NextFin News - SoftBank Group has lifted its margin loan secured by Arm Holdings shares to $25 billion, a move that turns the chip designer's 2026 rally into fresh dry powder for founder Masayoshi Son's artificial-intelligence offensive. The increase, reported September 18, 2026, adds roughly $5 billion to a facility that stood at $20 billion in November and leaves the loan at about one-tenth of the value of the Arm stake pledged against it. The question is no longer whether SoftBank can borrow against Arm. It is how much of the group's future depends on the stock never falling far.
The Financing Ladder: From $8 Billion to $25 Billion
SoftBank's use of Arm as collateral predates the chip designer's public listing. Ahead of Arm's September 2023 initial public offering, the group lined up roughly $8 billion in margin financing from eleven banks, including JPMorgan Chase, Barclays, BNP Paribas, Credit Agricole and Goldman Sachs, linking mandates for the listing to the loans. That facility was the first rung of a ladder SoftBank has kept climbing.
By March 2025, the group had raised about $13.5 billion from Arm margin loans, with $5 billion still undrawn, according to its earnings statement. Six months later, in its November 2025 investor briefing, SoftBank disclosed a $20 billion total facility with $8.5 billion drawn as of September 30, 2025. The latest increase to $25 billion — confirmed as the Arm-backed borrowing grew alongside the stock — takes the ceiling to a new high even as the collateral has appreciated faster than the debt.
Arm's market capitalization reached roughly $255 billion to $283 billion in mid-September 2026, according to market data, roughly quadrupling from its roughly $65 billion IPO valuation. SoftBank owns a controlling stake of about 86 percent in Arm through the Vision Fund, a position acquired in 2016 for $31.4 billion. At current prices the stake is worth well over $200 billion, which puts the $25 billion loan at a loan-to-value ratio of roughly 9 percent to 12 percent — comfortably inside the roughly 25 percent LTV ceiling that analysts say the group approaches across its combined AI-linked borrowings.
Where the Money Goes: An AI Balance Sheet Built on Collateral
The proceeds are being funneled into SoftBank's expanding artificial-intelligence commitments, which have become the defining project of Son's second act. The group has pledged as much as $30 billion toward OpenAI, the maker of ChatGPT, and in March 2026 closed a $40 billion bridge loan — its largest-ever dollar-denominated borrowing — to help fund that position. In mid-September 2026 it also secured an upsized $11.9 billion margin loan secured against its OpenAI stake, drawing commitments from around 20 banks. Stacked together, SoftBank's OpenAI-linked borrowing alone exceeded $50 billion for 2026.
Beyond OpenAI, the pipeline is broad. SoftBank raised a $15 billion one-year facility earlier in 2026 for U.S. AI investments, agreed to buy ABB's robotics business for $5.4 billion, and announced a $6.5 billion takeover of Ampere Computing. Analyst Sharon Chen estimated in a note that SoftBank's funding needs could exceed $30 billion given the deal pipeline, and that asset sales and asset-backed financing could reduce reliance on bond markets. JPMorgan estimates that debt tied to AI has ballooned to about $1.2 trillion, making it the largest segment of the investment-grade market — a pool into which SoftBank is now one of the deepest swimmers.
The centerpiece of that spending is the Stargate initiative, a $500 billion AI-infrastructure commitment announced at the White House in January 2025 with OpenAI and Oracle. SB Energy, SoftBank's infrastructure arm, is executing the data-center buildout; OpenAI and SoftBank each invested $500 million in SB Energy in January 2026, and the unit has confidentially filed for a U.S. IPO as it constructs a 1.2-gigawatt campus in Milam County, Texas. Son has framed the spending as unavoidable. Speaking on July 14, 2026, he said AI development will require investment of about $5 trillion a year globally by 2040 and dismissed talk of a bubble as "absurd." He has also called the AI revolution "50 times bigger" than the dot-com boom. "We will go from a human-centric world to an agent-centric world," he said. "The age when humans are the highest life form on earth will end. For better or for worse, it will happen and it can't be stopped."
