SoftBank Group's Record NAV Hides a Harder Earnings Story

SoftBank Group’s net asset value reached a record 72.3 trillion yen in the June quarter, propelled by an extraordinary rally in Intel shares, even as net income fell nearly 18 percent.

Key Takeaways

• An unrealized 1.33 trillion yen gain on Intel shares, which nearly tripled in three months, pushed SoftBank’s net asset value to an all-time high of 72.3 trillion yen.
• Net income attributable to shareholders dropped 17.7 percent to 347.3 billion yen, as rising compensation costs, finance charges and currency losses outpaced investment gains.
• SoftBank pushed its OpenAI stake toward 13 percent with 20 billion dollars in fresh capital during the quarter, as Codex usage grew twenty five fold in just seven months.

The Arithmetic of a Record

SoftBank Group closed its books on the quarter ended June 30 with a number Masayoshi Son has spent a career chasing: a net asset value of 72.3 trillion yen, the highest the company has ever reported. It is the figure management prefers above all others, a running tally of what the portfolio is worth once debt is subtracted, and by that measure the quarter was triumphant. By the measure that actually belongs to shareholders, it was not. Net income attributable to owners of the parent fell 17.7 percent to 347.3 billion yen, a decline that sits uneasily beside a record valuation and demands its own explanation.

The reconciliation is instructive. Net sales rose 10.9 percent to 2.02 trillion yen, and total gains on investments very nearly quadrupled to 1.86 trillion yen, yet almost none of that flowed through to the bottom line. Selling, general and administrative expenses climbed 70 percent, largely on share based compensation for engineers at Arm and for staff at the energy subsidiary building SoftBank’s American data centers. Finance costs nearly doubled. A weaker yen produced a net foreign exchange loss, and a derivative charge tied to warrants issued as part of a data center lease agreement added further drag. Minority shareholders in Arm, SoftBank Corp and PayPay also claimed a larger slice of consolidated profit than a year earlier, up 20.6 percent, a quiet reminder that SoftBank increasingly shares its growth with partners it does not fully own.

One Stock Carries the Quarter

Strip away a single position and the quarter looks considerably less impressive. The Investment Business of Holding Companies segment, which houses SoftBank’s directly owned stakes, swung from a loss of 20.9 billion yen a year earlier to income of 1.05 trillion yen this quarter, almost entirely on the back of Intel. Its shares rose from $44.13 at the end of March to $139.63 at the end of June, a move chief financial officer Yoshimitsu Goto described on the earnings call as a positive surprise rather than any deliberate strategy playing out. Inside the Vision Funds, a similar dynamic held: a 2.2 billion dollar mark up in ByteDance carried SVF1 to a solid quarterly gain, while SVF2 slipped into loss as PayPay and Symbotic shares softened, offset only partly by markups on unlisted holdings.

Arm tells the more interesting story. Consolidated under IFRS alongside Ampere and Graphcore, the AI Computing segment posted a loss of 200.8 billion yen, a deterioration of 168.4 billion yen from the prior year, driven by research spending and compensation for a fast growing engineering staff. Set against that is Arm’s own standalone report, filed separately on Nasdaq under American accounting rules: record quarterly revenue of 1.29 billion dollars, up 22 percent, and adjusted earnings per share up 29 percent to 45 cents, ahead of guidance. Arm’s stock fell 8 percent the day it announced those numbers anyway. The chip designer at the center of SoftBank’s AI strategy is executing about as well as a company can, and the market is still deciding what that is worth.

The OpenAI Wager Compounds

No commitment illustrates SoftBank’s appetite better than OpenAI. The company funded the first 10 billion dollar installment of a 30 billion dollar pledge made in February during the quarter, lifting cumulative investment to 44.6 billion dollars against a fair value of 89.6 billion dollars, an unrealized gain of 45 billion dollars, with the valuation still anchored to the price of the last funding round. A second 10 billion dollar tranche followed in July, a third is scheduled for October, and by year end SoftBank’s cumulative stake will approach 65 billion dollars and roughly 13 percent ownership. Mark Agni of SB Investment Advisers put the rationale plainly to analysts, describing OpenAI as more of a partner than an investment and explaining that the firm has chosen to concentrate its exposure to large language models there rather than spread it across rivals.

The conviction is not abstract. Combined weekly users of OpenAI’s Codex and its newly launched ChatGPT Work grew roughly twenty five fold in seven months, with Codex alone climbing from about 400,000 weekly users in December to 10 million in July. SoftBank is building the physical capacity to keep pace with that growth: a five gigawatt data center program in northern France, unveiled at the Choose France summit with a first phase commitment of 45 billion euros, a ten gigawatt project in Ohio nearing a signed lease, and a site in Texas already leased to OpenAI and under construction. The bet is no longer confined to a capitalization table. It now runs through concrete and turbines.

Discipline Behind the Spending

For all the aggression of its dealmaking, SoftBank’s balance sheet management remains conspicuously conservative by its own standards. The loan to value ratio, net debt measured against gross portfolio value, stood at 13 percent at quarter end, comfortably inside the 25 percent ceiling the company has set for itself, and S&P revised its outlook on SoftBank to stable from negative during the period. The financing was nonetheless substantial: 5.56 trillion yen raised and 3.94 trillion yen repaid on a consolidated basis, including full repayment of the prior year’s bridge loans used to fund the Ampere acquisition and the initial OpenAI commitment, alongside new hybrid bonds, foreign currency senior notes and a fresh 20 billion dollar drawdown earmarked largely for further OpenAI funding.

Consolidated cash fell to 3.93 trillion yen from 5.36 trillion yen at the start of the fiscal year, the result of 2.54 trillion yen in net investing outflows directed chiefly at OpenAI and at power and data center assets in the United States. SB Energy, the subsidiary building much of that American infrastructure, has raised roughly 19 billion dollars in project financing and filed confidentially in May for a listing of its own. A new joint venture with SoftBank Corp will pursue a domestic cloud computing business built on that same power infrastructure, and the previously announced acquisition of ABB’s robotics unit is expected to close before year end. None of this reads as a company retreating from risk. It reads as one choosing precisely where to take it.

A Verdict Written in the Share Price

Investors delivered their own judgment on the day. SoftBank shares fell 4.41 percent to 5,695 yen from a previous close of 5,958 yen, and management conceded that its pro forma net asset value had already slipped to roughly 58.3 trillion yen by August 5 as broader markets retreated from their late June highs. The stock’s 52 week range, from 3,116 to 9,074 yen, says as much about the company as any single quarter’s results: these are shares that move with the daily marks on Arm, Intel and a portfolio of private AI companies, not with a steady operating cash flow.

That volatility exists inside a considerably larger ambition. At June’s shareholder meeting, Son set a fifteen year target of lifting net asset value to 1,000 trillion yen, roughly fourteen times where it stands today, and Goto used the earnings call to reject the suggestion that SoftBank’s data center buildout amounts to speculative excess, arguing instead that computing supply remains structurally short of demand. The risks closer at hand are specific rather than philosophical. Arm’s new AGI processor has already drawn more than 2 billion dollars in orders against an original guide of roughly 1 billion, yet wafer capacity, not demand, is what will determine how much of that order book becomes revenue, and the Ohio and France projects still hinge on leases and financing that have not yet been signed. The record is genuine. So is the exposure that produced it.