Abstract

On May 13, 2026, SoftBank Group (9984) announced that its consolidated net profit for the fiscal year ending March 2026 reached 5.0022 trillion yen—4.3 times the previous year's figure—setting a new all-time record for annual profit by a Japanese company. While the result was boosted by enormous investment valuation gains at the Vision Fund, driven by a sharp surge in the valuation of its investee, the U.S.-based OpenAI, deep-rooted negative factors are also surfacing one after another, including a downgrade in its credit rating outlook, concentration risk stemming from its "OpenAI single-pillar strategy," and internal conflict over the Stargate project.

Breaking down the ¥5,002.2 billion figure: first off, valuation gains on the prized OpenAI stake account for the bulk of it

First, it is worth reviewing the earnings highlights for SoftBank Group (hereafter SBG). Consolidated net profit for the full fiscal year ending March 2026 came to ¥5.0022 trillion, swelling to 4.3 times the level of the previous fiscal year (which was on the order of ¥1.153 trillion). This exceeds the roughly ¥4.9879 trillion the company recorded in the fiscal year ending March 2021, marking the highest annual net profit ever for a Japanese company. It also surpasses the roughly ¥4.9449 trillion net profit that Toyota Motor posted in the fiscal year ending March 2024, meaning that an investment company—rather than a manufacturer—has reclaimed this position. Looking at the most recent January–March quarter (the fourth quarter) alone, net profit swelled to ¥1.8296 trillion, 3.5 times the figure for the same period a year earlier, vastly exceeding the median analyst estimate (around ¥295.2 billion) that Bloomberg had compiled in advance.

SBG's earnings become easier to understand if one first grasps that a profit of this magnitude is not operating profit earned by selling goods, the way Toyota or Sony do. SBG is, after all, an investment holding company, and most of its profit is the "increase in the price of the startup shares it holds" recorded as accounting profit. This fiscal year, the company that contributed most to inflating those "gains on the appreciation of held shares" was OpenAI of the United States, the world's largest generative AI firm. According to CNBC, SBG's Vision Fund (the combined total of SVF1 and SVF2) posted investment gains of about $46 billion (roughly ¥6.9 trillion) for the full year, the bulk of which were OpenAI-related. As of the end of March, the valuation of the OpenAI shares held by SBG reached about $79.6 billion (roughly ¥11.9 trillion), which works out to a cumulative unrealized gain of about $45 billion (roughly ¥6.75 trillion) against the roughly $34.6 billion (roughly ¥5.19 trillion) in capital it had originally invested.

To put it in an easy-to-understand analogy, SBG has become a company where "simply by holding the promising stock that is OpenAI, accounting profit piles on all at once when shares are marked to market value at the end of the fiscal period." When OpenAI, the developer of ChatGPT, leapt to a valuation of about $852 billion (roughly ¥127.8 trillion; some reports cite a figure of $890 billion) in its February 2026 round, an enormous "gift" rolled in onto SBG's books—and that is the essence of these earnings results. The Nikkei has likewise cautioned that "most of the net profit comes from valuation gains on investees, and the revenue structure differs from that of operating companies such as manufacturers," positioning it as something distinct from Toyota-style profits earned from a core business.

Arm and PayPay, light and shadow — the gap in performance has actually widened

Apart from OpenAI, one standout is Arm Holdings, the Nasdaq-listed semiconductor IP heavyweight. For the fiscal year ending March 2026, full-year revenue rose 23% year over year to $4.92 billion (roughly ¥740 billion), setting a record high, while fourth-quarter revenue also climbed 20% to $1.49 billion (roughly ¥224 billion). Arm is the prized asset that SBG, starting in 2024, shifted out of the Vision Fund and into SBG itself (the holding-company investment business), and it is positioned as the very thing that gives SBG enough collateral value to fight a "war of attrition until OpenAI's IPO." The fact that data-center revenue more than doubled year over year is also evidence that Arm has begun to directly capture AI infrastructure demand, including the Stargate project. On the other hand, quarterly royalty revenue of $671 million (roughly ¥100 billion) fell short of market expectations (about $693 million), and Arm's own share price briefly fell 9% after the FY2026 outlook was announced. CEO Rene Haas said "the weakness is concentrated in the low-end market," and he cannot conceal the fact that a plateauing smartphone market is squeezing royalty income.

