OpenAI Can Wait for Its IPO. SoftBank's Debt Cannot.

OpenAI will not go public this year. Sam Altman said as much on Saturday, calling a 2026 IPO an "ill-advised moment" given AI safety concerns, and pointing the company toward 2027 instead.

The less obvious question is who pays for that patience.

SoftBank Group paid $41 billion for an 11% stake in OpenAI. If OpenAI lists at the $1 trillion valuation Altman once targeted, that stake is worth about $110 billion. Enough to erase SoftBank's entire debt pile. But the debt pile exists right now, and it is growing.

Total debt at SoftBank jumped 57% in one year to ¥25.4 trillion. EBITDA fell 43% to ¥914 billion. Free cash flow hit negative ¥2.16 trillion. The company announced a ¥1 trillion ($6.3 billion) bond sale at 4.3% to 4.9% interest to fund OpenAI commitments. The stock fell 5.3% the next day, down 26.4% over three months.

This is not a normal leverage profile. It is a venture-capital firm that has borrowed like a leveraged buyout. SoftBank is funding illiquid, long-duration bets with fixed-coupon bonds that demand annual interest payments regardless of whether OpenAI ships a safety protocol or goes public.

The structure only works if one thing happens: the private companies SoftBank backs turn into public companies at higher valuations, SoftBank sells shares into that process, and the cash pays down the debt. That was the model in 2023 when Arm went public. SoftBank sold about 10% of Arm and raised roughly $10 billion. Arm shares have since climbed from the $51 IPO price to around $241 today — a $258 billion market cap that makes SoftBank's remaining ~90% stake very valuable on paper.

But paper value does not pay bond coupons. SoftBank needs actual liquidity from actual exits. And right now, the biggest exit on its horizon has paused.

OpenAI filed confidential IPO paperwork in June, targeting a $1 trillion valuation. By September, Altman had changed the answer. He agreed with Anthropic CEO Dario Amodei that the frontier should be paced more carefully. The company's dual non-profit and for-profit structure requires decisions "not obviously in the interest of our business and our shareholders," he said — and going public would "complicate that math".

In other words, OpenAI has no urgent reason to sell shares. It raises enough privately. Its products generate revenue. Safety and governance can wait for the market to be ready.

SoftBank does not have that luxury. Its current ratio — short-term liquid assets divided by short-term obligations — is 0.8x. That means the company cannot cover its near-term bills from its near-term assets without borrowing more or selling something. It borrowed more in April ($3.5 billion across six tranches), April again (a $40 billion bridge loan for Anthropic), and August (the record ¥1 trillion sale). Rating agency S&P put SoftBank on negative outlook in March.

There is something structurally asymmetric here. The company with the product controls the timeline. The company with the debt does not.

This is not a new pattern for SoftBank. It ran the same model in the late 1990s during the telecom boom, loading up on debt to fund bets on mobile infrastructure. Masayoshi Son lost nearly ¥5 trillion before the thesis worked — a decade later, with the Japanese phone business that is now part of SoftBank Corp. The difference then and now is scale. The current AI commitments are more than $60 billion and counting, including a $100 billion data-center project called Stargate in the U.S.. Son, now 68, abandoned plans to step down within two years in June and said he would stay for another decade or more.

He can afford the time. The balance sheet might not.

The irony is that OpenAI's rival is doing exactly what OpenAI won't. Anthropic filed for its own IPO this summer and is expected to begin marketing in mid-October, with a listing before year-end. Some investors project a $2.3 trillion valuation. Anthropic posted its first-ever operating profit this quarter — $559 million on $11.5 billion in revenue — and its annual revenue run rate is estimated at $65 billion. It shares the same safety concerns. It just doesn't need to wait on them to go public.

SoftBank also has a stake in Anthropic, funded through that $40 billion bridge loan. So Anthropic's IPO could partially offset the OpenAI delay. But the Anthropic stake is new, and it's unclear what percentage SoftBank holds relative to the multiple investors in Anthropic's Series H round. The OpenAI stake was the centerpiece — $41 billion committed, 11% of a company that was supposed to list this year.

What the investor in SoftBank Group is really holding is a single hypothesis: that Son's ability to raise capital will outlast the time it takes for these private companies to mature and go public. The hypothesis worked last time. The debt was bigger this time. The interest costs are bigger too — the new 10-year bonds carry an 8.5% coupon, and the company is borrowing in dollars and euros at rates ranging from 6.4% to 8.3%.

That is a cost of capital that demands returns in the medium term, not the distant future.

The question for someone watching this stock is not whether AI will be big enough to justify SoftBank's bets. It almost certainly will. The question is whether SoftBank's balance sheet can survive the gap between commitment and payoff. One concrete number to track: can the company keep raising debt at acceptable rates, or will lenders start pricing in the risk of a company whose biggest asset is a delayed IPO?

If Anthropic lists successfully and OpenAI follows in 2027, the math works out. If something breaks the IPO window, or if OpenAI's timeline stretches further, or if the interest on ¥25 trillion of debt starts consuming the cash flow from SoftBank's telecom business — which, remember, saw its own EBITDA drop 43% — then the whole structure gets tested.

Patience is a luxury that belongs to the person who owns the product. The person who funded it with borrowed money is on a clock.