SoftBank repays a $25.9bn bridge: short-term debt goes, the OpenAI bet remains

A bridge loan carries a borrower across a short period until durable funding becomes available. SoftBank built this one to advance an additional $30 billion to OpenAI without waiting to sell assets or arrange every part of its replacement financing. On 9 September, the holding company announced that it will repay the $25.9 billion still outstanding on 15 September, even though contractual maturity was not until 25 March 2027.

The important word is "bridge". SoftBank is removing almost the entire temporary structure, but the asset purchased with the money remains on the balance sheet. The release identifies neither the cash, disposals nor new financing that will fund repayment. For SoftBank shareholders, the real issue is therefore not merely the disappearance of short-term debt. It is the transition from temporary funding to a more permanent structure when the group still has a third $10 billion OpenAI instalment due and other artificial intelligence and robotics projects to finance.

Why $25.9 billion does not mean $25.9 billion of deleveraging

Net debt is calculated by subtracting cash and certain liquid financial assets from interest-bearing debt. If SoftBank uses $25.9 billion of cash to repay $25.9 billion of debt, gross debt falls but cash falls too. All else being equal, net debt does not fall when the payment is made.

If SoftBank replaces the bridge with a new bond or another borrowing, the short-term debt disappears but gross debt remains. Maturity, interest rate, currency, guarantees and any assets supporting the financing change. Finally, if the group has generated cash through disposals, fund distributions or operating flows, net debt can genuinely fall, but the economic effect begins when the cash is created, not when the repayment transfer is executed.

This distinction avoids the central misreading. Repayment is positive information about liquidity and funding access: SoftBank is not waiting until March 2027 to deal with a very large maturity. It is not yet proof of a $25.9 billion reduction in net debt.

The bridge begins with OpenAI

On 27 February, SoftBank committed to invest an additional $30 billion in OpenAI through SoftBank Vision Fund 2. Three $10 billion instalments were scheduled for 1 April, 1 July and 1 October. Following the third, SoftBank's cumulative investment in OpenAI was expected to reach $64.6 billion, representing an estimated approximately 13% interest under the group's published assumptions.

Initial funding was to come from bridge loans and other banking arrangements, then be replaced over time by existing assets and other financing measures. On 27 March, SoftBank therefore signed a $40 billion unsecured facility involving more than thirty financial institutions around international and Japanese lead banks.

The new release supports a simple calculation: $30 billion drawn less $25.9 billion outstanding means $4.1 billion has already been repaid. It also shows that a third $10 billion drawing occurred after the two drawings explicitly linked to the April and July instalments. SoftBank does not say whether this third drawing prefunds October, serves general corporate needs or reflects an accelerated OpenAI timetable. That question must remain an unknown.

OpenAI is no longer a peripheral holding

At 30 June, SoftBank carried SVF2's OpenAI shares at JPY 14,549.2 billion. SVF2 investments measured at fair value totalled JPY 18,239.2 billion. Opulion calculation: 14,549.2 divided by 18,239.2 equals 79.8%. OpenAI therefore already represented approximately four fifths of SVF2's recorded investments at that date.

This is not OpenAI's direct weight in SoftBank's published adjusted holding value: the scopes and adjustments differ. It nevertheless shows why OpenAI financing matters directly to shareholders of the holding company. SoftBank is not funding a small experimental option. It is concentrating a major part of one of its principal investment vehicles in a single private company whose value is remeasured quarterly through profit or loss.

The risk is therefore not solely financial. It also involves concentration, the lack of a daily market price and the path to liquidity. An increase in OpenAI's value strengthens SVF2's portfolio. A decrease can reduce portfolio value, generate an accounting loss and complicate related financing.

The first visible replacement carries OpenAI risk itself

On 5 August, SVF2 signed a new $10 billion loan, with drawing expected during the month and repayment in August 2028. SoftBank presents it as financing using OpenAI shares. The accounts describe a more nuanced structure: the legal collateral is a cash collateral account at the borrowing vehicle, SoftBank Group guarantees the loan, and certain circumstances, including a significant fall in the fair value of the OpenAI preferred shares referenced by the agreement, can trigger cash collateral provision or mandatory prepayment.