PayPay, the domestic cashless service, listed on the Nasdaq Global Select Market on March 12 (ticker: PAYP), closing its first day at $18.16, about 14% above its $16 offering price, debuting with a market capitalization of roughly $12.1 billion (about ¥1.82 trillion). For SBG, it was a symbolic event in which a payments subsidiary that had long posted losses finally turned into a "convertible public asset," with quarterly transaction volume swelling to roughly ¥14.3 trillion (up 24% year over year). However, the cash flow gained from PayPay's listing is, by orders of magnitude, smaller than the enormous funding needs required for the additional investment in OpenAI discussed below.

Meanwhile, some listed portfolio holdings, such as the Southeast Asian ride-hailing service Grab Holdings, have produced valuation losses, and Bloomberg has also pointed out the dynamic in which "the rise in OpenAI's valuation offsets the valuation declines of other portfolio companies." Behind this headline of surpassing ¥5 trillion, it would be more accurate to read it as the "lopsidedness of the substance" — that listed-equity investments other than OpenAI are lackluster — having in fact widened.

How Each Newspaper Reported It — the Sober Assessment Seeping Through Behind the Headlines

Media coverage was polarized. Kyodo News, Jiji Press, and the Nikkei reported the factual elements calmly—"a record high for a Japanese company," "¥5.0022 trillion," "boosted by OpenAI's valuation"—and in the comment sections of Yahoo News, measured reactions stood out, such as "reaching ¥5 trillion is impressive, but the substance is unrealized gains." CNBC headlined its piece "SoftBank posts $46 billion gain at Vision Fund driven mainly by massive OpenAI bet," deliberately choosing the word "bet." Rather than praising the numbers, this phrasing emphasizes "the current reality in which all of the company's assets are concentrated in OpenAI," reflecting the basic temperature of the U.S. media.

On the morning of the earnings announcement, Bloomberg ran a story titled "SoftBank Group Reports Earnings Today, With Focus on Monetizing AI Investments and Cash Flow," noting that "regarding its investment in OpenAI, intensifying competition with the likes of Google and Anthropic of the U.S., as well as a damages claim from Elon Musk, continue to simmer." Its affiliated media also wrote that "credit default swaps (CDS) have widened more than 20% since the start of the year, standing out as exceptionally high even among Japanese companies," carefully presenting the gap between "the record-profit headline" and the chilly reality that "the credit markets have, in fact, begun to price in deterioration." Likewise, a column in Bloomberg's opinion section titled "OpenAI Is So Yesterday — Even for SoftBank's Son" appeared in early May, taking a rather cynical angle on the fact that Masayoshi Son himself has begun shifting his focus toward physical AI (robotics).

Under the headline "SoftBank Group's Finances: A War of Endurance Until OpenAI's IPO," the Nikkei has repeatedly laid out, even before the earnings release, the framing that the premise of the picture Son envisions depends on the scenario in which "OpenAI goes public within the deadline and its shares can be converted into cash." Kabutan News, which issued a flash report on the earnings, straightforwardly rated them as "full-year net profit up 4.3-fold, the highest in five years," while calmly adding at the end that "guidance for the current fiscal year is undisclosed." Given its nature as an investment company, SBG has traditionally not issued earnings forecasts, but at a time when massive cash outflows continue in the AI domain, many in the market interpret the stance of "not answering how much it will earn" as "the company itself cannot see ahead."

Silicon Valley VCs' take: "Already a de facto OpenAI proxy ETF"

In the U.S. venture capital industry, there is a growing tendency to treat SBG stock as "a publicly listed proxy for investors who cannot bet directly on the privately held OpenAI." After Andreessen Horowitz (a16z) established a new fund of roughly 15 billion dollars (about 2.25 trillion yen) in January 2026, single-handedly capturing about 18% of the U.S. VC market, dry powder for seed- and growth-stage AI startups remains abundant. Yet, as Bloomberg reported in January 2026 that this same a16z had assembled a "3 billion dollar AI bubble short-selling strategy," a move to bet back against the surge in AI valuations itself coexists alongside it. In short, even Silicon Valley's top VCs harbor the view that "the OpenAI-skewed growth in valuations will not last long."

