What are the OpenAI SB Energy warrants? The OpenAI SB Energy warrants are stock warrants that data center developer SB Energy granted to OpenAI as an inducement to sign as anchor tenant, valued at roughly $3.6 billion when issued in January 2026 and about $5.5 billion by June 30, according to draft IPO documents reviewed by The Wall Street Journal and reported on August 30, 2026. In plain commercial real estate terms, the landlord paid the tenant to sign the lease. For the wider context on how AI capital is rewriting deal structures, see our guide to AI in CRE finance and capital markets.
Key Takeaways
- SB Energy issued OpenAI warrants worth about $5.5 billion by June 30, 2026 to secure a 20-year anchor lease across roughly 8 gigawatts of planned Ohio capacity.
- This inverts the standard concession stack. Instead of free rent or a tenant improvement allowance, the landlord handed the tenant equity in the landlord itself.
- SB Energy reports a contracted backlog above $400 billion while operating zero completed data centers, so backlog and collected revenue are not the same number.
- CBRE put primary market data center vacancy at 1.4% in H1 2026 with 80.4% of under construction capacity preleased, so raw scarcity does not explain the payment.
- The transferable lesson is to underwrite what a landlord gave up to win an anchor tenant, not just the headline rent roll it produces.
What Actually Happened
SB Energy, the power and data center developer majority owned by SoftBank, granted OpenAI warrants to buy its stock. Issued in January 2026 at an estimated $3.6 billion, they had appreciated to roughly $5.5 billion by the end of June, and vest in tranches after the listing, keyed to market capitalization milestones.
What the warrants bought was a signature. OpenAI executed 17 separate lease agreements covering approximately 8 gigawatts of computing capacity on a 20-year term at SB Energy's southern Ohio campus in Pike County, built on the site of the decommissioned Portsmouth Gaseous Diffusion Plant. We covered the groundbreaking on that project in March in our piece on SoftBank's $500 billion Ohio AI data center campus. What is new today is not the campus. It is the price the developer paid to fill it.
The relationships run in several directions at once. OpenAI separately invested about $500 million in SB Energy and is expected to hold a single-digit percentage stake after the offering. SB Energy committed to buy at least $50 million of OpenAI software and services through 2028, including ChatGPT Enterprise, which makes the tenant a vendor to its own landlord. Nvidia invested $1.5 billion in SB Energy in August and agreed to guarantee up to $105 billion supporting the Ohio lease, according to Reuters, which is targeting $5 billion to $7 billion in IPO proceeds.
One caveat matters. The figures come from draft filings described by the Journal, and neither OpenAI nor SoftBank has confirmed them. The public S-1 would settle them.
Why Would a Landlord Pay Its Tenant in a 1.4% Vacancy Market?
The short answer is that SB Energy was not selling space. It was selling a credit story to the public markets, and an anchor tenant of OpenAI's profile is what makes that story financeable. The concession was priced against the IPO, not against the rent roll.
Look at the market backdrop. CBRE's North America Data Center Trends H1 2026 put primary market vacancy at a record low 1.4%, with inventory up 33.7% year over year to 10,903 megawatts and 80.4% of all under construction capacity already preleased, up from 74.3% a year earlier. Less than 1,500 megawatts remains available across primary markets, roughly six months of demand at current absorption.
In a market that tight, a landlord with an operating building does not pay anyone to sign. But SB Energy has no completed data centers in operation and roughly 800 megawatts under construction. It is pre-revenue infrastructure asking public investors for a valuation above $50 billion. The warrants are best understood as a cost of capital, not a leasing cost. They convert an unbuilt campus into a bankable, 20-year contracted asset that lenders and equity buyers can underwrite.
That is a familiar move at an unfamiliar scale. Developers have always bought anchor tenants with free rent and oversized improvement allowances. Equity is the most expensive version of the same trade, and the only one that shows up nowhere in the rent roll.
What This Changes About Underwriting Contracted Backlog
Contracted backlog is now a gross number that needs a net calculation behind it. SB Energy reports a backlog above $400 billion, obtained in part by issuing billions in equity to the counterparty generating it. Backlog acquired with dilution is worth less per dollar than backlog acquired at arm's length, and nothing in the headline distinguishes the two.
Three underwriting adjustments follow for CRE investors looking at any AI-anchored asset:
- Net the inducement against the income. Cap rate is NOI divided by purchase price, and NOI is gross revenue minus operating expenses. It does not capture equity a sponsor issued to originate that revenue. A lease that pencils to a 7% cap rate on headline NOI is not really a 7% cap rate if billions in warrants produced it. Amortize the inducement across the term first.
- Test whether tenant credit is independent of the landlord. Standard practice is to underwrite an anchor lease against the tenant's balance sheet. That logic weakens when the tenant is also a shareholder, a warrant holder, and a vendor to the landlord. Ask what the lease is worth if the equity relationship is unwound.
- Separate guaranteed cash flow from contingent cash flow. A take-or-pay obligation backed by an investment grade guarantor is a different instrument than a 20-year commitment from a counterparty whose incentive to perform is partly equity-driven. Lenders sizing debt run DSCR, which is NOI divided by annual debt service, and they discount income they cannot isolate from the sponsor.
This is the same discipline we applied to the supply side in our analysis of whether the AI data center boom is a bubble. The bull case rests on contracted demand being real and durable. Structures like this one do not disprove that, but they do mean the contracts need to be read, not counted. CRE investors who want that discipline built into their own diligence process can reach out to Avi Hacker, J.D. at The AI Consulting Network.
What This Deal Does Not Mean
It is worth being precise about the limits of the read. This is not evidence that demand is soft. Vacancy at 1.4% and preleasing at 80.4% say the opposite. Nor is it a template that generalizes down-market. No multifamily or industrial sponsor is issuing warrants to a tenant, and the structure only works when one counterparty can validate an entire IPO narrative.
What it does signal is a shift in who holds pricing power at the top of the market. When the tenant is scarcer than the space, the tenant captures the upside. Investors watching land and power plays, a theme we examined in our look at Coatue's $5.7 billion data center land venture, should expect more of the economics to migrate from rent into equity, options, and revenue shares. Those terms sit in the documents, not the rent roll, and tools like Claude, ChatGPT, and Perplexity are useful for pulling inducement language out of a long lease file quickly. The AI Consulting Network specializes in exactly this.
Frequently Asked Questions
Q: Did OpenAI pay SB Energy or did SB Energy pay OpenAI?
A: SB Energy paid OpenAI. SB Energy is the issuer of the warrants and OpenAI is the recipient, which makes this a landlord compensating its anchor tenant rather than a customer buying a stake. OpenAI did separately invest about $500 million of cash into SB Energy, which is a different transaction.
Q: Why does a warrant grant matter more than a normal lease concession?
A: Because it is invisible in the rent roll. Free rent and improvement allowances show up in the net effective rent calculations every CRE investor already runs. Equity issued to a tenant does not, so an asset can look fully leased at strong rents while the sponsor has transferred billions in enterprise value to get there.
Q: Are these figures confirmed?
A: Not fully. They come from draft IPO documents reviewed by The Wall Street Journal and reported on August 30, 2026, and Reuters noted it could not independently verify the report. OpenAI and SoftBank have not confirmed the numbers. The public S-1 filing would be the authoritative source.
Q: Does this suggest an AI infrastructure bubble?
A: Not on its own. It suggests concentration risk rather than overbuilding. The concern is not that the space is empty, it is that a few counterparties anchor an enormous share of contracted capacity, and their commitments are increasingly entangled with the equity of the developers building for them. For guidance on stress testing that exposure, connect with The AI Consulting Network.