AI faces the cost of waiting

The price of slowing down · Part 5

After Chinese chips and factories, this fifth instalment returns to financing: who can keep investing as timelines lengthen?

On 24 September, SoftBank set the terms of new borrowing intended partly to fund its investment in OpenAI. The longest dollar tranche carries an annual coupon of 9.75%. Issuance is scheduled for 29 September, so the transaction has not yet settled at this article’s cutoff. The terms nevertheless reveal an important feature of AI finance: an investor can provide patient equity to a laboratory by taking on obligations that require regular payments for years. SoftBank’s published terms · Reuters, 24 September

That same day, the Bank of Japan’s latest decision took effect. Its target for the overnight call rate rose to around 1.25%, following the decision announced six days earlier. Bank of Japan decision

The two numbers belong in the same story, but not in the same subtraction. One is a corporate dollar coupon paid over several years. The other is an overnight yen rate. Currency, maturity, borrower and risk all separate them.

Together, they challenge the idea that Japan is simply abandoning American technology for its own government bonds. A major Japanese group is still financing AI and accepting a substantial contractual cost to do so. Other Japanese investors may reconsider their holdings for entirely different reasons.

The common question is how much another year will cost before an investment delivers the cash expected from it. That does not settle the debate over AI safety. It identifies the financial conditions under which proposals to slow development would operate.

Patient equity can rest on impatient financing

SoftBank’s transaction involves two different relationships. It borrows from bond investors, then invests in OpenAI for an ownership interest. SoftBank’s lenders do not thereby become direct creditors of the laboratory. The investment agreement announced in February provides for another $30 billion, divided into three instalments. February investment agreement

The second $10 billion instalment was paid on 1 July, according to the subsequent execution notice. This is an established investment programme, not just a promise of future funding. July payment

An equity investor accepts uncertainty about distributions and the price at which a holding can eventually be sold. A bondholder expects the payments set out in a contract. Better technology may improve an equity investor’s prospects without producing cash that can immediately service the investor’s own borrowing. A higher valuation is not a cash receipt.

The three dollar tranches announced in September imply $941.25 million of coupons over a full year, assuming their full principal remains outstanding. This is a calculation from the disclosed amounts and rates, not an expense already incurred in 2026. It excludes the euro issuance, fees and any hedges. It is neither SoftBank’s consolidated financing cost nor a bill payable by OpenAI. Inputs for the calculation

The announced terms make the calculation reproducible:

Maturity	Dollar principal	Annual coupon	Coupons over one year
April 1, 2030	$1 billion	8.625%	$86.25 million
April 1, 2032	$4.5 billion	9.25%	$416.25 million
April 1, 2034	$4.5 billion	9.75%	$438.75 million
Total	$10 billion		$941.25 million

Source: SoftBank’s September 24 terms. For each tranche: principal × annual rate. Issuance expected on September 29, coupons payable semi-annually; the table shows a full year, excluding early redemption.

SoftBank can meet its payments from other assets and funding sources. Equity that looks patient from the laboratory’s perspective can therefore depend on more demanding payment schedules at the shareholder.

A delay might be covered with cash, asset disposals or further borrowing. Each option has conditions. An ownership interest in a laboratory and the debt used to help finance it need not share the same liquidity or exit date. The risk becomes clearer when the cash flows are followed across both contracts rather than grouped under a single fundraising headline.

The policy rate is only the starting point

US monetary conditions have also tightened. On 16 September, the Federal Reserve increased its target range by a quarter of a percentage point, to 3.75–4%. Both central banks therefore raised their short-term policy rates by 25 basis points. Those two changes, considered on their own, do not narrow the gap between the policy rates. Federal Reserve statement · Japanese policy rates before and after the decision

A ten-year investment is financed on a different timetable. Its price depends partly on the interest rates investors expect over that period, on compensation for holding a longer-term asset and on the borrower’s own risks. Today’s overnight target is not a ten-year corporate borrowing quote.

