Berkshire and SoftBank, one will "definitely die."

Author: Le Ming

On May 2, 2026, in Omaha. Out of eighteen thousand seats, only a little more than half were filled—previously, tickets for the Berkshire shareholder meeting were highly sought after, and hotels outside the venue were fully booked.

This time, 95-year-old Buffett did not host the event as he had in previous years; the new CEO Greg Abel stood in front of the main screen, answering investors' questions about why Berkshire held nearly $400 billion in cash.

In the same week, six thousand miles away in Tokyo, Masayoshi Son's team was doing something else: packaging SoftBank Group's unprofitable AI assets to prepare for a new company called Roze AI, with a target valuation of $100 billion, planning to go public in the second half of 2026 in the United States.

The reason is simple: SoftBank must continue to find money to fund the $64.6 billion check for OpenAI, which could eventually roll up to nearly $100 billion.

One holds $397.4 billion in cash, buys nothing, and waits for the market to crash; the other bears a parent company debt of 163.4 trillion yen (over $10 billion), betting that the market will not crash.

Berkshire: Too Much Cash is a Problem

What does $397.4 billion mean?

It accounts for nearly 40% of Berkshire's total market value and is twice the average cash level Buffett has maintained over the past twenty years.

Of this amount, $339.3 billion is directly held in U.S. short-term Treasury bonds, making Berkshire one of the largest non-government creditors of the U.S. Treasury.

This pile of cash was not passively accumulated; it was actively built up.

Over the past fourteen quarters, Berkshire has been a net seller of stocks each quarter. Apple, once its largest holding, has been continuously reduced for four consecutive quarters since the third quarter of 2024, with nearly 688 million shares sold, realizing over $100 billion.

Buffett's explanation has always been the same: he cannot find cheap things.

In his 2024 letter to shareholders, he wrote, "Generally speaking, nothing looks attractive." In a recent off-site interview at the shareholder meeting, he likened the current market to "a church connected to a casino," saying that among all the market sentiments he has experienced, this is the most akin to gambling.

The problem is that he has been making this judgment for more than a year.

Berkshire's stock performance over the past twelve months has lagged the S&P 500 by about 40 percentage points. This is not a small number—this is one of the largest relative drawdowns since Buffett took over Berkshire in 1965. The last time such a degree of underperformance occurred was in the final stages of the internet bubble in 1999.

At that time, Buffett said he would only buy "cutting-edge industries like bricks, carpets, insulation materials, and paint." Two years later, the Nasdaq fell 78%, and he was proven right.

But this time, investors have been waiting for more than two years. The market rises, and Berkshire remains still; the market rises again, and Berkshire still does not move. On the day Buffett officially stepped down as CEO on January 1 of this year, Berkshire's stock price slightly declined— the market expressed a restrained disappointment in the most restrained way.

This is the situation Abel faced when he took over. He is a Canadian accountant who rose to become Berkshire's vice chairman, having spent his life in heavy asset, regulated, slow businesses like utilities, railroads, and energy.

He is not Buffett; he knows he is not Buffett. In his first letter to shareholders, he repeatedly emphasized "continuity" and "decentralization." At the first shareholder meeting, his absolute answer to all suggestions about splitting the group was "impossible."

Abel's dilemma is that he cannot use this cash (because the market is too expensive), nor can he continue to pretend this money does not exist (because investors are voting with their feet).

If the market continues to rise over the next five to ten years, he will eventually have to face a question that has never been seriously discussed in Berkshire's history—what to do with the money. Either return it to shareholders in the form of special dividends or actually break apart and sell this monster stitched together by more than 60 subsidiaries.

Will Berkshire die? No, it will not die suddenly. Its assets are too diversified, its cash too abundant, and its debts too low; any external shock will be hard to truly penetrate it. But it will slowly and decently transform into something else.

SoftBank's Problem: Too Little Money and Still Betting

Masayoshi Son's situation is the opposite mirror of Abel's.

On February 27, 2026, SoftBank issued an announcement. The most critical sentence translates to: "SoftBank Group's cumulative investment in OpenAI is expected to reach $64.6 billion, with a shareholding ratio of about 13%."

$64.6 billion, 13%. This is the most expensive single bet of this era.

To understand the madness of this number, one needs to see how SoftBank can afford this amount.

The parent company's interest-bearing debt skyrocketed from 121.4 trillion yen in March 2025 to 163.4 trillion yen in December 2025. The so-called cash at the parent company is about 38 trillion yen, of which nearly one-third is actually unused committed credit lines, not real cash.

Where did this money come from? SoftBank borrowed $20 billion against its holdings of Arm stock; it also raised about $7.7 billion from pledging shares of its Japanese telecom subsidiary SBKK.

