2022.

The Vision Fund 2, funded entirely by SoftBank, invested through the 2021 period of very high private valuations and suffered accordingly.

Arm Holdings was instead listed on Nasdaq in September 2023 at US$51 a share, valuing it at approximately US$54 billion, with SoftBank retaining approximately 90 per cent. The share price rose substantially over the following two years as the market reassessed Arm's position in artificial intelligence and data centres, and the retained stake became by a wide margin SoftBank's largest asset, at times exceeding the entire market value of SoftBank itself.

From 2024 Son redirected the company decisively toward artificial intelligence, describing artificial superintelligence as the purpose of the remainder of his life.

Stargate, announced at the White House on 21 January 2025 with OpenAI, Oracle and MGX, commits an announced US$500 billion over four years to artificial intelligence data centres in the United States, with SoftBank as a principal financial participant and Son as chairman.

OpenAI. SoftBank led a round valuing OpenAI at US$300 billion in March 2025 with a commitment of up to US$40 billion, of which US$30 billion was its own. In October 2025 the board approved a second instalment of US$22.5 billion completing the US$30 billion, conditional on OpenAI completing its corporate restructuring, which it did that month. It also participated in the February 2026 round at a US$730 billion valuation.

SB OpenAI Japan, announced in February 2025, is a joint venture to develop enterprise artificial intelligence products described as Crystal Intelligence, with SoftBank committing approximately US$3 billion annually to deploy OpenAI systems across its own group companies.

Ampere Computing, which designs energy-efficient Arm-based server processors, was agreed in March 2025 for approximately US$6.5 billion, adding chip design to complement Arm.

ABB Robotics was agreed in October 2025 for an enterprise value of approximately US$5.375 billion, replacing the Swiss company's plan to spin the division off, with completion expected in mid to late 2026. The purchase is consistent with Son's stated view that artificial intelligence will be embodied in machines and that SoftBank should own the means of building them.

DigitalBridge Group, an American digital infrastructure investment firm, was agreed on 29 December 2025 in an all-cash transaction valued at approximately US$4 billion, taking it private.

SoftBank also took a 25 per cent stake in Twenty One, a bitcoin acquisition company formed through a merger with the Cantor Equity Partners special purpose vehicle announced on 23 April 2025 and led by Tether and Bitfinex.

The company sold its remaining NVIDIA holding in 2025 to fund the OpenAI commitment, a decision that was immediately profitable for OpenAI and that forwent very large subsequent gains, and which Son defended as concentrating on the asset he believed most important.

Structure and holdings

SoftBank Group is a holding company and its value is the sum of what it owns less what it owes.

Arm Holdings is the largest holding by a wide margin, at approximately 88 to 90 per cent of a company whose market value has at times exceeded US$150 billion.

SoftBank Corp., the listed Japanese telecommunications operator, provides stable cash flow, operates mobile, fixed and broadband services, owns Yahoo! Japan through LY Corporation following the merger with LINE, and holds PayPay, the dominant Japanese mobile payment service.

The Vision Funds hold public and private technology stakes, with the public holdings including Coupang, Grab, DoorDash and others, and the private holdings including ByteDance and a long tail of venture positions.

Others include Ampere, Fortress Investment Group, held until its sale in 2023, Boston Dynamics, held until its sale to Hyundai in 2021, PayPay, and the newer artificial intelligence and robotics commitments.

The persistent feature of the structure is the discount to net asset value: SoftBank's shares have consistently traded well below the sum of its holdings, frequently at half. The reasons advanced are the concentration in one person's judgement, the complexity, the use of margin loans against the holdings, the governance, and Japanese investors' general preference for operating companies over holding companies. Son has attacked the discount repeatedly through very large share repurchases.

SoftBank Corp. and the Japanese businesses

The Japanese operating businesses are the least discussed part of the group and provide the cash flow that supports everything else.

Telecommunications

SoftBank Corp., listed separately since 2018 with the parent retaining a controlling stake, is the third largest Japanese mobile operator after NTT Docomo and KDDI, with roughly a fifth to a quarter of subscribers, together with fixed broadband, enterprise services and a substantial network.

Its history is a sequence of acquisitions: Japan Telecom in 2004, Vodafone Japan in 2006, Willcom and eAccess subsequently, and the Y!mobile brand serving the value segment. The 2008 iPhone exclusive transformed its position, and the loss of exclusivity to KDDI in 2011 and to NTT Docomo in 2013 tested it.