Why Arm Is the Collateral of Choice
Arm is an unusually clean piece of collateral, and that is precisely why SoftBank keeps returning to it. The Cambridge-based company licenses processor architectures and collects royalties rather than manufacturing chips, producing margins that resemble software more than semiconductors. Arm reported a gross margin of 98.2 percent and an operating margin of 43.0 percent, according to chief executive Rene Haas in SoftBank's 2026 group report. Its technology is used in more than 99 percent of smartphones, and management has guided to roughly $25 billion in annual revenue by fiscal 2030 — about $10 billion from its intellectual-property and custom-silicon business and more than $15 billion from newer chip ventures.
The underlying revenue base is already scaling at a pace that justifies lender confidence. In its fiscal year ended March 31, 2026, Arm reported record revenue of $4.92 billion, a 23 percent increase and the third consecutive year of more than 20 percent growth since going public. Royalty revenue reached a record $2.61 billion, up 21 percent, while licensing and other revenue hit $2.31 billion, up 25 percent. In the fourth quarter alone, data-center royalties more than doubled year over year, and the company launched its first Arm-designed data-center chip with more than $2 billion of customer demand across fiscal 2027 and 2028. That is the profile lenders want to see behind collateral: a franchise with recurring, high-margin cash flows and a valuation that has already rerated rather than one still searching for a multiple.
For SoftBank, every dollar of Arm appreciation creates fresh borrowing capacity without selling a single share or diluting its control. It is financing without dilution, and that is the entire appeal. The alternative — issuing equity or selling assets — would either dilute existing shareholders or reduce SoftBank's grip on the asset that has become its largest holding, representing more than half of its net asset value.
The Mechanism: Recycling One Asset Into Many Bets
The structure of SoftBank's financing is the real story, and it runs in a loop. Asset prices rise. Collateral values increase. Lenders extend larger margin facilities. SoftBank deploys the cash into more AI exposure. More exposure ties the group's fortunes more tightly to the same AI cycle, which — if it continues — pushes asset prices higher still and repeats the loop.
This is not a one-off transaction. It is a deliberate capital model that replaces the old SoftBank playbook. The group once operated as a diversified technology fund that used leverage occasionally. It has become a concentrated, collateral-recycling balance sheet whose strategy depends on AI assets staying bid. The transmission channel is straightforward: Arm and OpenAI valuations are the input, margin facilities are the transmission belt, and a growing stack of AI commitments is the output. The model works as long as the input keeps rising or at least holds steady.
The second-order consequence is what most investors are not pricing. A margin loan against Arm does more than raise cash; it concentrates SoftBank's risk in a single theme twice over. The group already owns the asset. By borrowing against it to buy more AI exposure, SoftBank is effectively leveraging its existing AI bet to make new AI bets. If the AI capex cycle slows, Arm's growth narrative weakens and the stock falls — at the same moment that the OpenAI-linked assets funded by the loan come under pressure. The correlation that makes the strategy profitable in an up-cycle is the same correlation that removes the escape hatch in a down-cycle.
There is a third-order effect as well, and it runs through the cost of capital. As SoftBank's AI-linked debt stack grows, lenders price the concentration. The group's OpenAI margin borrowing illustrates the pattern: it targeted $10 billion, cut the target to $6 billion amid creditor hesitation, then ultimately secured an upsized $11.9 billion facility. Creditor appetite is strong but not unconditional. If the stack keeps growing, the next increments of borrowing will carry wider spreads, tighter covenants, or both. That is how leverage quietly becomes a drag on returns before it ever becomes a crisis.
Cyclical Windfall, Structural Shift
Two forces are at work here, and they point in different directions across time horizons. The Arm share rally is cyclical: it is a valuation wave driven by AI enthusiasm, and valuation waves mean-revert. History offers three relevant comparisons. After Arm's 2023 IPO the stock roughly doubled in its first weeks before giving back most of the gain over the following year. During the 2022 semiconductor downturn, chip-design and semiconductor-equipment stocks fell 40 percent to 60 percent from peaks. And in the 2000 dot-com collapse, semiconductor names lost 70 percent to 90 percent. None of these cycles is a perfect analog, but all three show that even the highest-quality chip franchises can see their multiples compress violently when sentiment turns.
