SoftBank: AI is creating value faster than it produces cash

SoftBank is no longer merely a technology investor selecting companies. It wants to finance the full artificial intelligence chain: models, chips, data centres, power and the robots that will use that computing capacity. Its first quarter of fiscal 2026 shows why this ambition can create enormous value. It also shows why it is already absorbing enormous amounts of money.

The paradox is captured by two pictures of the same quarter. At 30 June, SoftBank reports record net asset value of ¥72.3 trillion. Net asset value, or NAV, simply means the estimated value of its holdings minus adjusted net debt. Yet net income declines, holding-company cash falls by ¥1.2 trillion and adjusted net debt rises by ¥2.61 trillion.

This is not an accounting contradiction. Portfolio value increased faster than the cash available to finance it. For SoftBank Group shareholders, that distinction unlocks the entire quarter.

First, the setting: what does SoftBank actually own?

SoftBank Group is Masayoshi Son's Japanese investment holding company. It should not be confused with SoftBank Corp., the Japanese telecommunications operator in which the holding company remains a shareholder. The group primarily owns Arm, the British chip-architecture designer, and technology stakes held through SoftBank Vision Fund 1, Vision Fund 2 and specialist vehicles.

Since 2025, its strategy has shifted towards what management calls artificial superintelligence. In practice, OpenAI provides exposure to models, Arm to computing architecture, Ampere to processors, US projects to power and data centres, and ABB Robotics to AI in the physical world.

That chain explains why a SoftBank quarter cannot be read like an ordinary industrial company. Revenue and earnings matter, but changes in holdings values can dominate profit, while the corresponding cash requirements appear elsewhere in the accounts.

1. The quarter produces ¥1.86 trillion of investment gains

SoftBank reports ¥2.02 trillion of revenue, up 10.9%. Income before tax is ¥589.2 billion, down 14.6%, and net income attributable to shareholders is ¥347.3 billion, down 17.7%.

The most striking line is investment gains: ¥1.86 trillion, including ¥1.33 trillion from Intel and ¥358.4 billion from ByteDance. Those gains do not mean SoftBank received the same amount in its bank account. Part reflects increases in the estimated or quoted value of investments it still owns.

The difference matters. A realised gain follows a sale or another event that converts value into available money. An unrealised gain increases earnings and NAV but remains exposed to the next market price or valuation.

2. OpenAI is worth twice its cost in the accounts, without producing $45 billion of cash

At 30 June, SoftBank reports a cumulative OpenAI cost of $44.6 billion and a fair value of $89.6 billion. The $45 billion difference is therefore a cumulative unrealised gain.

The number is enormous, but its meaning is precise. It is neither a dividend received, sale proceeds nor immediately available liquidity. Fair value is the accounting estimate of what the stake is currently worth using relevant transactions and assumptions. It can support group wealth and, depending on a financing structure, borrowing capacity. It pays a bill only after monetisation or when accepted as support for credit.

This distinction extends our analysis of the OpenAI bridge repayment. Temporary financing allowed SoftBank to invest quickly. Repaying the bridge removes a short-dated maturity, but it does not automatically turn the OpenAI gain into cash.

3. Arm represents six yen out of every ten in adjusted holdings

Arm is worth ¥49.91 trillion in SoftBank's published sum of the parts. Total adjusted holdings are ¥83.11 trillion. Opulion calculation: 49.91 divided by 83.11 equals 60.1%.

Despite hundreds of fund positions and a strategy spanning the full AI chain, the holding company's wealth remains dominated by one listed company. Concentration works both ways. A rise in Arm can quickly improve NAV and mechanically reduce leverage. A fall can remove a material part of the cushion supporting debt.

SoftBank itself illustrates this sensitivity. Between 30 June and its 5 August pro forma estimate, NAV falls from ¥72.3 trillion to ¥58.3 trillion. Opulion calculates a 19.4% decline. The dates differ, and this is not an accounting loss for the quarter. It only shows how quickly net wealth can move when the principal listed assets change price.

4. While NAV rises, the balance sheet supplies the money

Adjusted net debt reaches ¥10.81 trillion, versus ¥8.20 trillion at the end of March. It therefore rises by ¥2.61 trillion, or 31.8%, in three months. Holding-company cash declines from ¥3.50 trillion to ¥2.30 trillion, a fall of ¥1.20 trillion or 34.3%.

Holding-segment finance costs increase from ¥146.6 billion to ¥281.9 billion, a calculated rise of 92.3%. This does not mean every yen reflects new permanent debt: currencies, timing and instruments matter. It nevertheless shows that financing capacity carries a rising cost.

SoftBank still publishes a loan-to-value ratio, or LTV, of 13.0% at 30 June. It divides ¥10.81 trillion of adjusted net debt by ¥83.11 trillion of adjusted holdings. The reported snapshot remains moderate, but its denominator depends heavily on Arm and other market valuations.

Our article on SoftBank's two readings of debt explains why this 13% cannot be compared directly with Fitch's 28% scenario for March 2027. Dates, assumptions and methodologies differ.

5. Infrastructure turns strategy into cash requirements

AI models do not operate alone. They require chips, buildings, electricity, networks and equipment. SoftBank says it has already advanced approximately $6 billion for US power-generation and data-centre equipment.

Management presents these outlays as a bridge: the holding company supplies cash first and expects the projects to be refinanced when operational. That mechanism may release the balance sheet later. Until then, the money has left and construction, timing and refinancing risk remain somewhere in the group.

Other announced commitments include a further $10 billion for OpenAI in October, $5.4 billion for ABB Robotics and $3.1 billion for DigitalBridge. Their calendars and scopes do not support treating the sum as one immediate bill. Together, however, they show the scale of SoftBank's transition from a technology portfolio to an industrial AI infrastructure strategy.

6. What the quarter changes for shareholders

The first reading would say SoftBank has won its bet because NAV is at a record and OpenAI is worth twice its accounting cost. It is incomplete. The second would focus only on rising debt and conclude the group is weakening. It is equally incomplete.

Opulion's reading connects both: SoftBank owns more value, but that value is concentrated and partly illiquid; its strategy simultaneously requires more real financing. Economic success will therefore depend less on a single quarterly figure than on the group's ability to convert part of that wealth into durable, affordable funding without forced sales.

Three pieces of evidence will matter in the next accounts: the precise source of funds replacing the OpenAI bridge, actual refinancing of infrastructure advances, and the combined path of cash, net debt and Arm concentration.