This loan equals 38.6% of the $25.9 billion bridge balance due. It is therefore a credible part of the broader takeout process. It would nevertheless be incorrect to state that it finances exactly $10 billion of the bridge repayment: SoftBank does not make that connection in its 9 September release.

The transfer of risk matters. The March bridge was short term and unsecured. The August loan runs to 2028 but introduces explicit sensitivity to OpenAI's value and a holding-company guarantee. Maturity becomes longer; funding also becomes more tightly connected to the asset that motivated the investment.

The JPY 1,000 billion bond examined in our analysis of ABB Robotics financing belongs to the same broader transformation of liabilities. It cannot, however, be presented as the source of the OpenAI repayment: SoftBank officially allocates its proceeds to domestic bond redemption and partial funding of ABB Robotics.

Repayment exceeds the liquidity published at 30 June

At 30 June, SoftBank reported an effective cash position of JPY 2.3 trillion, including JPY 489.5 billion of undrawn capacity and taking account of prime-brokerage funding associated with certain bond investments. The announced repayment equals JPY 3.9819 trillion at the release's exchange rate. It is approximately 1.7 times that end-June liquidity.

This comparison is not a 15 September balance sheet. SoftBank could have raised, moved or generated further resources in the meantime. It nevertheless demonstrates that repayment cannot be read as a routine withdrawal from the cash balance published in the previous quarter. Replacement funding, monetisation or another liquidity source must complete the equation.

The cost of this balance-sheet acceleration was already visible. In the first quarter, consolidated finance costs almost doubled to JPY 328.7 billion. Interest expense at SoftBank Group increased by JPY 135.5 billion to JPY 265.8 billion, mainly because of the $11.5 billion Arm share-backed loan, the $20 billion of bridge drawings described in the group's August communication and a higher bond balance, in a more expensive rate environment.

Early repayment may therefore avoid several months of bridge cost. The precise saving is unknown because the bridge rate, any prepayment fees and the cost of replacement resources are not published. Calculating an interest saving without those data would be misleading.

LTV remains comfortable, but it needs two columns

SoftBank measures holding-company risk using LTV: adjusted stand-alone net debt divided by the adjusted equity value of holdings. At 30 June, net debt was JPY 10.81 trillion and holdings were JPY 83.11 trillion, giving 13.0%. Policy is to remain below 25% in normal conditions, with 35% as the upper threshold even in an emergency.

At first sight, the distance looks wide. At constant net debt, adjusted holding value would need to fall to JPY 43.24 trillion to reach 25%, a static decline of 48.0%. The 35% threshold would correspond to JPY 30.89 trillion, a decline of 62.8%. These are sensitivities, not forecasts: they assume, among other things, unchanged debt, exchange rates, investments and adjustments.

The second column is denominator composition. At 30 June, Arm represented JPY 49.91 trillion, or 60.1% of adjusted holding value. SVF2 represented JPY 19.29 trillion, or 23.2%. Together they formed 83.3% of the denominator. Headroom therefore depends heavily on Arm and on a vehicle in which OpenAI has become dominant.

The bridge repayment clearly improves the liquidity calendar. Its effect on LTV can only be established when the funding source and subsequent figures are published. If payment consumes cash, the net numerator changes little. If debt treated as asset-backed replaces holding debt, adjustments to numerator and denominator may change. If assets are sold, the outcome depends on the price received and the use of cash.

Opulion view: less timing risk, not yet less economic risk

SoftBank has done what is expected from a holding company able to reach multiple capital markets: it has not allowed $25.9 billion of debt to run into its final months. It removes a massive maturity from the short-term calendar after broadening its lending group, arranging longer funding around OpenAI and continuing to issue bonds.

Our inference is that timing risk decreases, while economic risk moves. It moves into replacement funding costs, guarantees, portfolio concentration and OpenAI's ability to justify a value that has become central to SVF2. This reading must be tested, not assumed, when SoftBank publishes its next balance sheet.

The paradox is therefore more precise than the press-release headline. The banking bridge disappears, but the journey is not complete. A third $10 billion OpenAI instalment remains scheduled for 1 October. The ABB Robotics acquisition and infrastructure projects continue to require capital. For SoftBank shareholders, the next decisive disclosures are the exact source of the $25.9 billion, net debt after payment, LTV treatment of new financing and the annual cost of the liabilities replacing the bridge.