SB Energy는 OpenAI에 55억 달러를 지불하여 임대를 확보했으며, 이는 컴퓨팅 용량이 순수 기가와트보다 주요한 희소성 제약 요인임을 강조한다.
A SoftBank-backed data center venture with no operating facilities granted its anchor tenant equity stakes valued at nearly its entire IPO. That pricing reveals what the AI buildout is actually short of. According to draft IPO documents reviewed by The Wall Street Journal, SB Energy issued OpenAI warrants estimated at $5.5 billion to secure the AI company as a tenant. The firm is targeting a $5 billion to $7 billion listing as soon as September. Placed side by side, the customer acquisition cost consumes 92% of the capital the offering aims to raise. In this trade, capacity is not the bottleneck. What is scarce is a signature that lenders will accept.
Every operator in the sector echoes the same assessment of the shortage: power is constrained, land near transmission corridors is limited, data center shells are unavailable, and turbines are booked through 2029. If physical capacity were the binding constraint, SB Energy would be auctioning its 4.25-gigawatt site in Pike County to the highest bidder and capturing the premium. Instead, it paid the bidder. Nvidia did the same, guaranteeing up to $105 billion of that tenant’s rent. AMD compensated the same developer with warrants. CoreWeave paid in stock at its own IPO. With five suppliers competing for one customer, capital flows backward down the supply chain.
The Journal reported on August 30 that SB Energy, majority-owned by SoftBank, granted OpenAI a block of warrants as an incentive to anchor its data center campuses. SB Energy initially issued these warrants in January 2026 at an estimated value of $3.6 billion. By June 30, their valuation had risen to approximately $5.5 billion. They vest in tranches following the public listing, contingent upon SB Energy achieving specific market capitalization milestones.
This remeasurement dominates the company’s income statement. SB Energy’s net loss expanded to roughly $3.2 billion in the first half of 2026, compared to about $250 million a year earlier, driven primarily by changes in the estimated value of its warrant liabilities. Meanwhile, revenue from its renewable energy division—which constructs solar and battery systems—reached nearly $140 million for the half, a 66% increase. The company currently has 800 megawatts of compute capacity under construction and operates zero data centers.
OpenAI occupies four distinct positions at the table. It leases the capacity. It contributed $500 million to SB Energy and is expected to hold a single-digit stake post-offering. It holds the warrants. Additionally, SB Energy has committed to purchasing at least $50 million of OpenAI software and services by 2028, including ChatGPT Enterprise, effectively turning the tenant into a vendor for its own landlord.
The draft prospectus offers no ambiguity regarding these relationships. SB Energy explicitly describes itself as substantially dependent on OpenAI, warning that any deterioration in the AI company’s financial condition would negatively impact it. The filing also discloses that funding the Ohio campus relies meaningfully on Nvidia standing behind the project through a residual value guarantee.
Nvidia formalized its involvement on August 17, securing land, power, and shell capacity at SB Energy’s PORTS-Pike Technology Campus in Pike County, Ohio. The facility is being developed on the site of the decommissioned Portsmouth Gaseous Diffusion Plant. SB Energy will build, own, and operate the campus, which OpenAI will take on a 20-year lease. Nvidia will serve as the exclusive AI compute provider, deploying its DSX AI factory platform with an initial allocation of 4.25 IT-gigawatts and an option on the remaining 3.75. Capacity will phase in starting in 2028. Goldman Sachs and JPMorgan advised SB Energy, while Morgan Stanley advised Nvidia.
Nvidia also invested $1.5 billion in SB Energy alongside SoftBank and OpenAI. Its second-quarter filing capped the guarantee obligations for the Ohio campus at $105 billion, to be triggered in phases as conditions are met, with the first expected in fiscal 2029. The Journal subsequently reported that Nvidia committed an additional $3 billion through two private transactions tied to the listing, one of which grants it the right to purchase shares at a 10% discount to the offering price.
Beneath these arrangements lies the largest figure rarely discussed: the power commitment. SB Energy and SoftBank agreed to construct at least 10 gigawatts of new generation capacity. The Department of Energy estimates that at least 9.2 gigawatts of this will be natural gas-fired, backed by $33.3 billion in committed Japanese funding. The partners also agreed to fund $4.2 billion in new transmission infrastructure with AEP Ohio, structured so that SB Energy bears the cost rather than Ohio ratepayers. As previously noted, this payment structure is exceptionally rare in AI power financing.
