SoftBank is easy to recognize and difficult to describe. A shopper in Tokyo sees a mobile carrier. A chip designer sees Arm. A founder remembers the Vision Fund, whose checks made ordinary venture rounds look pocket-sized. A public-market investor sees a holding company whose assets can move by trillions of yen before the phone business has finished breakfast. All of those views are correct, but the clearest answer in 2026 is this: SoftBank Group is a machine for placing capital around the next computing platform.
That machine has changed its favorite component repeatedly. Masayoshi Son founded the company in 1981 as a distributor of personal-computer software. The name was literal - a bank of software, imagined as infrastructure for an information society. SoftBank later built around the Japanese internet, broadband and mobile communications, then reorganized its ambitions around giant technology funds. Its latest reinvention is more concentrated. Rather than scatter capital across every app that might grow, the group wants positions across the machinery that makes artificial intelligence possible.
The thesis has four layers: AI models, AI chips, AI infrastructure and physical AI. OpenAI anchors the model layer. Arm, Ampere and Graphcore sit in computing. Stargate, new data-center plans and the proposed purchase of DigitalBridge address land, power, connectivity and financing. Robotics investments, plus the planned acquisition of ABB's robotics business, bring software into warehouses and factories. It is less a portfolio than a supply chain, with SoftBank trying to occupy the handoffs.
March 31, 2026
net sales
March 31, 2026
A holding company with an operating pulse
SoftBank Group does not sell a single flagship product. It owns, controls and invests. Revenue flows from consolidated operating businesses; investment gains and losses arrive as portfolio values change or shares are sold. The parent judges its work primarily through net asset value, or NAV: the estimated value of holdings minus adjusted net debt. At the end of March 2026, SoftBank calculated ¥48.3 trillion of equity holdings, ¥8.2 trillion of adjusted net debt and ¥40.1 trillion of NAV.
That accounting lens explains behavior that looks odd from the outside. A telecom company would optimize subscriber economics. A fund manager would collect fees and carry. SoftBank can sell part of a listed asset, borrow against another, acquire an operating company and put the proceeds into a private AI developer. It is a public company offering shareholders a bundle of operating cash flow, listed stakes and private valuations, with debt amplifying both the reach and the risk.
Who actually uses SoftBank?
The answer changes by floor. Japanese consumers and businesses buy mobile service, broadband, enterprise technology, online advertising, ecommerce and payments from SoftBank Corp. and companies including LY and PayPay. Semiconductor companies license Arm architecture and intellectual property for phones, cars, connected devices and data centers. Cloud operators can use Ampere's energy-efficient server processors. Founders seek capital and introductions from the Vision Funds. Enterprises in Japan are the intended customers for Crystal intelligence, the OpenAI collaboration being developed through SB OAI Japan.
These businesses solve different problems, but SoftBank sees one common constraint: useful technology stalls without distribution and infrastructure. A good model needs computing capacity. Computing capacity needs chips, data centers, power and networks. A capable robot needs models and edge compute, then a buyer willing to redesign a workflow. SoftBank's proposed answer is coordination. Capital links independent companies; operating subsidiaries provide routes into the market; ownership of infrastructure reduces dependence at strategic bottlenecks.
AI models
OpenAI investment and the SB OAI Japan joint venture put SoftBank near research, products and enterprise adoption.
AI chips
Arm supplies pervasive compute architecture; Ampere and Graphcore add server CPU and AI processor capabilities.
Infrastructure
Stargate, neocloud operations, data-center projects and planned digital-infrastructure assets address capacity and connectivity.
Physical AI
Robotics turns inference into movement, bringing the AI thesis from screens into factories, logistics and services.
“Our primary objective at SBG is to maximize net asset value.”Masayoshi Son, Chairman and CEO
The difference is the size of the loop
Other technology investors can fund a model company. Hyperscalers can build data centers. Semiconductor groups can design chips. Telecom operators can distribute enterprise services. SoftBank's distinction is an attempt to connect all four roles while preserving the independence of each company. Management calls the approach the “Cluster of No. 1” strategy: assemble businesses with strong technologies, encourage collaboration, but avoid forcing them into a single brand or a tightly integrated bureaucracy.