For 24 September, the Federal Reserve’s ten-year Treasury constant-maturity series records 5.18%. The Asian Development Bank’s market table shows 3.08% for Japan’s ten-year government bond on the same date. These are nominal market yields in different currencies, from different series and trading sessions that close at different times. They provide context, not a guaranteed return from switching between the two. Federal Reserve, DGS10 · AsianBondsOnline, Japan

Nominal rates do not, on their own, establish whether monetary policy is restrictive. Expected inflation matters for the real return. A rate can be higher than before without that observation alone resolving whether it is high enough to restrain the economy.

A corporate borrower must also compensate lenders for its particular risk. Financing can become more expensive because the reference rate rises, because repayment looks less secure, or because both happen together. A later policy-rate cut would not necessarily make borrowing cheaper for a company whose creditworthiness had meanwhile deteriorated.

The contract determines how quickly the change reaches the borrower. The coupon on an existing fixed-rate bond is not reset at every central bank meeting. Higher rates initially affect new funding and floating-rate debt, then borrowing that must be refinanced. A company may be insulated today while approaching a more difficult refinancing date.

Investors also gain alternatives for money they have not yet committed. They may demand more equity for the same investment or decline to tie up funds for as long. That changes the terms on which a company can raise money. It does not retrospectively remove cash from a funding round that has already been paid in.

This is why “there is still plenty of money” can be true without answering the financing question. The relevant issue is the price and conditions at which that money is available to a particular borrower or project.

The cost of another two years

Consider a hypothetical project that will pay 100 in five years, with no intermediate distributions. At a required annual return of 5%, that future payment is worth 78.35 today. At 10%, it is worth 62.09.

The calculation asks how much would need to be invested today at the chosen return to become 100 on the payment date. Discounting translates a future sum into its present equivalent at that required return.

Now delay the same payment by two years, leaving the required return at 10%. Its present value falls to 51.32. The additional wait reduces the calculated value by approximately 17.4% relative to the five-year, 10% case. Neither discount rate is an estimate of the actual cost of capital of any AI company.

A delay need not destroy value. More testing might prevent a failure, improve the product or make its future revenues more dependable. An assessment would then have to change the expected cash flows or their uncertainty as well as the date. The illustration holds those factors constant to isolate the effect of waiting.

That distinction is crucial to AI development. Postponing expenditure that can still be avoided may preserve cash. Postponing revenues from a facility that has already been financed can do the opposite. Companies with different investment commitments may therefore have opposing financial interests in a measure described by the same word: a slowdown.

A lower valuation must also be distinguished from an immediate funding shortage. The hypothetical project can still pay 100. What has changed is the price at which someone would buy that claim today. The change becomes a cash constraint if the company needs to raise further money against it, if collateral is reassessed, or if an investor has to sell.

The timeline is what connects valuation arithmetic to a financing problem. The same delay can be manageable for an investor with spare cash and disruptive for one approaching a payment deadline.

Japanese investors have different objectives

An insurer with future benefits to pay in yen may find Japanese bonds more useful as their yields rise: matching the currency and timing of assets to liabilities can reduce risks. An international equity manager has another mandate. An industrial group taking a strategic stake may not be seeking the highest available bond coupon.

Japan’s public pension reserve fund, GPIF, provides a clear example of a long-horizon allocation framework. Its reference portfolio for the period beginning in April 2025 retains four asset classes at 25% each: domestic bonds, foreign bonds, domestic equities and foreign equities. These are strategic targets with permitted deviations, not four accounts required to remain equal every day. The framework also allows review. GPIF’s policy portfolio

That mandate is not an instruction to sell every foreign equity when a Japanese bond offers a higher yield. Diversification, future pension requirements and allocation rules still matter. The return available on one security is not the sole objective of a retirement portfolio.

There can even be a reason to wait before buying domestic bonds. An existing bond paying a fixed coupon becomes less attractive when new issues offer more. Its market price generally has to fall to attract a buyer. A higher entry yield can therefore draw in new investors while imposing losses on earlier holders.