The Japanese mobile market has been subject to sustained political pressure on prices, most consequentially from 2018 when the government of Shinzō Abe demanded reductions of around 40 per cent, and Japanese operators have since competed on price in a way they had not.

PayPay, launched in 2018 as a joint venture with Paytm and Yahoo Japan, became the dominant Japanese mobile payment service through an extremely aggressive promotion in which a proportion of every purchase was refunded, funded by SoftBank. It has more than 60 million users, has changed Japanese payment behaviour in a country that had remained unusually attached to cash, and is expected to be listed.

LY Corporation is the result of the 2021 merger of Yahoo! Japan with LINE, the messaging application that is the dominant communication service in Japan, Thailand and Taiwan. It is jointly controlled by SoftBank and Naver of South Korea, and the arrangement came under Japanese government pressure in 2024 following a data breach originating at Naver, with the ministry asking SoftBank to increase its control, which Naver resisted.

Robotics

Pepper, a humanoid service robot introduced in 2014 through SoftBank Robotics, was sold to businesses and to households in Japan and Europe. It was widely photographed, commercially unsuccessful, and production was halted in 2021. The venture is an early example of Son's conviction that robots would become general, arriving about a decade before the technology could support it.

Boston Dynamics was acquired from Google in 2017 and sold to Hyundai in 2021 for approximately US$1.1 billion, SoftBank retaining a minority. The company's quadruped and humanoid robots are the most widely recognised in the world and its commercial results were modest during SoftBank's ownership.

The ABB Robotics acquisition agreed in October 2025, at an enterprise value of approximately US$5.375 billion, returns SoftBank to the category at industrial scale, bringing one of the four largest industrial robot manufacturers.

Arm Holdings

Arm is SoftBank's most valuable asset and deserves separate treatment.

What Arm does

Arm designs processor architectures and the cores that implement them, and licenses them to companies that build chips. It does not manufacture and, until recently, did not sell chips.

Two kinds of licence exist. A core licence permits a company to use an Arm-designed processor, such as the Cortex series, in its own chip. An architecture licence permits a company to design its own processor implementing the Arm instruction set, which Apple, Qualcomm, NVIDIA and Amazon do.

Arm earns a licence fee upfront and a royalty on every chip shipped, typically a small percentage of the chip's price. The royalty is individually trivial and collectively enormous: approximately 30 billion Arm-based chips ship annually and the cumulative total exceeds 300 billion.

Why it matters

Arm's designs are in essentially every smartphone, in a growing share of servers, in most vehicles, in microcontrollers, in network equipment and in an increasing proportion of personal computers.

Its position derives from power efficiency. The instruction set was designed in the 1980s at Acorn Computers in Cambridge for a machine without a cooling fan, and the resulting simplicity produced processors that do more work per unit of energy than the x86 designs that dominated computing. That mattered little when computers were plugged into walls and mattered decisively when they ran on batteries, and it matters again now that data centres are limited by electricity rather than by space.

The company's business model is also unusual in that it profits from its licensees competing with each other. Apple, Qualcomm, Samsung, MediaTek and others compete fiercely in phone processors and all pay Arm.

Under SoftBank

SoftBank acquired Arm in 2016 and increased its research spending substantially, expanding into servers, automotive and infrastructure, which required investment that a public company optimising quarterly results might not have made and which has since paid.

The attempted sale to NVIDIA for approximately US$40 billion, agreed in September 2020, was abandoned in February 2022 after the United Kingdom's Competition and Markets Authority, the European Commission and the American Federal Trade Commission all objected, on the ground that Arm's licensees compete with NVIDIA and would have been licensing from a competitor.

Arm listed on Nasdaq in September 2023 at US$51 a share. The listing in New York rather than London was a political disappointment in the United Kingdom, where Arm is regarded as the country's most important technology company.

Arm has begun designing complete chips rather than only licensing cores, which places it in competition with some of its own licensees and is the principal strategic question about its future.

Its litigation with Qualcomm over whether Nuvia's architecture licence transferred on acquisition was decided largely in Qualcomm's favour by a jury in December 2024.

The Vision Fund portfolio

The Vision Funds together invested approximately US$140 billion across more than 470 companies, and the outcomes divide sharply.