Partners at major VCs such as Sequoia Capital, General Catalyst, Founders Fund, and Khosla Ventures have stated publicly that "SBG's strategy stakes its fate too heavily on the success or failure of a single company's IPO" and that "it is risky to assume the story of the lucky successes with Yahoo! and Alibaba can be replicated," with comments invoking the memory of WeWork's collapse being especially conspicuous. Among Wall Street research houses, TD Cowen's Krish Sankar maintains a "Hold (neutral)" rating and a stern price target of 13 dollars, implying roughly 31% downside from the current share price. Even bulls like Wedbush Securities' Dan Ives append the caveat that "this is the gateway to an AI infrastructure revolution, but the gap between valuation gains and cash realization has not been closed."

Among VC insiders, the fact that a16z, D. E. Shaw Ventures, and others lined up as underwriters of the round OpenAI raised in March is spoken of positively as "proof that the Silicon Valley side is climbing aboard on the same terms as SBG." On the other hand, SBG's committed amount (up to 64 billion dollars = roughly 9.6 trillion yen) far exceeds the individual investment amounts of other VCs, and the overwhelming view is that "even though they appear to be in the same boat, only SBG is standing at the prow." Silicon Valley's skepticism is directed not at "whether OpenAI will list and grow as expected," but rather at "whether SBG can endure the nearly two years until that listing while still carrying its leverage."

Persistent Negative Factors (1): Warnings from Ratings and Credit Markets

What tends to get overlooked while one is intoxicated by the opening figure of "over 5 trillion yen" is the cold gaze coming from the credit markets. In March 2026, following the announcement of an additional $30 billion investment in OpenAI, S&P Global Ratings revised its outlook on SBG's long-term issuer rating from "Stable" to "Negative." While the long-term rating itself was kept at BB+, in non-investment-grade territory, S&P explicitly stated that it "will move to a downgrade if the likelihood increases within the next 6 to 12 months." The "LTV (loan-to-value, the ratio of debt to assets on a market-value basis) of 25%" limit that SBG voluntarily imposes on itself is expected to reach 27–28% once the additional investment is completed, approaching a figure that risks breaching even its own discipline.

The credit markets are moving ahead of the rating agencies. SBG's 5-year CDS (credit default swap) spread has widened by more than 20% since the start of 2026, trading at a level that stands head and shoulders above other Japanese companies. This means investors are now paying a higher "insurance premium against the risk that SBG's debt will not be honored in the future," and it indicates that market participants, decoupling from the "dazzle of the numbers" in the earnings results, are steadily beginning to price in the deterioration of its financial health. According to analyst estimates, a refinancing demand (refinancing wall) of roughly $50 billion (approximately 7.5 trillion yen) looms before the end of 2026, and clearing this is the first hurdle.

Persistent Negative Factors (2): The $40 Billion Bridge Loan and "Margin Loan Reduction"

A symbol of concerns over its financial health is the unsecured bridge loan of roughly $40 billion (about ¥6 trillion) that SBG arranged in March 2026 from lenders including JPMorgan, Goldman Sachs, Mizuho Bank, Sumitomo Mitsui Banking Corporation, and MUFG Bank. The term is 12 months, and if SBG cannot repay it by March 2027, it will have to find an alternative means of financing or sell holdings to raise cash. Bloomberg writes that "the 12-month brevity reflects a funding scheme designed on the premise that OpenAI's IPO will be realized within that window." As of these earnings results, the outstanding balance is estimated at around $17.5 billion (about ¥2.6 trillion).