“Attractive” needs a holder, a horizon and an entry price. It also needs an assumption about whether the investor can wait for repayment. A government bond still carries market risk when it must be sold before maturity. A nominal coupon does not, by itself, preserve purchasing power against inflation.

The possible return home is therefore not a switch that automatically flips when the Bank of Japan meets. Investors can reach different decisions from the same yield curve because they have different obligations and different tolerance for losses along the way.

The dollars must eventually turn back into yen

An investor paying liabilities in yen but earning dollars has another calculation to make. A 5% dollar return does not guarantee a 5% gain in the currency that will ultimately be needed.

In a second hypothetical example, ¥15,000 buys $100 at an assumed exchange rate of ¥150 to the dollar. A one-year investment earns 5% and returns $105. At an unchanged exchange rate, that becomes ¥15,750. If the dollar instead buys only ¥135, the proceeds are ¥14,175: a dollar gain has become a 5.5% loss in yen.

The investor can fix in advance the rate at which the future dollar proceeds will be exchanged. This is a currency hedge. But the forward exchange rate reflects the difference between interest rates in the two currencies. In an ideal market without costs or risk, an investor cannot simply collect the higher rate and eliminate the exchange-rate exposure for free. BIS analysis of covered interest parity

Keep the dollar rate at 5% and assume a one-year yen rate of 1%. The forward rate consistent with these assumptions is approximately ¥144.286 per dollar. Selling the $105 at that rate produces ¥15,150, a 1% gain. This forward rate is calculated for the illustration; it is not a September 2026 market quote.

The arithmetic has an often overlooked implication. With the dollar rate unchanged, a higher yen rate can reduce the cost of hedging a dollar investment. Raising the yen rate in the example from 1% to 1.25% increases the covered final proceeds from ¥15,150 to ¥15,187.50. The comparable domestic investment would also earn more. The Japanese rate increase alone does not create a new relative advantage for one over the other.

The yen borrower and the hedged holder of a dollar asset are therefore in different positions. The borrower faces a higher funding cost. The asset holder can obtain better hedging terms, other things equal. Treating them as the same investor produces a supposedly automatic capital outflow that the mechanics do not support.

Real markets introduce further differences. Hedging demand, dealers’ balance-sheet costs and liquidity can push forward prices away from the textbook relationship. The BIS examines these deviations as the cross-currency basis. A long bond hedged through successive short contracts also leaves the cost of future hedge renewals uncertain. Mechanism and limitations

We have not reconstructed an executable currency-hedging quote for 24 September. Subtracting two central bank rates from a Treasury yield would not produce a guaranteed yen return. That calculation requires the appropriate market quotes, maturities and client terms.

Hedging also answers only the exchange-rate question. It does not remove the underlying bond’s interest-rate risk, ensure that an equity investment will succeed, or make an illiquid asset immediately saleable.

A carry trade involves a different risk

A strategy funded by borrowing yen may seek a higher return on an asset in another currency without fully hedging the exchange rate. That is the basic idea of a carry trade. Its result depends on funding costs, the asset’s performance and the exchange rate at which yen must be bought back.

If the yen strengthens, the debt becomes more expensive in the asset’s currency. If the asset falls as well, the investor’s remaining equity cushion shrinks further. A lender may require additional collateral. The investor must then provide cash or sell, possibly long before a long-term investment thesis can be tested.

The BIS documented this amplification through deleveraging and margin calls during the market turbulence of August 2024. It is evidence of a mechanism, not a forecast of a repeat in September 2026. BIS assessment published on 27 August 2024

Nor should all of Japan’s foreign assets be relabelled carry trades. A holder’s country of residence does not establish whether there was yen borrowing, how large it was or whether currency risk was hedged. A separate BIS note explains the limitations of measuring these trades through available statistics. Measuring carry trades

The possible connection to AI is straightforward. Forced selling can reach liquid technology stocks and influence the terms available for subsequent fundraising. But assigning a specific amount of carry-trade financing to a particular laboratory would require tracing positions that the aggregate data do not reveal. A speculative yen position is not an identified ownership interest in OpenAI.