The financing behavior, by contrast, is structural. SoftBank is no longer a diversified fund that occasionally uses leverage; it has become a concentrated balance sheet whose strategy depends on AI assets appreciating. That regime shift will not reverse on its own. Unwinding it would require asset sales, equity issuance, or a sustained pause in new commitments — none of which Son has signaled. The short-term leg is the Arm rally; the long-term leg is the leverage structure built on top of it. Confusing the two is the most common error in reading this story.
"We will go from a human-centric world to an agent-centric world," Masayoshi Son said in July 2026. "The age when humans are the highest life form on earth will end. For better or for worse, it will happen and it can't be stopped."
The Counter-Thesis: Plenty of Cushion Left
The strongest argument against alarm is simple arithmetic. With Arm's market capitalization near $255 billion to $283 billion and SoftBank's stake worth well over $200 billion, a $25 billion loan is a modest fraction of the collateral. Even a 30 percent decline in Arm shares would leave the loan comfortably covered, with the stake still worth roughly $140 billion to $160 billion. Arm's 98 percent gross margin and its presence in more than 99 percent of smartphones make it one of the most defensible assets in the technology complex, and SoftBank's broader balance sheet includes listed holdings — T-Mobile, Deutsche Telekom, SoftBank Corp., Intel — that it can sell before any margin call touches the Arm position.
There is also a timing argument. Margin loans are typically structured with maintenance covenants and periodic resets rather than hard maturities, giving SoftBank room to manage collateral values over time. And Son has repeatedly shown a willingness to monetize winners: he sold SoftBank's remaining Nvidia stake for about $5.8 billion in October 2025, and SB Energy's planned IPO would convert an illiquid infrastructure stake into cash. OpenAI itself is widely expected to list within the next couple of years, which would give SoftBank a liquid market for its largest private position.
That case is credible — up to a point. The risk is not the Arm loan in isolation; it is the stack. Add the $11.9 billion OpenAI margin loan, the $40 billion bridge loan maturing in March 2027, the $15 billion one-year facility, and the Ampere and ABB deals, and SoftBank's AI-linked obligations for 2026 approach or exceed $100 billion. If Arm and OpenAI valuations fall together — a correlated drawdown in the same AI theme — the cushion shrinks across the whole structure at once. Diversification is the only free lunch in finance, and SoftBank has spent 2026 giving it away.
What to Watch: The Signals That Would Break the Thesis
The falsifying signal is specific and quantifiable: if Arm's market capitalization falls below roughly $180 billion — about a 30 percent decline from mid-September levels — the collateral cushion compresses enough that lenders could demand additional collateral or SoftBank could face pressure to deleverage. That threshold is not arbitrary; it is roughly the level at which the combined AI-linked loan stack begins to consume a dangerous share of the flagship collateral's value.
A second signal is the maturity wall. The $40 billion bridge loan runs to March 2027, and the $15 billion facility is a one-year instrument. How those are refinanced — with longer-term bonds, asset sales, or more margin borrowing — will reveal whether the strategy is sustainable or simply being rolled forward. A successful $10 billion to $20 billion bond sale, which the group has discussed, would be a positive sign. A third margin loan against the same assets would not be.
The scenarios split cleanly by time horizon. In the base case, AI capex continues to grow, Arm's revenue compounds toward its fiscal-2030 guidance, and SoftBank refinances its bridge and one-year facilities into longer-dated bonds while the Arm loan sits quietly as cheap, collateralized funding. In the upside case, OpenAI lists at a valuation that validates the stack, SB Energy's IPO unlocks billions, and Son's "agent-centric world" thesis pulls the whole portfolio higher. In the downside case, AI spending growth decelerates, Arm's multiple compresses below $180 billion, and the group faces a choice between selling assets into a weak market or accepting tighter, costlier financing.
Short term, SoftBank's shares have ridden the AI wave, up roughly 46 percent in 2026 and more than 200 percent over twelve months as of September. Medium term, the group's earnings will increasingly reflect the cost of its debt and the performance of its AI bets rather than its legacy fund portfolio. Long term, the question is whether Son has built a durable AI balance sheet or simply concentrated risk in the most crowded trade of the decade.
The $25 billion Arm loan is not just funding. It is a statement that SoftBank would rather bet the farm than miss the AI boom. The market will decide whether that is conviction or leverage dressed up as vision.
Explore more exclusive insights at nextfin.ai.
