NextFin News - SoftBank's $50 billion data-centre listing has hit a wall. SB Energy, the SoftBank-backed developer at the centre of the AI infrastructure boom, has slowed its planned U.S. initial public offering, a delay that lands just as OpenAI - its anchor tenant and strategic partner - leans toward pushing its own market debut into 2027. The timing is awkward for a company asking public investors to underwrite a roughly $178 billion construction programme on the back of contracts that have not yet produced a single dollar of data-centre revenue.
The situation: a $439 billion backlog with no operating facilities
SB Energy filed its registration statement with the U.S. Securities and Exchange Commission on August 31, 2026, describing itself as an integrated data-centre and power infrastructure company. The headline numbers are staggering: roughly $439 billion of contracted backlog, about $430 billion of it tied to data centres covering 8.8 gigawatts of signed capacity, with leases running a weighted average of 19.6 years. Only about $10 billion of the backlog comes from power projects.
But the financials tell a different story. For the six months ended June 30, 2026, the company reported revenue of $138.7 million, up 66.4% from $83.3 million a year earlier, almost entirely from its legacy power business. It posted a net loss of $3.21 billion, compared with a net loss of $215.5 million in the same period of 2025, and an operating loss of roughly $551.6 million. Critically, none of its data centres is operational, and the data-centre business has generated no revenue.
The loss expansion is not a cash-burn story in the way it first appears. A breakdown of the filing shows much of it reflects non-cash accounting charges, including roughly $2.57 billion from warrant revaluation and about $589.5 million in stock-based compensation. That distinction matters: it means the reported loss overstates the cash actually leaving the business. But it does not change the harder problem. The company estimates it must spend approximately $178 billion to convert its backlog into operating assets, roughly $174 billion of that for the data-centre segment.
The customer concentration is extreme. OpenAI affiliates lease about 8,777 of the company's 8,827 megawatts of signed data-centre capacity - roughly 99%. The flagship project is the PORTS-Pike Technology Campus in Pike County, Ohio, an 8-gigawatt site where 17 leases with an OpenAI affiliate were signed on August 17, 2026, each a 20-year triple-net lease with rent commencement phased between 2028 and 2032. The first meaningful data-centre revenue is not expected until the fourth quarter of 2026, when the initial phase of the Cosmos Technology Campus in Travis County, Texas - a 50-megawatt site leased to a SoftBank affiliate - reaches rent commencement.
The filing is blunt about the dependency. SB Energy warns it is "substantially dependent on OpenAI as a tenant and strategic partner," and adds that this concentration means near-term revenues, project-level financing arrangements, and development plans are significantly linked to OpenAI's continued performance. That dependence runs in both directions. OpenAI has been issued warrants valued at roughly $5.5 billion - up from $3.6 billion when they were issued in January - and Nvidia has committed to invest $1.5 billion in a private placement at the IPO price. Both companies would own significant stakes in a listed SB Energy, binding the fortunes of the AI era's two most valuable private companies to a single infrastructure vehicle.
Why the IPO is slowing
The delay reflects a simple problem of demand. SoftBank is seeking a valuation of roughly $50 billion for SB Energy, with the IPO expected to raise between $5 billion and $7 billion. But investor enthusiasm for AI infrastructure, while still strong, has become more discriminating. The company's own filing shows that approximately $357 billion of its contracted revenue is not expected to be recognized until 2034 or later - a timeline that asks public-market investors to wait nearly a decade for the bulk of the promised cash flow.
OpenAI's decision to push its own listing into 2027 compounds the difficulty. Sam Altman, OpenAI's chief executive, has described a 2026 IPO as "ill-advised," citing safety and alignment priorities. For SB Energy, the consequence is twofold. First, a public OpenAI would have provided a visible valuation anchor and a liquid currency for the ecosystem; without it, investors must underwrite SB Energy's OpenAI exposure largely on faith. Second, the delay signals that even the most sought-after private AI company is not immune to public-market scrutiny - a warning for any infrastructure bet whose thesis depends on AI tenants maintaining their private-market valuations.
There is also the matter of SpaceX. The satellite operator's June listing raised more than $85 billion and reached a debut valuation of $1.77 trillion, only to see its shares slide afterward. The volatility of the year's marquee AI-adjacent IPO has made new investors more cautious about entry pricing - precisely the problem for a company that needs to clear a $50 billion valuation to fund its ambitions.
"Investors have to be convinced that hundreds of billions of contracted demand can be turned into cash flow over the coming years," said IPOX Research Associate Lukas Muehlbauer.
That is the test SB Energy now faces, and the delay suggests SoftBank does not yet believe it can pass it on its preferred terms.
The mechanism: backlog is not cash flow
The central question SB Energy's IPO poses is whether a signed lease is the same thing as an investable asset. It is not - not until the facility is built, powered, and delivering uptime under contract. The transmission mechanism runs like this: a data-centre lease creates a contractual revenue claim, but that claim only becomes cash flow after the company spends the capital to build the facility, secures the power, and hands over operating capacity. Between signature and revenue lies the entire execution risk.
This is where the structure of the deal matters. SB Energy's leases are long-dated triple-net agreements, meaning the tenant bears most operating costs. That is attractive once the asset is running. But the company still carries the development risk: land, power interconnection, equipment lead times, construction, and financing. Its own prospectus flags long lead times for power equipment and growing local opposition to data centres as factors that could hinder delivery.
