SoftBank CDS rises to 432 basis points, Masayoshi Son continues to increase AI investment

On October 20, according to foreign media reports, the price of SoftBank's 5-year credit default swaps (CDS) rose to 432 basis points, hitting the highest level since November 2022.

CDS can be simply understood as default insurance for debt—the higher the price, the greater the cost investors must pay to guard against default risk.

Today, the market's concern about SoftBank's credit risk has reached its highest level in nearly four years.

Yet Masayoshi Son has not stopped spending.

Earlier this month, SoftBank had just completed a new $10 billion investment in OpenAI, bringing its cumulative investment to $64.6 billion. Not long ago, a batch of new bonds it issued carried a maximum coupon of 9.75%.

In addition, according to foreign media reports, OpenAI is striving to complete a new round of financing before Anthropic's expected mid-November listing, but because some investors are hesitant about the high valuation, the target valuation for this round has been cut from $1.4 trillion to $1.25 trillion, a shrinkage of $150 billion. SoftBank, as an existing investor, was previously reported to be in talks to continue participating.

And SoftBank's stock price has already fallen by more than half from its June high.

A

Among major Japanese companies, SoftBank's default insurance price now ranks first, more than 130 basis points higher than second-ranked Nissan Motor.

Some analysts who have long followed SoftBank have expressed concern about Masayoshi Son's investment strategy.

On October 14, MST Financial analyst David Gibson, in an interview with the media, directly compared SoftBank to a game of roulette. He said that SoftBank used to bet on many numbers on the roulette table. Now, it only bets on red or black, either losing everything or winning everything. In Gibson's view, SoftBank is concentrating more and more bets on a small number of AI assets.

For investors who truly need to consider whether to buy SoftBank bonds, such risk is not easy to accept.

James Athey, a fixed income portfolio manager at British asset management company Marlborough, said that SoftBank's investment strategy is too aggressive and too concentrated, with almost all bets placed. Even when considering investing in high-yield bonds, he would not choose SoftBank first.

Another analyst, Amir Anvarzadeh, believes that Masayoshi Son's bet this time is bigger than ever before, and directly advised investors to short SoftBank stock.

However, Masayoshi Son should not be unfamiliar with such a situation.

Four years ago, SoftBank also experienced a similar crisis.

In 2022, global tech stocks suffered a sell-off, the valuations of a batch of tech companies in which Masayoshi Son held heavy positions shrank sharply, and the Vision Fund posted huge losses one after another. Financial reports released in November of that year showed that the Vision Fund lost about $7.2 billion in the July-to-September quarter, while in the previous quarter it had just lost about $17 billion.

Losing more than $24 billion in just half a year would make anyone sweat.

Masayoshi Son, who had once been searching everywhere for investment opportunities, had to rein in his offensive posture. SoftBank sharply cut new investments, sold a large number of Alibaba shares to improve its financial condition, and launched layoffs in September of that year, cutting about 150 people across the Vision Fund and the group's international business.

By the earnings press conference on November 11, Masayoshi Son announced that for a period going forward, he would no longer attend SoftBank's quarterly earnings briefings and would devote more energy to its chip design company Arm.

Masayoshi Son once said that he is naturally a person who likes to attack. But that time, he chose to retreat.

SoftBank began to preserve cash, reduce investments, and wait for the market to improve. It was not until 2023 that Masayoshi Son gradually regained his offensive rhythm.

Now, four years later, SoftBank's credit risk indicator has once again returned to a similar level.

Only this time, Masayoshi Son seems to have no intention of stopping.

B

In February this year, SoftBank announced an additional $30 billion investment in OpenAI, planned to be paid in three installments, completed in April, July, and October respectively. Together with previous investments, SoftBank has invested $64.6 billion (about 431.5 billion yuan) in OpenAI and holds about a 13% stake.

Of course, such a large amount of money could not all come from cash on hand.

In March, SoftBank signed a $40 billion bridge loan agreement with banks, mainly to support investment in OpenAI—a so-called bridge loan is to first borrow short-term funds for emergency use and later repay them through other financing methods.

SoftBank successively drew $30 billion from it. On September 15 this year, it repaid the remaining $25.9 billion of borrowing early, and at the end of the month it canceled the unused $10 billion facility. At this point, the $40 billion bridge financing agreement has been settled.

But repaying the bridge loan does not mean SoftBank has shaken off the corresponding financing burden; part of the funding need was resolved through long-term bonds, other financing, and asset monetization.

On September 24, SoftBank announced the issuance of about $11.1 billion in dollar- and euro-denominated bonds, among which a $4.5 billion dollar bond maturing in 2034 carried an annual coupon of as high as 9.75%.

This is the highest coupon on bonds ever issued by SoftBank. By comparison, a bond maturing in 2033 issued by Google parent Alphabet in August this year carried a coupon of only 5.2%.

Wall Street has already begun to demand that Masayoshi Son pay a higher price for his bold bet.

But SoftBank management does not seem to think this money is expensive to borrow. On October 14, Alex Clavel, head of SoftBank Vision Fund Management, explained in an interview with the media why SoftBank would rather bear high interest than sell the assets it holds.

He said: "We borrow money largely because we don't want to sell those assets that we believe will continue to appreciate."

In the view of SoftBank management, OpenAI's future upside is large enough that even if the cost of borrowing money is now close to 10%, it is still worth continuing to hold.

Moreover, Masayoshi Son is not satisfied with the $64.6 billion already invested. According to foreign media reports, OpenAI is accelerating the raising of a new round of financing, and SoftBank is also in talks to participate. It is not yet clear how much SoftBank plans to invest, but if it continues to follow on, it will be another new funding requirement.

In addition to OpenAI, SoftBank has also spent no small amount elsewhere.

