To The Moon

Data center developer SB Energy, a unit of Japan's SoftBank Group led by legendary investor Masayoshi Son, is considering postponing its initial public offering until as late as next year. The company is building a massive artificial intelligence data center in Ohio for OpenAI.

Tech outlet The Information reported on the 22nd that SB Energy is weighing a delay of its IPO until December of this year or into next year. The outlet had reported in August that the company was aiming to list as early as September, indicating the timeline has slipped noticeably. The shift reflects rapidly cooling investor interest in IPOs of data center and AI cloud computing companies. Cloud computing stocks have traded erratically recently, with yields on some CoreWeave corporate bonds surging, prompting other data center developers to reconsider their listing plans. Australian cloud services provider Firmus withdrew its IPO plan earlier this month.

According to the report, SB Energy publicly filed its prospectus on the 2nd of last month and submitted a revised version on the 21st of the same month. Companies typically launch a roadshow for IPO investors about two weeks after a public filing with the U.S. Securities and Exchange Commission. But when market conditions turn volatile, the preparation period from filing to actual issuance can stretch longer. SB Energy is seeking to finance its Ohio data center through the IPO, planning to raise $5 billion to $7 billion, or roughly 6.61 trillion to 9.25 trillion Korean won, to build the facility that OpenAI will lease. The data center will use Nvidia chips, with a planned scale of 8.5 gigawatts, backed by a $105 billion credit guarantee from Nvidia.

Investment banks value SB Energy at about $60 billion, or roughly 79.35 trillion Korean won. However, The Information said IPO investors have already signaled they find that valuation hard to accept, because it rests entirely on expected future revenue while the company has no data centers in operation. The Ohio data center accounts for the bulk of SB Energy's total contracted data center revenue of $430 billion. The company plans to bring the facility online in phases starting in 2028, generating revenue through lease agreements that run 15 to 20 years once operations begin. By comparison, SB Energy generated $139 million in revenue in the first half of this year, or $278 million on an annualized basis. At a $60 billion valuation against that revenue, its price-to-sales ratio would exceed 200 times. A higher price-to-sales ratio means a company is being valued more richly relative to its current revenue.

Citing a person familiar with the matter, The Information said SB Energy executives are closely watching Anthropic's large IPO, expected next month. Anthropic is seen as the first AI startup to go public; if its shares perform well after listing, SB Energy could move ahead with its offering as early as December. The $60 billion valuation meets a cold reception: pressure in AI financing is spreading from the bond market to the IPO market. SB Energy's consideration of a listing delay exposes the financing tension in AI infrastructure, where long-term orders are enormous but near-term cash is insufficient. The SoftBank-owned data center developer originally sought to list as early as September and now is considering delaying to December or even 2027, with the $5 billion to $7 billion it plans to raise earmarked for the Ohio project leased by OpenAI.

Investor resistance to the roughly $60 billion valuation centers on the timing of revenue recognition: the company holds about $430 billion in data center contracts, but the related facilities are not yet in operation, and the main project is scheduled to come online in phases starting in 2028. Capital markets are reassessing how much today's valuation these forward rents can support, and how much funding is still needed before construction is complete. The shift comes after global assets just absorbed a prolonged shock from rising interest rates. The 10-year U.S. Treasury yield had climbed as high as 5.34%, the highest since 2002, significantly raising the discount threshold for future cash flows. The latest moves, however, show a change: a sharp selloff in tech stocks on Oct. 22 triggered safe-haven buying, pushing Treasury yields down more than 19 basis points in a single day, to about 4.97% during Asian hours on the 23rd.

As a result, the more accurate market narrative now is that although the benchmark yield trajectory has retreated, credit risk premiums and equity financing pressures for AI infrastructure-related companies remain pronounced. Stock markets have already priced in this concern. On Oct. 22, the Philadelphia Semiconductor Index fell 6.7%, down more than 26% from its June closing high; the Nasdaq Composite dropped 2.60%. SoftBank Group fell about 6.9% in Tokyo the same day, roughly halving from its June peak, and its credit default swap (CDS) prices had earlier risen to a near four-year high. On the 23rd, SoftBank recovered most of its intraday losses and ended down just 0.02%, but the short-term price repair has not changed the market's scrutiny of its financing burden. Credit markets offer a more direct pressure gauge for SB Energy, the data center developer under SoftBank Group.

A report on Oct. 22 showed that the yield on a CoreWeave bond maturing in 2031 reached 15.76%, with its five-year CDS spread climbing to 1,115 basis points; some media previously reported that its CDS pricing at one point implied a cumulative five-year default probability of nearly 63% — a probability that includes credit risk premiums and model assumptions and should not be read directly as the company's actual chance of failure. SoftBank itself is paying a clearly higher financing cost: in September it announced the issuance of about $11.1 billion equivalent in foreign currency bonds, including 7.5-year dollar notes issued at par with a coupon of 9.75%, with proceeds mainly used to pay for its investment in OpenAI and for general corporate purposes. CoreWeave's secondary market yields and SoftBank's new bond coupon together show that investors are demanding higher returns before they will keep funding AI expansion.

Orders lock in the future, but financing determines whether the company can make it to delivery. SB Energy's key constraint is the mismatch in timing between construction spending and rental income. Land, power connections, substation facilities, server rooms and cooling systems must be funded first, and long-term leases can only begin generating revenue after construction is complete and delivery conditions are met. The $430 billion in total contract value represents business scale spanning many years; it is not a cash reserve, nor the present value of that revenue. A Reuters Breakingviews analysis noted that only about one-tenth of the related data center contract revenue is expected to be realized within six years; this also means that even if demand is real, the project must still raise enough funds before production begins and bear the financing costs that accumulate along the way.

Dividing the $60 billion valuation by $278 million in annualized revenue corresponds to a price-to-sales (P/S) ratio of about 216 times; however, current revenue comes mainly from the existing energy business, while future valuation depends mainly on data centers not yet in operation, a difference in business stage. What truly determines how much valuation the company can bear is how much cash future rents can leave after deducting construction, operating and financing costs, and when that cash arrives. When benchmark rates for the relevant tenor and issuer credit spreads raise the funding threshold, investors will demand a lower entry valuation; an IPO delay in turn postpones equity replenishment, further increasing the developer's reliance on other financing channels.

Nvidia's officially disclosed $105 billion is a conditional guarantee cap, covering specified lease and power payment obligations tied to about 4.25 gigawatts of IT load at the Ohio campus. It typically takes effect as each lease phase begins and is triggered if a specified tenant defaults; it is not cash already delivered to SB Energy, nor does it cover the project's full cost. The 8.5-gigawatt planned scale cited in the report is not the same measure as the officially disclosed guaranteed scope; gigawatts measure power capacity and cannot directly represent model computing performance. The investment implication of SB Energy's listing plan adjustment is that the AI supply chain is drawing a sharper distinction between "holding orders" and "having the funds to complete orders." The more solid the financing arrangements and the closer a project is to being energized and delivered, the greater the certainty that its procurement demand translates into revenue for chips, power equipment and data centers. If Anthropic lists successfully, it can test the market's acceptance of AI companies' growth and valuations and improve the environment for subsequent offerings; but SB Energy will ultimately still need to convince investors with its own project economics and financing structure to pay today's price for forward rents.