The capital math is unforgiving. To build its contracted pipeline, SB Energy needs roughly $178 billion. If it maintains a typical 10% equity stake in project financing, analysts estimate it will need to secure an additional $7 billion in equity beyond the IPO proceeds, alongside substantial debt. A $500 million share sale to Japanese investors - disclosed in a September 15 filing, with proceeds earmarked for general operating costs - is a rounding error against that requirement. The IPO is not the finish line; it is the entry fee.
Cyclical or structural: the buildout is structural, the valuation window is cyclical
This is the judgment that determines how to read the delay. The demand for AI data-centre capacity is structural. It is driven by a durable shift in how the global economy consumes compute, backed by long-term contracts with some of the world's best-capitalized companies. Power constraints, equipment bottlenecks, and local permitting fights are real, but they are friction on a trend, not evidence that the trend is reversing. The buildout will happen. Analysts at UBS estimate $511 billion will be spent by 2030 to close the power-demand gap, driven largely by data centres.
The valuation window, however, is cyclical. Public-market appetite for pre-revenue infrastructure stories priced on decade-out cash flows expands and contracts with liquidity, with the performance of comparable listings, and with the perceived safety of the anchor tenant. The delay is not a verdict on AI demand; it is a verdict on the price at which that demand can be sold to public investors right now. SoftBank's read of the situation appears to be that waiting for a better window is cheaper than accepting a lower valuation - a cyclical call on a structural asset.
That distinction is important because it means the delay is likely a timing event, not a thesis break. But timing events can still destroy returns if the capital math changes while you wait. If power costs rise, if equipment lead times lengthen, or if OpenAI's own trajectory wobbles, the $178 billion buildout becomes more expensive and the $439 billion backlog becomes worth less in present-value terms.
The second-order problem: entanglement risk
The first-order read of SB Energy is straightforward: it is a leveraged bet on AI data-centre demand. The second-order read is more uncomfortable. SB Energy, OpenAI, and Nvidia are binding themselves together so tightly that a problem at any one of them becomes a problem at all three.
OpenAI holds $5.5 billion of warrants. Nvidia has committed $1.5 billion and agreed to backstop up to $105 billion for the Ohio project. SoftBank is the controlling shareholder. If OpenAI delays its IPO because its private valuation cannot be defended in public markets, then the warrants OpenAI holds in SB Energy - and the tenant commitments that underpin SB Energy's backlog - are exposed to the same repricing. The structure that makes SB Energy attractive - its deep entanglement with the AI ecosystem's winners - is also its concentration risk.
There is a further twist. If OpenAI delays its IPO to grow into a higher valuation, it may seek to renegotiate the economics of its data-centre commitments or slow its buildout pace. SB Energy's backlog is only as good as OpenAI's willingness and ability to take delivery. A tenant that is itself under public-market pressure is a tenant with more incentive to delay, renegotiate, or walk.
The counter-thesis: SoftBank is right to wait
The strongest argument against reading the delay as weakness is that SoftBank has done this before, successfully. Masayoshi Son has a track record of timing large listings, and the company's confidence in securing backing is grounded in that history. From this angle, the delay is not a retreat; it is discipline. The AI infrastructure theme is not going away, demand is contracted for two decades, and the strategic investors - OpenAI and Nvidia - are not going anywhere. Waiting for a better window is the rational move for a seller who does not need to sell.
There is evidence for this view. The Japanese share sale proceeding alongside the delay shows the company is still moving forward with listing mechanics. The contracts are signed. The tenants include the most valuable private companies in AI. And the power-demand gap is real.
The counter-thesis holds only if one condition is met: that the window actually reopens on better terms. If public-market appetite for AI infrastructure cools further, or if OpenAI's 2027 listing itself disappoints, SoftBank could find itself waiting not months but years - and watching its capital requirements grow while it waits.
The falsifying signal is specific. If SB Energy proceeds with a U.S. listing at a valuation materially below $50 billion - say, under $40 billion - it would indicate that the delay was not discipline but a failed attempt to hold price, and that the market is discounting the backlog for execution and concentration risk. Conversely, if the company lists at or above $50 billion within the next two quarters, the delay will have been a successful timing call.
What to watch
In the short term, watch the IPO pricing. The size of the raise, the valuation, and the quality of the anchor book will tell you whether the delay bought SoftBank a better price or merely more time. Watch also for any revision to the Cosmos rent-commencement date in the fourth quarter of 2026 - the first real test of whether contracted capacity becomes revenue on schedule.
Over the medium term, the key metric is conversion: how much of the $439 billion backlog moves from signed lease to operating asset, and at what capital cost. The company needs roughly $178 billion to build what is under contract; any slippage on that number, or any need to raise equity on worse terms, dilutes the thesis.
Over the long term, watch OpenAI. The tenant is the thesis. If OpenAI lists successfully in 2027 at a strong valuation, SB Energy's entanglement becomes an asset. If OpenAI's listing struggles or its buildout slows, the concentration that makes SB Energy unique becomes its vulnerability.
Conclusion
The base case is that SB Energy lists, but at a price that reflects the gap between signed contracts and operating cash flow. The upside case is a clean $50 billion-plus listing on the back of a strong AI IPO window and a credible OpenAI listing timeline. The downside case is a prolonged delay or a down-round valuation that forces SoftBank to fund the buildout with more of its own capital.
The delay is a reminder that in infrastructure, the contract is the beginning of the risk, not the end of it. SB Energy has signed the future; now it has to build it, fund it, and convince public investors that a lease signed today is worth the price being asked. The AI buildout is real. The question is who pays for it, and when.
Explore more exclusive insights at nextfin.ai.