This distinction prevents a common analytical shortcut: turning a plausible channel of market stress into a claim that the funding of an entire industry has already disappeared.

August brought equity purchases and debt sales

Japanese transaction data provide a check on the broader narrative. In August 2026, designated major investors made net purchases of ¥1,298.3 billion in foreign equities and investment-fund shares, alongside net disposals of ¥143 billion in medium- and long-term foreign debt and ¥1,018.7 billion in short-term debt. The overall balance was still ¥136.6 billion of net acquisitions. The release was published on 8 September. Ministry of Finance monthly release, page 1

Asset categories moved in different directions, with an overall balance still showing net acquisitions. These aggregates cover many countries and assets; identifying American AI exposure would require the underlying securities and fund holdings.

The survey covers designated reporting institutions. Securities are classified by the issuer’s residence, rather than simply by their currency. Net purchases of foreign fund shares do not identify the businesses held inside those funds. Statistical coverage

The release covering 13–19 and 20–26 September is scheduled for 1 October. As of 27 September, the series therefore cannot yet measure the complete response around the latest Bank of Japan decision. Publication schedule

The distinction between transactions and valuations matters too. Foreign holdings can be worth fewer yen because exchange rates have moved, even when no securities were sold. A fall in the reported value of a portfolio is not enough to establish repatriation.

Conversely, funding can become more expensive without an exodus of existing holders. New buyers may simply require better terms. For a company that needs another round of capital, the conditions offered by the next investor can matter more than whether previous investors have sold.

Taken in isolation, a securities sale transfers money to another holder; it does not make the money disappear. What can become scarce is the willingness to finance a particular risk at the proposed price, or the ability to sell quickly without a large discount. Neither dimension of liquidity can be read directly from a headline measure of available cash.

SoftBank changes the source and maturity of its funding

The September financing is revealing in another respect: the Japanese group is borrowing dollars and euros in international markets outside Japan. Its nationality does not make this transaction an automatic recycling of Japanese household savings or a cheap yen-funded trade. Offering terms

The financing history also shows why announcements should be traced rather than added together. A $40 billion bank facility, agreed in March and due on 25 March 2027, provided borrowing capacity. Its maximum size was not an immediate $40 billion cash disbursement. March bridge facility

On 9 September, SoftBank announced an early repayment of the then-outstanding balance, scheduled for the 15th. The new bond programme is intended partly to finance the final OpenAI instalment and to allow the group to cancel remaining undrawn bank capacity. Refinancing an investment commitment, repaying debt and issuing equity at the laboratory are separate events. Early-repayment announcement · Stated use of bond proceeds

Replacing transition financing with longer maturities can reduce the need to find another lender quickly. The price of that time becomes explicit. Assessing the group’s solvency requires comparing contractual payments with its cash, other assets and repayment schedule.

The credit assessments reported by Reuters also differ. CreditSights highlights asset concentration and pressure on cash flow; S&P considers that the delay to OpenAI’s listing does not immediately impair SoftBank’s credit quality. These are attributed assessments, not guarantees. Credit views reported by Reuters

OpenAI also names SoftBank among the partners in its fundraising announcement of March 2026. Both parties thus describe the same financing relationship. OpenAI announcement

SoftBank is continuing its investment programme despite higher Japanese rates. This is one allocation decision in a market where investors have different resources and obligations.

Waiting can favour companies that already have the money

The connection between rates and competition remains worth examining. Consider two laboratories facing the same additional year of research or mandatory testing. One already has enough funding. The other must still raise it. An identical technical requirement need not impose an identical financing burden.

The first may absorb the delay or support its staff through another business. The second may have to negotiate in a less favourable market, surrender more ownership or reduce its programme. A common obligation can therefore have different competitive effects depending on the funding already available. The competitive effect would depend on balance sheets and timing; the intentions behind a rule require a separate investigation.

