OpenAI Is SoftBank’s Investment, Customer and Collateral. What Could Go Wrong?

Masayoshi Son’s biggest investment in AI is also becoming the customer for his biggest AI infrastructure project.

This investment is helping SoftBank report huge gains. The shares have helped the company raise new money. And the customer contracts are helping it justify spending billions of dollars on data centres and electricity.

The same private company is at the centre of all these deals: OpenAI.

SoftBank has invested a total of $64.6 billion in the company that makes ChatGPT. It owns about 13 per cent of OpenAI. At the end of September, before it completed the final $10 billion part of its investment, SoftBank valued its $54.6 billion stake at $99.6 billion. This meant it had an estimated gain of about $45 billion that it had not yet turned into cash.

SoftBank has also raised $10 billion through what it calls a “loan using OpenAI shares”. But there is an important detail. The shares have not been officially offered as security for the loan. Instead, the borrower’s cash account is listed as security. However, if the value of the OpenAI preferred shares mentioned in the agreement falls sharply, SoftBank may have to add more cash as security or repay the loan early.

OpenAI is also the customer under long-term rental agreements for data centres backed by SoftBank in Texas and Ohio. At the PORTS-Pike campus in Ohio, SB Energy is expected to build, own and run an eight-gigawatt AI complex. OpenAI will rent it under a 20-year agreement.

This is more than simply putting a large amount of money into a promising startup. OpenAI has become an investment, a source of reported gains, a reference for borrowing and a customer for SoftBank’s infrastructure projects. All of these roles are part of the same business strategy.

When all these parts grow together, the arrangement can be very powerful. But if the main assumptions turn out to be wrong, problems in one part could spread to other parts of SoftBank.

How large is SoftBank’s OpenAI investment?

SoftBank completed three additional investments of $10 billion each in April, July and October 2026. It bought the shares at a company valuation of $730 billion before the new money was added. This brought its total investment to $64.6 billion.

That is an enormous amount of money to put into a single private company. It is more than the yearly economic output of several countries. It is also a big change from the original idea behind the Vision Fund, which was to spread investments across hundreds of technology companies.

The size of the investment is even more striking when compared with Vision Fund 2. At the end of September, the fund reported total investment returns of $144.4 billion against an original investment cost of $122.6 billion. OpenAI alone accounted for $99.6 billion of the reported value of those returns.

At the end of September, Vision Fund 2 reported total investment returns of $144.4 billion from an original investment cost of $122.6 billion. OpenAI alone accounted for $99.6 billion of that return value Credits: Getty

SoftBank’s own figures show how much Vision Fund 2’s reported recovery depends on the value assigned to OpenAI. During the reported six-month period, OpenAI’s value stayed the same as it was at the end of the previous financial year. This means the $45 billion gain built up over time, rather than being created in the latest quarter. It was also still an estimated gain, not money SoftBank had received.

This difference is important. A private company can become more valuable on paper without producing cash for its investors. SoftBank must sell some shares, wait for OpenAI to list on the stock market or find another way to turn its investment into money it can use.

It chose the third option.

Are OpenAI shares really security for SoftBank’s loan?

In practical terms, yes, although the legal arrangement is more complicated than the headline suggests.

In August, a subsidiary of Vision Fund 2 borrowed $10 billion. The money was intended for the general business needs of SoftBank and the fund. SoftBank Group guaranteed the loan, which must be repaid in August 2028.

The OpenAI shares have not been formally pledged to the lenders. Instead, the loan is backed by a cash account owned by the borrower. However, the agreement includes rules that are directly linked to the market value of OpenAI’s preferred shares. If that value falls sharply, the lenders can demand more cash security or require the loan to be repaid early.

This means OpenAI’s value is more than just a number in SoftBank’s investment records. It can affect how much cash SoftBank must set aside to support its $10 billion loan and whether the loan can remain unpaid until its due date.

This is the main contradiction in the strategy. SoftBank can use the rising value of an investment that it cannot easily sell to raise cash without giving up its shares. This allows it to keep the potential gains that Son expects from OpenAI. But it also means the loan depends on the same company value that SoftBank wants to see rise in the future.

This is not exactly the same as a normal margin loan. In a margin loan, shares are officially pledged and the lender may take or sell them if their value falls. But the financial risk is similar. If OpenAI’s value falls far enough, SoftBank may have to provide more cash or repay its debt at the same time that the value of its investment is also falling.

How did OpenAI become a customer of SoftBank?

Son does not just want to own part of the AI technology. He also wants SoftBank to help provide the electricity, land and buildings needed to run it.

SB Energy’s planned PORTS-Pike technology campus in Ohio is the clearest example. OpenAI has signed 17 rental agreements covering about eight gigawatts of computing capacity. Each agreement has an initial term of 20 years. Rent will begin in stages as the required buildings, power sources and connections to the electricity grid are completed.

OpenAI says it will start paying rent only when the completed computing capacity is ready to use. It expects to pay for these commitments using its business revenue, cash flow and money raised from investors. The first 800 megawatts are expected to be ready in 2028. The rest of the project is planned to be completed in stages.