The structure resembles the silver mark in SoftBank's logo - two parallel bars borrowed from the flag of Kaientai, a 19th-century Japanese trading and naval venture. The romantic interpretation is a fleet of captains sharing a direction. The practical version is governance at arm's length. Arm keeps its own public listing. Ampere kept its name after acquisition. The planned DigitalBridge deal would leave its platform separately managed. Autonomy can protect specialist cultures and customer trust, even as SoftBank supplies capital and strategic pressure.
Its expertise, then, is not simply picking startups. It is financial architecture: buying companies, creating funds, syndicating commitments, listing subsidiaries and borrowing against assets without pretending every business has the same clock. The 2016 Arm purchase showed the long horizon. SoftBank paid roughly £24 billion, invested through years of private ownership and returned Arm to Nasdaq in 2023 while keeping control. Arm is now both an operating semiconductor platform and a substantial component of group NAV.
A strategy with visible seams
The same design creates concentrated exposure. SoftBank's cumulative OpenAI investment stood at $44.6 billion by June 2026, including indirect investment and net of disposals, with another $10 billion paid in July and a third 2026 tranche planned. Completion would take the expected cumulative commitment to $64.6 billion and ownership to about 13 percent. The company is funding the expansion through a mixture of cash, asset use and bridge financing. In August, Vision Fund 2 also arranged a loan using OpenAI shares.
That makes valuation, liquidity, interest expense and foreign exchange more than footnotes. In the June quarter, investment gains reached ¥1.859 trillion, but finance costs nearly doubled from a year earlier and derivatives produced a ¥391.6 billion loss. Net income attributable to owners of the parent was ¥347.3 billion, down 17.7 percent. SoftBank can report healthy operating sales and still have its quarter shaped by Intel shares, ByteDance marks, the yen and hedges. The portfolio is the product, and the product is volatile.
The three dials to watch
OpenAI and Arm increasingly influence the value and narrative of the entire group.
Large commitments rely on debt markets, asset-backed structures and the continued liquidity of valuable holdings.
Owning adjacent layers does not automatically create customers, cheap power, compatible road maps or useful robots.
Where it sits in the market
SoftBank occupies a peculiar public-market niche. It overlaps with Prosus and Tencent as a technology holding company, with Temasek and Mubadala as a strategic allocator, with growth-equity firms as a startup investor, and with telecom, semiconductor and infrastructure operators through its subsidiaries. None is a clean comparison. Berkshire Hathaway also mixes operating businesses and investments, but SoftBank's technology concentration, private-company marks and use of asset-backed financing make the ride different.
For founders, SoftBank can offer large checks, global relationships and potential commercial partners. For enterprise customers, the group can package connectivity, cloud capacity and AI deployment, especially in Japan. For consumers, the benefits are less visible as a group strategy but familiar in practice: phones connect, payments clear, search and shopping pages load. For investors, SoftBank is a way to buy exposure to assets that are otherwise private, alongside Arm and domestic operating cash flows. The price of that access is complexity.
Software distribution
A “bank” for the personal-computer software age.
Internet, broadband, mobile
Yahoo Japan, low-cost broadband and Vodafone Japan turned distribution into networks.
Chips and giant funds
Arm and the Vision Funds shifted the center from operations toward global technology assets.
The integrated AI wager
Models, computing, infrastructure and robotics become one thesis.
The useful lesson behind the spectacle
SoftBank's scale makes it tempting to treat every move as a referendum on one man's optimism. A more useful observation is organizational. The company repeatedly separates mission from method. Its mission - “Information Revolution - Happiness for everyone” - has barely changed. The method has changed constantly: wholesale distribution, internet stakes, broadband, mobile networks, venture funds, semiconductor ownership and now AI infrastructure.
That is something smaller companies can borrow without borrowing billions. Keep the problem stable and let the vehicle expire. Measure the business according to what it is, not what it used to be. Preserve specialist autonomy when integration would smother expertise. Then name the dependencies in a strategy: if the product requires compute, power, distribution and customer trust, a model roadmap alone is not a business plan.
SoftBank still has to prove that its four layers form a system rather than a costly collage. Data centers take years. Robotics integrations are stubborn. Model economics and private valuations can change faster than financing agreements. Yet the strategic direction is legible. The company that once put software on shelves now wants to own the architecture, intelligence, buildings and machines behind a new computing era. The old “bank of software” has become a bank of dependencies.