SoftBank Turns to Japan's Retail Savers as Its AI Ambitions Outrun the Balance Sheet

SoftBank issues record ¥1 trillion retail bonds to fund OpenAI and Stargate, amid rising global yields and volatile stock performance.

The math behind Masayoshi Son's artificial-intelligence empire is getting harder to ignore. SoftBank Group is preparing to sell one trillion yen — roughly $6.3 billion — of bonds aimed squarely at Japanese individual investors, a record for the conglomerate and nearly double the 600 billion yen it raised in a similar deal just five months ago.

The seven-year notes are expected to carry a coupon between 4.3 and 4.9 percent, with pricing set for September 4 and a subscription window running from September 7 through September 16. The structure is telling: rather than lean further on bank loans or sell down stakes in listed assets, SoftBank is tapping the savings of ordinary Japanese households to help fund its deepening commitment to OpenAI.

That commitment is substantial. Cumulative investments in or pledges to OpenAI are reported at between $34.6 billion and $64.6 billion, alongside $19 billion directed at the Stargate data-center project. A newly formed battery joint venture with South Korea's Cosmos Lab and DeltaX, based in Osaka, targets production by the fiscal year ending March 2028 and mass manufacturing the following year, with a revenue goal of more than 100 billion yen — about $6.38 billion — by fiscal 2030. A separate entity, Roze AI, has also been established for AI infrastructure work.

The sheer breadth of these outlays explains why the bond market has become central to SoftBank's funding strategy. But it also exposes the group to a global rate environment that is turning less forgiving. Thirty-year US Treasury yields have been hovering near levels last seen in 2007, British gilts are trading at 1998-style yields, and Japan's own long-dated government bonds are approaching historic highs. Tokyo's budget for fiscal 2027 already pencils in record interest costs of 16.6 trillion yen following the Bank of Japan's rate rise to 1.0 percent — the highest in 31 years.

Investors have taken notice. The stock has been volatile, with an annualized volatility reading of 75 percent that underscores just how jittery trading in the name has become. In Tokyo on Tuesday, shares climbed more than 2 percent to around 5,087 yen, helped by a broader recovery in Japanese and South Korean indices after early losses, as US yields eased and the dollar softened. In German trading, the stock was up 2.8 percent on the day at 27.70 euros — though the weekly picture remains down 8.2 percent, and the 30-day stretch shows a 4.8 percent decline.

The contrast between the parent company and its telecom subsidiary is becoming one of the more striking features of the group's profile. Fitch affirmed SoftBank Corp.'s rating at BBB+ with a stable outlook on Monday, citing a strong mobile position with a 25 percent market share and synergies with LY Corp. The agency projects EBITDA growth of 7 percent CAGR for the telecom arm between fiscal 2027 and 2030, with net leverage expected to improve from 2.6–2.7 to 2.4–2.5. That is steady, income-generating reality — a world away from the venture-capital intensity of the holding company above it.

The immediate test, though, comes from the semiconductor complex. Nvidia's quarterly results are due Wednesday, an event that tends to set the tone for AI-adjacent names across Asia. The Philadelphia Semiconductor Index fell 2.7 percent on Monday, and should Nvidia disappoint, Japanese tech names like SoftBank, Advantest, Kioxia and Tokyo Electron could face renewed selling pressure.

The September retail bond will therefore serve as something of a referendum on how much confidence Japanese savers still place in SoftBank's AI strategy. Investment banks including Goldman Sachs expect AI-related bond issuance to accelerate further in the coming years, even as borrowing costs climb. For SoftBank, the balancing act is acute: it must raise fresh capital for its OpenAI ambitions at precisely the moment when debt financing becomes more expensive — and when the market is demanding evidence that the billions being committed will eventually translate into earnings, not just headlines.