A January 2026 BIS bulletin identifies the move from financing AI investment through existing operating cash flows towards greater reliance on debt as an important issue. Its authors connect the sustainability of that financing to the earnings companies eventually deliver. Financing the AI boom

Japan is also part of the underlying industrial economy. In its September assessment, the Bank of Japan identified AI-related demand as a contributor both to economic activity and to pressure on prices. Technology investment therefore helps shape the environment in which interest rates are set. Rate changes cannot simply be treated as a wholly unrelated shock arriving from outside the sector. Bank of Japan economic assessment

In its April report, based on data available through the end of March, the Bank of Japan assessed Japan’s financial system as broadly stable and its banks as able to withstand its stress scenarios. That dated system-wide assessment has a different scope from a review of an individual borrower in September. Financial System Report

Who can afford to wait?

A safety argument should be assessed against the risks it documents and the effectiveness of the proposed response. The cost of that response belongs in the assessment, but does not replace it. A genuine risk may justify expenditure. Expensive safeguards are not evidence that the risk was invented.

Financial analysis asks a different set of questions: who can afford to wait, who still needs lenders or investors, and which obligations continue to fall due? It can identify potential beneficiaries of a constraint without assuming that they designed it.

Connecting calls to slow development with Japanese capital repatriation would require additional investigation: dated internal decisions, their motivations and demonstrably changed funding conditions. The sources assembled here document constraints and transactions against which such a hypothesis could be examined.

The practical distinction is between spending that a pause can still avoid and revenue that a pause would postpone. The former can preserve cash; the latter can extend a period in which payments continue before the expected income arrives. Assessing a slowdown’s financial benefit begins with locating it on that timeline. Lenders can know their next payment date without knowing how the debate over AI’s future will end.

Sources and method

Research cutoff: 27 September 2026. Market yields refer to 24 September; monthly transaction data cover August. Statistics may be revised. SoftBank and OpenAI disclosures describe their own transactions and do not independently establish future investment returns. No direct interviews were conducted.

The simulations isolate mechanisms. They do not value a company, estimate an executable current hedged return or assign a default probability. Amounts are nominal and before tax unless otherwise stated. A bond yield, coupon, policy rate, investment hurdle rate and hedged financing cost are not treated as interchangeable.

• SoftBank Group, Issuance of Foreign Currency-Denominated Senior Notes (2026-09-24).
• Bank of Japan, Change in the Guideline for Money Market Operations (2026-09-18).
• SoftBank Group, Follow-on Investments in OpenAI (2026-02-27).
• SoftBank Group, Execution of Follow-on Investment (Second Tranche) in OpenAI (2026-07-01).
• Federal Reserve, Federal Reserve issues FOMC statement (2026-09-16).
• Federal Reserve / FRED, DGS10: Treasury ten-year constant-maturity market yield.
• Asian Development Bank, AsianBondsOnline: Japan Market Watch.
• GPIF, 基本ポートフォリオの考え方 / Policy portfolio framework.
• Bank for International Settlements, Covered interest parity lost: understanding the cross-currency basis (2016-09).
• Bank for International Settlements, The market turbulence and carry trade unwind of August 2024 (2024-08-27).
• Bank for International Settlements, Sizing up carry trades in BIS statistics (2024-09).
• Japan Ministry of Finance, International Transactions in Securities: monthly release (2026-09-08).
• Japan Ministry of Finance, International Transactions in Securities: coverage and series.
• Japan Ministry of Finance, International Transactions in Securities: release schedule.
• SoftBank Group, Execution of Bridge Facility Agreement Primarily for the Follow-on Investments in OpenAI (2026-03-27).
• SoftBank Group, Early Repayment of Bridge Loans (2026-09-09).
• OpenAI, OpenAI raises $122 billion to accelerate the next phase of AI (2026-03-31).
• Bank for International Settlements, Financing the AI boom: from cash flows to debt (2026-01-07).
• Bank of Japan, Financial System Report, April 2026 (2026-04-21).
• Reuters / Investing.com, SoftBank issues $11.1 billion in bonds in OpenAI financing push (2026-09-24).
• Bank of Japan, Change in the Guideline for Money Market Operations: reference (2026-09-18).