Furthermore, in early May, it was reported that SBG was moving to cut its target for arranging a margin loan—collateralized by OpenAI shares and sized at up to $10 billion (about ¥1.5 trillion)—by as much as 40%, down to roughly $6 billion (about ¥900 billion) at most. The reason is simple: the lending side—banks and hedge funds—has shifted to a cautious stance, judging that "the valuation is too high to accept unlisted OpenAI shares as collateral." To market participants, the gap—"even though OpenAI's valuation has risen this much, the size of the borrowing arranged against it as collateral has conversely shrunk"—appears to be evidence that SBG's largest fundraising network is creaking from within. Should the margin loan fail to come together as expected, multiple media outlets point out that SBG would be forced to choose to additionally sell or pledge NVIDIA shares, T-Mobile shares, SoftBank Corp. (9434) shares, PayPay shares, and even Arm shares—its second source of revenue.

Persistent Negative Factors (3): Internal Conflict and Executive Departures in the Stargate Project

The "Stargate Project" — a $500 billion (roughly ¥75 trillion) AI data center consortium formed by OpenAI, Oracle, SoftBank, and MGX — which Mr. Son announced with great fanfare alongside President Trump in January 2025, has already seen internal conflict surface, in contrast to the initial euphoria. According to reports by overseas specialist media such as Tom's Hardware and The Decoder published from February 2026 onward, the three companies have failed to bridge their differences over ownership, operating rights, and revenue distribution for the data centers, and site acquisition and construction schedules have fallen behind.

Although they ultimately reached an outward compromise — "SBG will own and develop the sites, OpenAI will lead the design of the facilities, and operating rights will be consolidated under OpenAI through long-term lease agreements" — this is an arrangement that can hardly be called advantageous for SBG, since SBG is saddled with the infrastructure investment while the speed of monetization depends on OpenAI's rollout of its services. Furthermore, winbuzzer reported that in April 2026, three senior executives on the OpenAI side who had been leading Stargate departed in quick succession, meaning that the on-the-ground execution structure is wavering just ahead of the IPO.

The amount SBG has already committed in connection with Stargate is said to total more than $100 billion (roughly ¥15 trillion) over five years. Recalculated on an annual basis, this requires an additional cash outflow of about $20 billion (roughly ¥3 trillion), a level that cannot be covered even by the roughly ¥1.04 trillion in annual operating profit of its core carrier business (9434). A delay in Stargate is not merely a talking point for the AI industry; it is a landmine that directly damages SBG's cash flow plan for the fiscal year ending March 2027 and beyond.

Persistent Negative Factors (4): Intensifying Competition for OpenAI and "Evidence of Stalling Revenue Growth"

To begin with, OpenAI itself—the very "object of the bet"—is starting to show visible cracks. According to a report by tradingkey, the month-over-month growth rate of OpenAI's weekly active users (WAU) sharply decelerated from 42% in January 2026 to just 13% as of September 2026. Not only has the company failed to present any outlook for turning a profit within 2026, but CFO Sarah Friar, in an interview in May 2026, stated that "2027 is the realistic timeframe for an IPO," effectively walking back the original scenario of a "Q4 2026 listing." A temporal mismatch with the deadline of SBG's $40 billion bridge loan (March 2027) is also gradually beginning to draw attention.

The competitive landscape is growing ever harsher. Google (Alphabet) with its Gemini 3.0 line, Anthropic (with Amazon as a major investor) with its Claude 5 line, and China's DeepSeek with boldly low-cost models are each closing in on OpenAI, and the stronghold where OpenAI can claim to be the "overwhelming No. 1" has come to lean heavily on the B2C brand power of products like ChatGPT and Sora. Furthermore, co-founder Elon Musk continues to pursue a damages lawsuit against OpenAI, and combined with the moves of the U.S. Department of Justice and the Federal Trade Commission (FTC), the legal uncertainty surrounding the IPO review process cannot be dispelled. OpenAI's valuation is premised on an extreme expansion of future cash flows, but the moment that premise collapses, a considerable portion of SBG's ¥5 trillion in book profit is structured to "reverse course" and come back as valuation losses.