Substantial successes include Coupang, the Korean e-commerce company, which listed in 2021 and returned many times the investment; DoorDash; ByteDance, in which the Vision Fund holds a position whose value has grown enormously; Grab in Southeast Asia; Auto1; Guardant Health; and Arm itself, held directly by the group.

Substantial failures include WeWork, which went bankrupt; Greensill Capital, a supply chain finance company that collapsed in 2021 and produced a political scandal in the United Kingdom; Katerra, a construction technology company that went bankrupt in 2021 after approximately US$2 billion of investment; Zume, which made pizza in robotic trucks and closed; Wag, a dog-walking application; Brandless; and Oyo, the Indian hotel company, which was written down very substantially.

Ambiguous outcomes include Uber, in which SoftBank invested approximately US$7.7 billion, sold at a loss during 2022 and which subsequently recovered strongly; Didi, damaged severely by Chinese regulatory action in 2021; and a long tail of private companies carried at marks that the public market has not tested.

The fund's structural problem has been that very large cheques at high valuations require very large outcomes to produce returns, and that the number of companies capable of such outcomes is smaller than the number the fund invested in. The preferred coupon owed to the Saudi and Emirati investors meant that mediocre gross returns produced poor net ones for SoftBank.

The fund also changed the venture industry while it operated. Its willingness to pay any price raised valuations across the market, obliged conventional venture firms to raise larger funds, and contributed to the conditions of 2021 whose correction has occupied the industry since. Several of its investments were made in companies that had not asked for the money and in some cases were told that if they did not accept it a competitor would.

Investment has slowed substantially since 2022, and the group's capital has been redirected to artificial intelligence.

Masayoshi Son

SoftBank is more completely an expression of one person than almost any company of its size, and an account of it that does not address Son is incomplete.

He was born in 1957 into a Zainichi Korean family in a shack beside a railway line in Kyushu, in circumstances he has described publicly and without embellishment. His father raised pigs and later ran pachinko parlours. Discrimination against Koreans in Japan at the time was substantial and open.

He left for the United States at 16, finished high school in California in three weeks by persuading the principal to let him take the examinations, and studied economics at Berkeley. He has said that he decided at university to make his fortune by 30, to build a business by 40, to accumulate by 50, to pass it on by 60 and to retire by 70, and has publicly revised the last several times.

His method has been consistent: identify a technological shift, form a conviction about it that is more extreme than the consensus, and commit very large sums quickly. The approach produced Yahoo, Alibaba, the iPhone exclusive and Arm. It also produced WeWork, Sprint, the Vision Fund's long tail and the sale of NVIDIA.

He has described his framework as a 300-year vision, and presents at shareholder meetings using slides with a distinctive combination of exponential curves, drawings and personal declaration that has no equivalent in Japanese or Western corporate practice.

His personal risk-taking has repeatedly extended to his own balance sheet. He has pledged a substantial proportion of his SoftBank shares against personal borrowing, and he holds a large personal interest in the Vision Fund's performance through an arrangement that magnifies both directions.

Assessments differ sharply. He has made two of the best investments in history and several of the worst, and the honest summary is that his returns are dominated by a very small number of outcomes, which is characteristic of venture investing and unusual in a company of SoftBank's scale, because most companies of that scale cannot survive the losses.

Governance and finances

SoftBank's finances are unusually difficult to assess because the company is a holding company whose principal assets are marked to market, so that reported profit and loss reflect valuation movements rather than operations.

It reported a loss of approximately ¥1.7 trillion for the year to March 2023 and very large profits in other years, and the swings are a function of the Vision Fund's marks and of Arm's share price.

Debt has been a persistent concern. SoftBank has borrowed substantially at the holding company level and has used margin loans and prepaid forward contracts against its holdings, notably Alibaba and Arm, which provide cash without selling and which create exposure if the underlying falls. Rating agencies have held the company below investment grade for extended periods and Son has disputed their methodology, arguing that loan-to-value against a portfolio of liquid assets is the correct measure.

Governance has been criticised persistently. Independent directors have resigned, including Jack Ma's departure in 2020 and several others; the board has been described as unable to constrain the chief executive; the WeWork investment was made in a manner that the company's own subsequent statements acknowledged lacked adequate process; and the SB Northstar trading unit, through which Son took very large options positions in American technology shares in 2020 and which was described in the press as the Nasdaq whale, was conducted with a personal interest and was wound down after losses.

SoftBank has responded with enhanced disclosure, a reduced number of new Vision Fund investments and a focus on loan-to-value, and Son retains effective control through his holding and his position.

