The numbers that matter in the SoftBank OpenAI bet are not the ones printed in red.

By Blumefield | November 10, 2026

A Loss the Market Decided to Ignore

On paper, the quarter that defined the SoftBank OpenAI bet looked like a disaster. SoftBank Group reported a net loss of ¥1.12 trillion, roughly $7 billion, for the three months to the end of September, its first quarterly loss in two years. A year earlier the Japanese technology investor had booked a healthy profit. Net sales actually rose about 16 percent to ¥2.22 trillion, so the damage came entirely from the investment side of the house, where Masayoshi Son has spent eighteen months rewiring the company around that single wager.

And yet the shares climbed almost 14 percent in Tokyo. That gap between a grim headline number and a buoyant market reaction is the whole story. Investors have stopped pricing SoftBank on its reported earnings and started pricing it on the SoftBank OpenAI bet, a holding now large enough to make the rest of the balance sheet look like background noise. The loss, once you read past the first line, turned out to be the cost of going all in.

Where the Red Ink Came From

Most of the quarter's pain was accounting pain, not cash walking out the door. The single biggest hit was a ¥597 billion charge, about $3.7 billion, tied to a realignment of the gas-fired power equipment SoftBank had been buying to feed American AI data centers. Through a vehicle called Silver Bands 7, the group had committed to turbines and generators on a scale that outran its own project pipeline. It wrote down ¥457 billion of non-refundable advances and swallowed another ¥140 billion unwinding and modifying contracts. Cumulative payments through that vehicle have already passed ¥1.5 trillion, a reminder that the physical plumbing of the AI boom is being bought years ahead of the demand it assumes.

The second drag was stranger. SoftBank raised $7.35 billion in August against its Intel shares using prepaid forward contracts, a way to pull cash forward without selling. Because Intel's stock then rallied hard, the hedge moved against the company and generated a derivative loss of ¥322 billion. In other words, one of its bets worked so well that the financing wrapped around it produced a reported loss. A further ¥204 billion loss came from warrants handed to OpenAI as incentives on data-center leases. None of these are the marks of a portfolio in trouble. They are the friction of a company that has bent everything toward the SoftBank OpenAI bet while borrowing, hedging and building at the same time.

The $45 Billion That Changed Everything

Strip away the noise and one number explains the market's calm. At the end of September, SoftBank's stake in OpenAI carried a cost basis of $54.6 billion and a fair value of $99.6 billion. That is a paper gain of roughly $45 billion, a multiple of about 1.82 times invested capital, and it sits on the balance sheet quietly because accounting rules meant no fresh gain flowed through the quarter's profit line. The market, unlike the income statement, is happy to count it.

The SoftBank OpenAI bet is also still growing. The group has committed to a $30 billion follow-on in three $10 billion tranches across April, July and October, lifting its total investment to about $64.6 billion and its ownership to roughly 13 percent. That later money went in at a $730 billion pre-money valuation, a figure that would have sounded absurd a year ago and now functions as the anchor for SoftBank's entire equity story. The conviction is not new. Last year the company sold its entire stake in Nvidia, the most obvious winner of the AI trade, to concentrate firepower on OpenAI instead. That decision looks either visionary or reckless depending on the week.

Borrowing Against the Future

Concentration on this scale has to be financed, and this is where the SoftBank OpenAI bet starts to look less like investing and more like a highly leveraged machine. The group reported net asset value of about ¥59.8 trillion, close to $380 billion, with loan-to-value still modest at around 16 percent. Those are comfortable numbers. The question is how quickly they could stop being comfortable if the OpenAI mark ever reversed.

The financing moves tell their own story. SoftBank took out a $10 billion loan secured against OpenAI shares held inside its second Vision Fund, lifted a margin loan backed by Arm by $5 billion, issued billions in bonds, and layered on acquisitions including the $1.75 billion leveraged purchase of ABB's robotics arm and a $3.1 billion deal for DigitalBridge. It now owns the chip designer Ampere outright. Each piece is defensible on its own. Together they describe a company pledging its best assets to feed the SoftBank OpenAI bet, with the Vision Funds, once the centerpiece, now reduced to a supporting role after their investment gains fell nearly 80 percent year on year.

Son's All-or-Nothing Machine

Masayoshi Son has never hidden what he is building. He talks openly about artificial superintelligence and frames SoftBank around four pillars: the AI model, the AI chip, the AI infrastructure and physical AI, meaning robotics. OpenAI supplies the model, Ampere and Arm the silicon, the Stargate data-center program the infrastructure, and acquisitions like ABB the robots. The SoftBank OpenAI bet is the keystone that holds the arch together. Son has described his own method as working the gas and the brake at once, though this quarter looked mostly like gas.

The danger is circularity. SoftBank is helping finance data centers that OpenAI will lease, buying power equipment those centers will need, and marking its returns against a valuation that other AI-soaked investors helped set. If the AI capital cycle keeps expanding, the structure compounds beautifully. If it stalls, the same leverage that magnifies the upside works in reverse, and a paper loss on the OpenAI position could force real decisions about real debt. For now, markets have cast their vote, and the verdict is that one spectacular investment is worth more than a quarter of red ink. Whether that holds is the sharpest question in technology finance, and it is one worth following across Blumefield's continuing coverage of the AI capital cycle.

Son spent years being mocked for missing the first internet giants and for the WeWork debacle. He has answered by making the boldest concentrated bet of his career. The next few quarters will decide whether the SoftBank OpenAI bet is remembered as the trade that defined the AI era or the one that finally caught up with him. Readers can track the fallout as it unfolds at Blumefield.