Persistent negative factors (5): The "¥5 trillion in unrealized gains" is not cash

What decisively distinguishes SBG's profit from that of other operating companies is that the majority of it consists of "unrealized valuation gains based on IFRS fair value measurement." Unlike Toyota or Mitsubishi Heavy Industries, where the cash received from selling products becomes profit, this is merely a case of recording as profit an amount representing "this is how much we would supposedly earn if we sold," based on the market price at the time, simply because the price of OpenAI shares happened to be high at the end of the fiscal period. If OpenAI does not actually go public as expected and its valuation falls by even 30%, SBG will post a massive valuation loss in the following period. The market has not forgotten the "round-trip" history in which SBG, in the fiscal year ending March 2022—immediately after posting its then-record-high profit in the fiscal year ending March 2021—recorded a loss of 3.5 trillion yen (its largest ever) at the Vision Fund.

In addition, the roughly $46 billion in investment gains that SBG posted this period may look glamorous as net income under IFRS, but it naturally does not turn into a positive on a cash flow basis. FY2025 financial costs (interest expenses) rose more than 1.5-fold to 229.4 billion yen (a sharp jump from 148.9 billion yen in the same period the previous year), and the price of expanded leverage is steadily swelling in a corner of the financial statements. Rather than "being able to make the next investment with the money earned," the flip side of the record-high profit is "dressing up appearances with unrealized valuation gains while covering cash with debt."

Future developments, and when and what will be newly announced

There are broadly four notable upcoming events.

First, OpenAI's formal IPO filing. OpenAI's CFO, Sarah Friar, has signaled an implementation in 2027, and there are observations that the assumed market capitalization at the time of listing could reach up to $1 trillion (roughly ¥150 trillion). Since the deadline for SBG's $40 billion bridge loan is March 2027, whether a "filing submission" (S-1 filing) emerges within 2026 will be the immediate milestone.

Second, the IPO of "Roze (tentative name)," which is drawing attention as a strategic subsidiary of SBG. This is a new company that bundles the robotics business acquired from ABB (acquisition amount approximately $5.375 billion = roughly ¥806 billion, with closing expected in the mid-to-late part of 2026) together with physical AI-related assets under its umbrella such as SoftBank Robotics, Berkshire Grey, AutoStore, Agile Robots, and Skild AI. The Financial Times reported in April 2026 that Mr. Son is aiming for a valuation of up to $100 billion (roughly ¥15 trillion). The stance is to target a U.S. IPO as early as within the year, or by 2027 at the latest.

Third, the medium-term management plan for fiscal 2026 (the term ending March 2027). SBG itself continues its traditional stance that "as an investment company, it does not disclose earnings forecasts," and even in the May 13 announcement no outlook for the current term was presented. Meanwhile, the subsidiary SoftBank Corp. (9434) released a new medium-term management plan, "Activate AI for Society," on May 11, setting the start of operations for its AI data centers in Sakai City and Tomakomai City (Sakai will be equipped with 140MW, 110 exaFLOPS, equivalent to roughly 100,000 of NVIDIA's latest GPUs) for fiscal 2027. It places its fiscal 2030 consolidated operating profit target at ¥1.7 trillion. For SBG, the extent to which it can incorporate this subsidiary's AI infrastructure business into its consolidated value will be the main battleground from the next earnings cycle (the first quarter in August) onward.

Fourth, Stargate's announcement of additional sites and concession negotiations. The three parties—OpenAI, SoftBank, and Oracle—announced in September 2025 the addition of five sites within the United States (totaling a scale of 7 gigawatts together with Abilene, Milam County in Texas, and others, exceeding $400 billion over three years), but converting the procurement commitment into real demand requires the conclusion of contracts with each site's power company and government by the latter half of 2026. If delays or cost overruns occur here, they will materialize in the earnings results as asset impairments or provisions.

In the near term, the FY2026 first-quarter results scheduled to be announced in August 2026 and the second-quarter results scheduled for announcement in November will be the next touchstones. Rather than basking in the afterglow of "record-high profits," investors will, over the coming one to two quarters, rigorously question "when SBG will present a concrete cash-generation scenario for repaying the bridge loan," "when Stargate's full-scale operation will return as recognized profit," and "whether Arm's stock price can capture data center demand and continue rising." Behind the superficial glamour of surpassing ¥5 trillion, the accurate reading would be to regard SBG's true moment of truth as beginning, rather, from here.

Sources