Stargate and the artificial intelligence commitment

SoftBank's commitment to artificial intelligence since 2024 is the largest single strategic redirection in its history and is worth setting out in detail, because it now accounts for most of what the company is doing.

The thesis

Son's stated position is that artificial general intelligence will arrive within a small number of years and artificial superintelligence within roughly a decade, that this will be the most consequential event in the history of technology, and that SoftBank should hold positions across the whole stack required to produce it.

He has described the remainder of his working life as devoted to this, and has said that his earlier work, including Alibaba and Arm, was preparation.

The positions

Models. The US$30 billion committed to OpenAI makes SoftBank among its largest investors. The February 2025 SB OpenAI Japan joint venture will deploy OpenAI systems across SoftBank's own companies at a cost of approximately US$3 billion a year and sell enterprise products in Japan under the name Crystal Intelligence.

Processor architecture. Arm Holdings, approximately 88 to 90 per cent owned, licenses the instruction set on which an increasing share of data centre processors is built, including Amazon's Graviton, NVIDIA's Grace and Microsoft's Cobalt.

Server chips. Ampere Computing, agreed in March 2025 for approximately US$6.5 billion, designs Arm-based server processors and gives SoftBank the ability to build chips rather than only to license the designs.

Data centres. Stargate, announced on 21 January 2025 with OpenAI, Oracle and MGX and endorsed at the White House, commits an announced US$500 billion over four years to American artificial intelligence infrastructure, with Son as chairman. DigitalBridge, agreed on 29 December 2025 for approximately US$4 billion, brings digital infrastructure investment expertise and assets.

Embodiment. ABB Robotics, agreed in October 2025 at an enterprise value of approximately US$5.375 billion, brings one of the world's largest industrial robotics businesses, consistent with Son's argument that the technology will be applied physically and that the shortage of labour in ageing economies including Japan's makes robots a necessity rather than a preference.

The funding

The commitments are very large relative to the company's resources and have been funded by disposal and by borrowing.

SoftBank sold its entire NVIDIA holding in 2025, realising several billion dollars and forgoing the subsequent appreciation, a decision Son defended on the ground that OpenAI was the better investment and that holding both was not possible.

It has also used margin arrangements against Arm, raised debt at the holding company, and disposed of remaining Vision Fund positions.

The resulting structure concentrates the company's value in a small number of private and illiquid positions funded partly by debt secured against a single listed holding whose price is itself driven by expectations about artificial intelligence. If those expectations are met the arrangement produces extraordinary returns; if they are not, the correlation between the assets and the collateral is the problem.

The criticism

Three criticisms are made and each is serious.

The first is circularity. SoftBank invests in OpenAI; OpenAI buys computing from data centres SoftBank helps finance; those data centres use chips built on Arm designs SoftBank owns. The arrangement is coherent as a strategy and produces a structure in which the same expectation supports every part, so that a disappointment propagates through all of it simultaneously.

The second is concentration. SoftBank has spent four decades diversifying across hundreds of investments and has now concentrated in a single theme at a scale that exceeds its equity.

The third is timing. Son's record includes entering markets too early, with Pepper and with the network computer era investments, and entering them at the top, with the Vision Fund in 2021. The artificial intelligence commitment was made after the sector's valuations had risen very substantially, which is the pattern of the second failure rather than the first.

Son's answer, given consistently, is that the opportunity is larger than any previous one and that a bet proportionate to conventional risk management would be too small to matter.

Corporate history in detail

The near-death of 2001

SoftBank's market value fell from approximately ¥20 trillion at the peak of the internet boom in February 2000 to a small fraction of it by 2002, a decline of roughly 99 per cent, which is among the largest ever suffered by a company that survived.

Son has said that he briefly became, on paper, the wealthiest person in the world for a few days in early 2000 and then lost more money than anyone in history. The company's survival depended on the Alibaba and Yahoo Japan holdings, which were not marked in the public market at the time, and on his willingness to borrow against them.

The response, entering telecommunications, was a deliberate move from an asset that could be repriced to a business producing cash, and it worked.

The Sprint episode

The acquisition of Sprint in 2013 was intended as the first step in a combination with T-Mobile US that would create a third American carrier of viable scale.

The Department of Justice and the Federal Communications Commission indicated in 2014 that they would not permit it, on the ground that four carriers were better for consumers than three. SoftBank was left owning a business with a weak network, heavy debt and a subscriber base that was leaving, in a market where the two leaders could outspend it indefinitely.