On March 27, 2026, SoftBank signed an unprecedented $40 billion bridge loan, led by JPMorgan, Goldman Sachs, Mizuho, Sumitomo Mitsui, and Mitsubishi UFJ, later expanded to eight banks—this is one of the largest bridge loans in Asian history. Of this money, $30 billion is directly used to co-invest in OpenAI. The term is 12 months, meaning that by March 2027, SoftBank must repay $40 billion.

This is why Son seems a bit "abnormal" this year: he cleared all of SoftBank's Nvidia shares, cashing out $5.8 billion in a one-time sale in October 2025. He admitted during a speech in Tokyo in early December 2025, "I didn't want to sell a single share of Nvidia, but I needed more money to invest in OpenAI. I cried while selling Nvidia."

Son is "selling everything" to raise money for OpenAI: SoftBank also sold its holdings in T-Mobile—selling 56.9 million shares in the first three quarters of the 2025 fiscal year, cashing out $12.7 billion; in the fourth quarter, another 12.5 million shares were sold for $2.3 billion. Deutsche Telekom was also completely sold off, and Alibaba was also fully divested. At the end of April this year, SoftBank began to arrange a $10 billion margin loan secured by OpenAI equity, with an interest rate of nearly 8%.

At the same time, SoftBank is also issuing bonds everywhere: in November 2025, SoftBank issued a bond of 500 billion yen with a coupon of 3.98%; in April 2026, it issued another subordinated bond of 418 billion yen with a coupon of 4.97% for the first five years—this is SoftBank's most expensive retail bond and the highest coupon in the history of Japanese non-financial corporate yen retail bonds—indicating that the market has begun to have "doubts" about SoftBank's debt.

The response from the credit market was direct: SoftBank's five-year credit default swaps surged to 355 basis points in early March, hitting an 11-month high.

Son's recent "savior" is hoping that OpenAI can go public as soon as possible; otherwise, if the debt pressure becomes long-term, it could really trigger a crisis for SoftBank.

However, although OpenAI's CEO Sam Altman advocates for an IPO in the fourth quarter of 2026, CFO Sarah Friar suggests delaying it until 2027—this public split between the CEO and CFO itself tells the market: this company is internally uncertain about whether it is ready.

"Inevitable Death"

Berkshire's death will be gentle.

It will not go bankrupt—its subsidiaries are all high-quality cash cows, and its debt level is extremely low. Even if the AI capital expenditure bubble bursts, even if data center demand halves, even if the S&P 500 falls by 50%, Berkshire's cash pile is enough for it to devour any cheap assets over the next decade.

Its death is a death of identity—the compounding myth of Buffett's "buying good businesses at bargain prices" may no longer hold in a world where valuations are always above a 30 times price-to-earnings ratio and where ten-year Treasury yields have shattered all traditional valuation models.

Abel may execute well as a "rational CEO"—continuing operations, continuing small buybacks, and continuing to do some mergers and acquisitions on the margins—but Berkshire, as a narrative of capitalist discipline, died the moment Buffett stopped writing letters. Its shell remains, but its soul has exited.

SoftBank's death, on the other hand, may be violent. It has three death triggers, and any one of them being pulled could trigger a chain reaction:

The first trigger is OpenAI. If its IPO is delayed until 2027 or even 2028, if Amazon's $35 billion commitment linked to the IPO ultimately cannot be fulfilled, if OpenAI's revenue growth stalls in any quarter—even just one quarter—the valuation of SoftBank's 13% stake in OpenAI will be reassessed downward.

The second trigger is Arm. Arm is currently SoftBank's only truly liquid asset that is still highly priced by the market—valued at about $200 billion, with SoftBank holding 87%.

Arm's royalty revenue increased by 26% year-on-year in the third quarter of the 2026 fiscal year, and data center-related royalties doubled, making it one of the pillars supporting SoftBank's entire valuation story.

But Arm is also the most easily repriced target at present—once it returns from its current 70 times forward price-to-earnings ratio to a "normal" semiconductor company valuation, the coverage ratio of SoftBank's $20 billion Arm mortgage loan will collapse.

The third trigger is refinancing itself. The $40 billion bridge loan matures in March 2027, and before that, SoftBank must achieve at least one of the following: OpenAI goes public, Roze AI goes public, sell another batch of assets, or issue a similarly sized bond to roll it over.

But each path is more expensive than the previous year—SoftBank's retail bond coupon has risen from 3.98% in 2025 to 4.97% in 2026.

Looking at each of these three triggers individually, the probability does not seem high—OpenAI is likely to go public, Arm is unlikely to collapse immediately, and the credit market is unlikely to close its doors to SoftBank overnight. But they have a characteristic that cannot be ignored: they are highly correlated, rather than three independent events.

If the bubble does not burst, Son will ultimately be deified: Roze AI goes public with a $100 billion valuation, OpenAI successfully IPOs, and he will realize his narrative of ASI (Artificial Super Intelligence) at the age of 70.

Meanwhile, Berkshire will be gently, continuously, and irreversibly marginalized by the market under Abel's steady management—until one day, some successor has to do what Buffett refused to do his entire life.