Reading the capital stack in sequence reveals a coherent logic. SB Energy operates no data centers, generates no data center revenue, and its renewable business earns roughly $140 million per half-year. It cannot fund an 8 IT-gigawatt campus from internal cash flow, nor would any lender issue a construction loan against a speculative shell. What makes the debt viable is a 20-year lease with a counterparty that investment-grade credit committees can underwrite.
OpenAI is the counterparty everyone wants but nobody can accurately price. It generates genuine revenue at a scale unmatched by any other AI lab, yet it burns cash at a magnitude no traditional commercial real estate tenant approaches. To bridge this gap, the parties constructed a synthetic credit rating from complementary assets. Nvidia supplied balance-sheet strength through a capped guarantee. SoftBank provided equity and political leverage. SB Energy contributed the only asset it had left: equity upside in itself.
The warrant structure is the elegant component. SB Energy incurs no upfront cash cost. If the company fails, the warrants expire worthless, resulting in zero incentive expense. If it succeeds, the tenant realizes gains, but only after SB Energy reaches market capitalization thresholds that first enrich the founders and SoftBank. For a pre-revenue developer acquiring a creditworthy signature it cannot afford in cash, the instrument is nearly optimal.
The pricing mechanics, however, invite scrutiny. Customer acquisition costs typically behave predictably: paid at signing, fixed, and amortized against lifetime value until the ratio improves. SB Energy’s model operates inversely. The warrants vest on market capitalization milestones and sit on the balance sheet as fair-value liabilities. Consequently, every dollar of equity value the company creates transfers a portion to the tenant and expands the reported loss. Structurally, the income statement is short the company’s own share price.
Any investor considering the IPO should weigh this carefully. The dilution schedule is directly indexed to the exact metric the offering exists to maximize. Success acts as the trigger. The $3.2 billion first-half loss is not an operational failure; it is the accounting manifestation of customer acquisition cost appreciating.
The escalation trajectory, largely overlooked in current coverage, began with CoreWeave. In March 2025, it signed an $11.9 billion five-year contract with OpenAI and issued $350 million of its own stock through a private placement at its listing, receiving no proceeds from that issuance. At the time, CoreWeave was targeting a raise exceeding $4 billion. The signing consideration represented less than 9% of its intended capital raise.
Fifteen months later, SB Energy paid $5.5 billion against a $6.0 billion midpoint target—a staggering 92%. Across two pre-IPO infrastructure suppliers, the same anchor customer, and identical instrument logic, the price of the signature increased roughly tenfold as a percentage of the offering. The instruments differ materially: CoreWeave issued common stock while SB Energy issued milestone warrants, and CoreWeave ultimately priced below its target. Yet the directional trend remains undeniable.
The per-unit capacity arithmetic is even starker. Applied to SB Energy’s 800 megawatts under construction, the $5.5 billion translates to approximately $6.9 million per megawatt of tenant acquisition cost before a single server rack ships. Spread across the full 4.25-gigawatt first phase in Ohio, the figure drops to roughly $1.29 million per megawatt. While substantial, this represents a meaningful fraction of industry-standard build costs, which range around $10 million per megawatt.
A fourth campus definitively resolves the scarcity debate. SB Energy has approximately 900 megawatts planned in Scurry County, Texas, with no named customer. A developer sitting on unallocated capacity while simultaneously paying $5.5 billion to secure another tenant is not suffering from a gigawatt shortage. It is facing a shortage of tenants whose credit profiles clear institutional hurdles.
The $50 million software commitment serves as the clearest signal. It equals just 0.9% of the warrant value. No party negotiates a $50 million purchase order into a transaction of this magnitude for purely commercial reasons. Its purpose is to establish a secondary revenue direction on paper, reinforcing the symbiotic relationship.
The concentration risk is disclosed, which is preferable to most peers, yet remains severe. SB Energy’s equity thesis hinges on two counterparties. One is an unprofitable enterprise recording annual losses in the tens of billions. The other is the semiconductor manufacturer guaranteeing that tenant’s rent, whose own $500 billion financing platform and 25% residual backstop exist because the marginal buyer of an AI factory ran out of