Sprint consumed capital and management attention for six years. Marcelo Claure was installed as chief executive and reduced costs substantially without changing the competitive position. The merger with T-Mobile was eventually agreed in 2018 and approved in 2020 after litigation by state attorneys general, with SoftBank receiving shares in the combined company which it subsequently sold.

The strategic reading is that Son bought an asset whose value depended on a regulatory decision he could not control, at a price that assumed the decision would go his way.

The Nasdaq whale

In 2020 SoftBank established a trading unit, SB Northstar, which bought approximately US$4 billion of American technology shares and, more consequentially, very large call option positions on them, a strategy that amplified the market's rise and that was identified in the financial press in September 2020 as having contributed to unusual movements in technology share prices.

The positions were profitable and then not, the unit was wound down in 2022, and the episode drew criticism because Son held a personal interest in it and because a holding company borrowing against its assets to buy derivatives on shares it did not own was outside any stated strategy.

Succession

Nikesh Arora joined from Google in 2014, was designated as Son's successor, and left in June 2016 when Son announced that he intended to remain for at least another decade. Arora subsequently became chief executive of Palo Alto Networks.

Marcelo Claure, Rajeev Misra and others have held senior positions and departed. Misra, who built the Vision Fund and who came from Deutsche Bank and UBS, left in 2022 to run his own fund while retaining a Vision Fund role.

No successor has been designated since Arora, and the question of what SoftBank is without Son is the most consequential unanswered one about the company.

Investment philosophy

Son's approach is distinctive enough to be worth describing, because it explains both the successes and the failures.

Time machine management is his term for the practice of observing a business model working in one country and replicating it in another where conditions are less developed. He applied it repeatedly, bringing American internet models to Japan through Yahoo! Japan, and subsequently funding local versions of established models across Asia and Latin America, on the reasoning that the destination is known and the question is who arrives first.

The cluster of number ones is his description of the group's intended structure: a collection of companies each dominant in its own field, connected by shared ownership and by Son's coordination rather than by operational integration. The concept is closer to a Japanese keiretsu than to a Western conglomerate and it has rarely produced the synergies he has described.

Conviction at scale. Son's characteristic action is to decide quickly, on the basis of a judgement about a founder and a market, and to commit far more than anyone else will. He has described deciding on Alibaba within minutes and on Arm within a fortnight. The approach is what produced the outcomes that define his record, and it is also what produced WeWork, where the same speed applied to a business with a structural flaw.

The 300-year vision. Son has said since 2010 that SoftBank should be organised to last three centuries and that the point of the structure is to survive the departure of any individual. Critics observe that the structure is entirely dependent on one individual and that no successor has been established; the one designated successor, Nikesh Arora, left in 2016 after two years.

Asymmetric personal exposure. Son has repeatedly used leverage against his own holdings and has a large personal interest in the Vision Fund's performance. This aligns him with shareholders in the upside and creates a degree of concentration that ordinary governance would not permit.

Financial structure

SoftBank's accounts are among the more difficult of any large company to interpret, and the reasons are structural rather than presentational.

Why the numbers move

The group applies investment entity accounting to the Vision Funds, which means the holdings are carried at fair value and every change passes through the income statement. A quarter in which technology valuations rise produces an enormous profit and one in which they fall produces an enormous loss, neither of which corresponds to anything the company did.

The effect is magnified by the private holdings, whose valuations are estimated using comparable public companies and discounted cash flows rather than observed in a market. When public technology valuations fell in 2022, the marks on hundreds of private companies fell with them, some by very large proportions, and the company reported losses of a scale that had no operating cause.

Investors have therefore generally ignored the reported profit and looked instead at net asset value, the sum of the holdings less net debt, which the company publishes, and at loan-to-value, the ratio of net debt to the value of the holdings, which management targets below 25 per cent and has allowed to rise above it during periods of commitment.

The discount

SoftBank's shares have traded at a persistent discount to net asset value, frequently between 40 and 60 per cent, which means the market values the company at substantially less than the sum of what it owns.

Explanations advanced include the concentration of decision-making in one person and the absence of a successor; the illiquidity and opacity of the private holdings; the leverage at the holding company; the tax that would be payable if the holdings were sold; the governance record; the complexity; and a general Japanese market discount applied to holding companies.