SoftBank is no longer just investing in companies that might benefit from AI. It is organising money, electricity, property and computing capacity around the future of one company in which it is already the biggest financial supporter Credits: Getty

Nvidia adds another part to the arrangement. It is the only provider of computing infrastructure for the project and has agreed to invest $1.5 billion in SB Energy. It is also providing financial support for the land, electricity and basic buildings needed for the first 4.25 gigawatts of capacity.

According to a regulatory filing by SB Energy, Nvidia has also agreed to guarantee some of the remaining value linked to OpenAI’s rental payments for this first 4.25 gigawatts. Under certain conditions, the total guaranteed value can be as high as $105 billion. This protection does not apply in every situation. It can come into effect if, for example, the tenant becomes unable to pay its debts or fails to make a payment and does not fix the problem. The guaranteed value also falls over the life of the rental agreements.

This arrangement shows how money is moving through the AI industry. SoftBank provides investment money and develops the infrastructure. OpenAI promises long-term demand for that infrastructure. Nvidia supplies the chips and some financial protection. Each company helps make the business case for the others stronger.

Where does this arrangement become risky?

The first risk is the value of OpenAI.

OpenAI must prove that its enormous private-market valuation is justified while spending heavily on computing capacity. If its valuation falls, SoftBank’s total asset value will also fall. The drop could also activate the protections in the $10 billion loan agreement. It would also reduce the paper gain that has greatly improved Vision Fund 2’s reported results.

The second risk is relying too heavily on one customer.

OpenAI is expected to rent space at both PORTS-Pike and SB Energy’s Milam County project in Texas. Long-term rental agreements make it easier for these projects to get financing. But they also mean that a great deal depends on one customer being able to meet very large payment commitments over many years.

Nvidia’s financial support reduces some of the risk at PORTS-Pike. But the project still needs permits, electricity transmission, power generation, project financing and completed buildings before rent can begin.

The third risk is whether the projects can be completed as planned. SoftBank’s latest financial results already show a warning sign.

A SoftBank subsidiary paid about $9.48 billion to reserve manufacturing capacity and power-generation equipment for data-centre projects in the United States. Later, the project requirements changed. SoftBank then recorded a loss of $3.73 billion, including money paid in advance that could not be recovered and costs from changing contracts. At the end of September, about $6.6 billion in related advance payments was still listed on its balance sheet.

This loss does not prove that SoftBank’s data-centre strategy has failed. But it shows that, in the rush to secure equipment that is in short supply, SoftBank committed money before all the long-term project details had been settled.

The fourth risk is the cost of borrowing money. During the quarter, SoftBank raised a record $56.5 billion through loans, bond sales and the sale of assets. Its interest and other financing costs more than doubled year on year in the first half of the financial year. The group’s total debt that carries interest also increased by about ¥7.9 trillion in just six months.

SoftBank says its loan-to-value ratio is 15.9 per cent, well below its internal limit of 25 per cent. It also has ¥5 trillion in cash. These figures give the company a large financial safety cushion. But they do not change the overall trend: more of its AI plans are being paid for with loans, financing backed by assets and commitments that last many years.

Is this circular financing?

Calling it circular financing goes too far if that means money is simply being passed around to create the appearance of business revenue.

OpenAI is a real operating company. It has users, business customers and rapidly growing revenue. Its rental agreements are contracts for physical infrastructure that is supposed to be built and delivered. Nvidia is also involved as a separate strategic investor and provider of financial support.

A more accurate way to describe the arrangement is that all the companies are financially connected to one another.

SoftBank’s investment in OpenAI becomes more valuable if OpenAI grows. OpenAI’s expected growth creates demand for data centres backed by SoftBank. Those rental agreements can help the projects attract loans. The rising value of SoftBank’s OpenAI investment can also help it get other loans. This gives SoftBank more money to keep investing in the same AI industry.

There is a business reason behind every part of this arrangement. The danger is that more and more of these parts depend on the same big assumption: OpenAI must grow quickly enough, raise enough money and earn enough cash to justify its valuation and pay for huge amounts of computing capacity.

Has Masayoshi Son built a growth engine or a chain of financial risks?

At this point, it could still turn out to be either.

If OpenAI’s revenue continues to grow, its value could increase SoftBank’s net asset value and help it raise money in the future. OpenAI’s demand could fill data centres backed by SoftBank. This would make it easier for those projects to get loans that do not depend on SoftBank’s own balance sheet and to earn money over many years. Arm and Nvidia could also benefit as the need for computing power increases. SoftBank would then own investments across several parts of a new industrial system.

But if OpenAI’s growth or valuation falls short of expectations, the effects could spread in the opposite direction. The value of SoftBank’s investment would fall. The conditions on its loans could become harder to meet. Getting money for data-centre projects could become more difficult. SoftBank might have to provide more cash at a time when its borrowing costs and debt levels are already attracting attention.

This is why SoftBank’s latest loss is not the most important result in its financial reports. The company’s quarterly profits have always changed sharply as the value of Son’s investments goes up and down.

The bigger change is in how the business is organised. SoftBank is no longer just investing in companies that might benefit from AI. It is arranging money, electricity, property and computing capacity around the future of one company in which it is already the biggest financial supporter.

Masayoshi Son has spent $64.6 billion buying a stake in OpenAI and its potential future gains. He is now building part of SoftBank’s financial structure around the belief that those gains will eventually arrive.