SoftBank pays up to 9.75% on a record $11 billion junk sale to avoid selling assets it expects to rise

The comparison Reuters prints is the useful one. SoftBank's 9.75% coupon is on a bond maturing in 2034 ^[2], and Alphabet's 5.2% is on a bond issued in August that matures in 2033 ^[3]. The gap is 4.55 points ^[20], or rather it is 4.55 points on paper, because the two bonds differ by a year of maturity. Reuters says SoftBank's higher yields reflect investor perceptions of risk and concentration in large positions ^[19]. Reuters does not give yields for the Amazon.com and Tencent sales it lists beside Alphabet's ^[10], so Alphabet is the only peer price in the report and nothing in it shows AI borrowers as a group moving into junk.

The two skeptics Reuters quotes by name are talking about SoftBank's size. "SoftBank is so aggressive, so concentrated and so all-in, I don't think we'd be looking there first anyway," said James Athey, a fixed-income portfolio manager at Marlborough ^[15]. "The bets are much bigger than we've ever seen," said Amir Anvarzadeh of Asymmetric Advisors, who recommends shorting the stock ^[16].

Credit has moved less than equity. SoftBank's shares are down 38% since their June record ^[8], while its credit default swap spreads, the cost of insuring its debt, are up 24% over the same period ^[9]. A price fall and a spread rise are different measures, so the two percentages do not compare cleanly. In our view the ordering still counts: shareholders have absorbed more of the repricing than the people insuring the debt.

Management's defence is that the price is small next to the value. "We've borrowed money in large part because we don't want to sell assets that we believe will keep rising in value," said Alex Clavel, CEO of SoftBank Investment Advisers ^[5]. "Given OpenAI's appreciation to date, 9.75 per cent is still very reasonable," Clavel said ^[6]. The coupon is cash due on a schedule. The appreciation is a mark on a private company, and with Altman saying OpenAI will not go public this year ^[12], no public price will test it in that time. Son's record on such marks is mixed: Alibaba was a major success, and he also directed $16 billion to WeWork, which went bankrupt ^[17].

SoftBank is paying interest where it once sold stock ^[4]. We think the 9.75% prices SoftBank's concentration and not yet the sector, and we can see three ways to be wrong. If SB Energy lists near the $50 billion Reuters says it may seek ^[13] and Arm's gains hold ^[11], the coupon will look cheap and the concentration premium too harsh. If another AI borrower with a sturdier balance sheet has to pay junk-bond yields, SoftBank is simply the first. If the CDS spread keeps rising while the shares steady, credit buyers are doubting repayment.