The company's response has been share repurchases on a very large scale, including programmes of ¥2.5 trillion and ¥1 trillion, which mechanically increase net asset value per share and which have narrowed the discount temporarily on each occasion without closing it.

Instruments

SoftBank has used financial instruments more inventively than most industrial companies, including margin loans secured against Alibaba and Arm shares; prepaid forward contracts, which provide cash immediately against shares to be delivered later at a price within a collar, and which allowed the company to monetise Alibaba without recognising a disposal; asset-backed financing against the Vision Fund portfolio; and extensive hybrid bonds sold to Japanese retail investors, which count partly as equity for rating purposes and which have made SoftBank one of the largest issuers of retail debt in Japan.

The instruments are lawful, disclosed and complex, and they are among the reasons the company's balance sheet is difficult to assess from outside.

SoftBank and Japan

SoftBank occupies an unusual position in Japanese business that is worth describing because it explains some of the company's conduct.

Japanese corporate culture has historically favoured consensus, incremental improvement, lifetime employment and deference to established hierarchies including the ministries. SoftBank has operated against each of these. It hired outsiders into senior positions, paid them at international rates, challenged NTT publicly and through the regulator, and pursued acquisitions on a scale and at a speed that Japanese companies generally did not.

Son's background as a Zainichi Korean is part of this. The Korean minority in Japan, descended largely from people who arrived during the colonial period, faced legal and social discrimination for decades, and the businesses they built were concentrated in sectors the mainstream would not enter. Son has spoken about this directly, including about the decision to take his Korean surname publicly, and it is generally understood that a person outside the establishment had less to lose by attacking it.

The company also has an unusual public profile. It owns the Fukuoka SoftBank Hawks, among the most successful teams in Japanese baseball, and its sponsorship and advertising, including a long-running campaign featuring a white dog as the father of a Japanese family, are widely recognised.

Its treatment by the Japanese establishment has changed with its fortunes: regarded as an upstart in the 2000s, as a national champion after Alibaba, as an embarrassment after WeWork, and as a strategic asset again since the artificial intelligence commitment, with the government's endorsement of Stargate and of the OpenAI joint venture reflecting an interest in Japan holding a position in a technology it has otherwise been slow to develop.

Subsidiaries and holdings in brief

Wholly or majority owned: Arm Holdings (approximately 88 to 90 per cent), SoftBank Corp. (controlling stake), SoftBank Vision Fund 1 and 2, Ampere Computing, SB Energy, SoftBank Robotics, Z Holdings and LY Corporation through SoftBank Corp., PayPay, Boston Dynamics until 2021, Fortress Investment Group until 2023, and, pending completion, ABB Robotics and DigitalBridge.

Significant minority positions: OpenAI, ByteDance, Coupang, Grab, DoorDash, Symbotic, Twenty One, and a long tail of Vision Fund holdings.

Disposed: Alibaba, NVIDIA, Sprint and the resulting T-Mobile stake, Uber, Didi in part, Yahoo!, Ziff-Davis, Comdex, Boston Dynamics and a large number of Vision Fund positions.

Comparisons

SoftBank does not resemble other large companies and the comparisons made are each partial.

It is sometimes compared to Berkshire Hathaway, another holding company built on one person's capital allocation. The differences are more instructive than the similarities: Berkshire buys profitable businesses at conservative prices and holds them without leverage at the parent level; SoftBank buys unprofitable ones at high prices with borrowed money.

It is sometimes compared to a venture capital firm, and the Vision Fund is one. But SoftBank is also an operating company, a telecommunications carrier and the owner of the world's most widely licensed processor architecture, and its permanent capital means it need not return money on a fund's schedule.

It is sometimes compared to the zaibatsu and their post-war successors, the Japanese industrial groups, and Son's description of a cluster of number ones is consistent with that tradition. The group lacks their bank at the centre and their cross-shareholdings.

The closest modern analogue is probably Prosus and Naspers, the South African and Dutch group whose value rests overwhelmingly on a single early investment in Tencent and which trades at a comparable discount to its assets for comparable reasons.