At the end of September, it had just completed the acquisition of data center investment management company DigitalBridge for about $3.1 billion. Previously, SoftBank had agreed to acquire ABB's robotics business for about $5.375 billion, with the deal planned to be completed in the second half of this year.

Recently, Masayoshi Son has set his sights on the US security giant Allied Universal. According to foreign media reports, SoftBank is in preliminary talks to acquire the company, with a potential equity valuation of up to about $18 billion, and if debt is included, the total transaction value could reach $30 billion.

OpenAI is still raising funds, the data center and robotics businesses are still expanding, and new acquisition plans have emerged again.

However, at least judging from previously disclosed financial data, SoftBank still has some financial room and has not yet reached the end of its rope.

As of the end of June this year, the investment assets held by SoftBank were worth about 83.11 trillion yen, and net debt calculated on a group basis was 10.81 trillion yen, corresponding to an LTV (loan-to-value ratio) of 13%.

This indicator measures the ratio of net debt to the value of held assets. SoftBank's management standard set for itself is no more than 25% under normal market conditions, and even under extreme circumstances, it should be kept within 35%.

13% still leaves some room from the 25% upper limit.

But these are all data as of the end of June. Since then, SoftBank has continued to complete huge investments and acquisitions, and the market prices of held assets have also fluctuated significantly. Especially for unlisted companies like OpenAI, even if equity valuations continue to rise, it does not mean SoftBank can turn them into cash at any time.

Sharon Chen, a credit analyst at Bloomberg Intelligence, expects that in the fiscal year ending March 2027, SoftBank's net cash outflow will reach at least $6 billion. If it continues to expand AI investment or completes the Allied Universal acquisition, SoftBank will need to raise more funds, and LTV may also gradually approach the 25% management upper limit.

Masayoshi Son does indeed have a large number of valuable assets, but valuable assets ultimately cannot be directly used to pay bond interest. Moreover, the prices of these assets are not stable.

If core investments such as OpenAI continue to appreciate, SoftBank can use the increasingly large asset value to support the next round of financing and investment.

C

SoftBank's creditors are not the only group feeling nervous.

On October 20, US time, according to foreign media reports, Wall Street is setting off a wave of buying default insurance for AI-related companies.

According to Morgan Stanley statistics, in July this year, the average weekly trading volume of CDS for companies such as Oracle, SpaceX, Alphabet, Amazon, Meta, Broadcom and Nvidia was less than $2 billion; by September, this figure had exceeded $9 billion.

At the same time, the AI industry is borrowing more and more money. According to statistics, since 2026, AI-related companies have raised about $466 billion in total through bond issuance, private credit and other methods, while in the same period last year it was only $101 billion.

These investors buying default insurance are not all doing so to protect the bonds they hold. Some equity funds and multi-asset funds have begun using CDS to hedge against the risk of AI stocks falling.

This concern does have a realistic basis. On the same day, The Wall Street Journal disclosed Anthropic's latest draft IPO prospectus.

In the first half of this year, Anthropic recorded a book net loss of $80.7 billion, a sharp expansion from $13 billion in the same period last year. However, $72 billion of that came from accounting expenses generated by changes in the value of financing instruments such as convertible notes, and does not mean the company actually burned that much cash.

More noteworthy is computing power spending. Anthropic spent nearly $9 billion on computing power and infrastructure in the first half of this year, already exceeding last year's full-year $7.3 billion. According to the draft prospectus, including transactions still under negotiation, computing power spending over the next several years may reach $750 billion, of which more than $600 billion involves signed agreements.

Although Anthropic's revenue in the first half of the year has grown to $16.3 billion, such a huge long-term computing power commitment will still test its future cash flow.

Take CoreWeave, a cloud computing company invested in by Nvidia, as an example. It borrows heavily to buy GPUs, build data centers, and then rents computing power to customers such as Microsoft.

As market concerns about AI debt risk heat up, CoreWeave's stock price continued to plunge in October. As of October 23, the cumulative decline was close to 40%. The credit market is also flashing warnings: its 5-year CDS quote exceeded 1,100 basis points, and some bond yields rose to around 16%.

AI companies need computing power, data centers need funds, and chip companies need these customers to keep buying GPUs.

Several companies appear to be doing their own businesses, but the funding needs behind them are becoming more and more closely linked.

On October 19, Mark Habib, head of digital infrastructure at S&P Global, also issued a warning in an interview with the media: the AI-related debts and obligations borne by large technology companies may be more than the outside world knows.

Some data center projects raise funds through complex financing structures, while large technology companies provide guarantees for related projects or enterprises. S&P is concerned that some risks are not fully reflected in these companies' public financial data.

However, S&P has not currently required a broad downgrade of the credit ratings of technology giants because of this.

What is more worrying is the reliance of these financing arrangements on future revenue growth.

SoftBank borrows money to invest in OpenAI; model companies such as OpenAI and Anthropic lock in computing power for the next several years in advance; cloud computing and data center companies then borrow money to buy chips and build infrastructure. These companies' debts are not borne by the same entity, but their investment returns all depend to varying degrees on the AI market being able to continue growing.

If future revenue falls short of expectations, model companies may find it difficult to bear the huge computing power contracts they previously signed and be forced to cut back on new procurement and investment; the financing capacity of data center operators may also be affected, and investment institutions may face the dual pressure of asset depreciation and rising financing costs.

The troubles encountered by one company may gradually be transmitted to other companies along funding and contractual relationships.

What Masayoshi Son is facing is precisely one of the difficult problems in this high-stakes gamble.

He believes that OpenAI will continue to appreciate and that AI will ultimately be able to create wealth far exceeding today's investment.

This article comes from the WeChat public account "Zimubang" (ID: wujicaijing), author: Yuan Xinyue, published by 36Kr with authorization.