Timeline

1981 Masayoshi Son founds SoftBank on 3 September to distribute software.
1994 The company lists in Japan.
1995 Invests in Yahoo! and acquires Comdex.
1996 Establishes Yahoo! Japan; acquires Ziff-Davis.
2000 Invests US$20 million in Alibaba.
2001 The dot-com collapse removes approximately 99 per cent of the market value; Yahoo! BB launches.
2004 Acquires Japan Telecom.
2006 Acquires Vodafone Japan for approximately ¥1.75 trillion.
2008 Secures the exclusive Japanese iPhone rights.
2013 Acquires 78 per cent of Sprint for approximately US$21.6 billion.
2014 Alibaba lists; SoftBank's stake is worth approximately US$60 billion; Pepper is introduced.
2016 Acquires Arm Holdings for approximately £24.3 billion; announces the Vision Fund in October; Nikesh Arora leaves.
2017 The Vision Fund closes at approximately US$100 billion; Boston Dynamics is acquired.
2018 SoftBank Corp. lists; PayPay launches.
2019 The WeWork listing is withdrawn in September and SoftBank rescues the company.
2020 Reports the largest annual loss in Japanese corporate history; sells approximately US$80 billion of assets; the Sprint and T-Mobile merger completes in April; agrees to sell Arm to NVIDIA.
2021 Yahoo! Japan and LINE merge; Boston Dynamics is sold to Hyundai; Greensill and Katerra collapse.
2022 The NVIDIA sale of Arm is abandoned in February.
2023 Arm Holdings lists on Nasdaq in September; WeWork files for bankruptcy in November.
2024 The remaining Alibaba holding is disposed of.
2025 Stargate is announced on 21 January; the OpenAI investment is agreed in March and completed in October at US$30 billion; Ampere is agreed in March for approximately US$6.5 billion; SB OpenAI Japan is announced in February; ABB Robotics is agreed in October for approximately US$5.375 billion; the NVIDIA holding is sold; DigitalBridge is agreed on 29 December for approximately US$4 billion.
2026 Participates in the OpenAI round at a US$730 billion valuation; the ABB Robotics transaction is expected to complete.

Reception

SoftBank is regarded very differently depending on the period under discussion and the observer.

In Japan Son is an unusual figure: a naturalised citizen of Korean descent who became among the country's wealthiest people, who challenged established institutions including NTT and the ministries, and who has spoken publicly about discrimination. He is admired for having broken the Japanese corporate pattern and criticised for the volatility of the results.

Among venture investors the Vision Fund's effect is generally regarded as having been destabilising: it inflated valuations, encouraged founders to raise more than they needed, and demonstrated that capital is not a substitute for a business model. Several of the practices it introduced, very large late-stage rounds and competitive pressure to accept them, have receded since 2022.

Among technology executives Son's record on the specific bets is respected, particularly Alibaba and Arm, and his approach is regarded as unrepeatable rather than as a model.

Among credit analysts the company has been a persistent concern, and its loan-to-value discipline, its asset sales in 2020 and its subsequent recommitment of capital to illiquid artificial intelligence positions have been debated continuously.

Among shareholders the persistent discount to net asset value is the standing complaint, and the very large buybacks have been the principal response.

Assessment

SoftBank is best understood not as a company but as a mechanism for converting borrowed money into concentrated bets on technological change, operated by one person over four decades.

The record of the mechanism is extraordinary in both directions. A US$20 million investment in Alibaba returned approximately US$60 billion. A US$100 million investment in Yahoo returned many billions. The Arm acquisition, criticised as expensive, is now worth several times what was paid. Against these, Sprint consumed six years and very large sums for little, WeWork lost approximately US$10 billion, the first Vision Fund's returns after its preferred coupon have been mediocre, and the sale of NVIDIA to fund OpenAI forwent one of the largest gains available in the market.

What distinguishes Son from an ordinary investor with a mixed record is scale and persistence. He was nearly destroyed in 2001 and returned. He was nearly destroyed in 2020 and returned. The structure survives its losses because the winners are large enough to absorb them, which works until a loss arrives that is not absorbed.

The present commitment is the largest he has made. SoftBank has pledged US$30 billion to OpenAI, is a principal participant in a US$500 billion infrastructure programme, has bought a chip designer and a robotics business, and has sold assets including NVIDIA to fund it. If artificial intelligence proceeds as Son believes, SoftBank will hold positions in the model developer, the processor architecture, the server chips, the data centres and the robots, which is a coherent and deliberate structure.

If it does not, the company will hold very large illiquid commitments funded partly by debt, and the mechanism will be tested more severely than it has been before. Son is 69, and he has described this as the work of the remainder of his life, which is both the conviction that makes the bet possible and the reason nobody at the company is placed to question it.