The Take — SoftBank's $5.5B warrants prove AI infrastructure is funding itself
SoftBank just handed OpenAI $5.5 billion in stock warrants to stay in a data center lease, and the most important word in that sentence is "just." This is not a customer buying compute. This is a landlord paying its tenant to remain a tenant, then preparing to sell the resulting revenue stream to public market investors as if it were organic demand. The AI infrastructure boom is real. The financing underneath it is more circular than anyone with an IPO to file wants to admit.
The facts, as we covered them this morning in SoftBank hands OpenAI $5.5B in warrants to lock in Stargate, are straightforward. SB Energy — the SoftBank subsidiary building the data center campuses behind Stargate — is about to file for a US IPO targeting $5 billion to $7 billion in proceeds. Buried in those IPO documents is a disclosure: SB Energy issued warrants to OpenAI worth roughly $5.5 billion, giving the model company the right to buy SB Energy shares on favorable terms. OpenAI is SB Energy's biggest tenant. SB Energy is building campuses to OpenAI's specifications. And now OpenAI has a financial incentive, denominated in equity upside, to keep leasing from the one company that is building for it.
This is not a new pattern. It is the same pattern we have been tracking since Nvidia started using its balance sheet to lock in demand. When Nvidia backed a $500 billion data center deal with a GPU value guarantee, the move was framed as confidence in the AI build-out. When Nvidia then halved its OpenAI backstop from $250 billion to $120 billion after investor pushback, the move was framed as prudent risk management. Both reads were correct, and both missed the structural point: the companies building AI infrastructure are increasingly financing the demand for that infrastructure, and the circularity is accelerating, not stabilizing.
Here is the loop, stated plainly. SoftBank invests in OpenAI — cumulative exposure now past $64 billion. OpenAI needs compute, so it leases from SB Energy, a SoftBank subsidiary. SB Energy needs to show revenue for its IPO, so it sweetens the lease with warrants that give OpenAI equity upside in SB Energy itself. OpenAI's lease payments become SB Energy's revenue. SB Energy's revenue becomes the basis for its IPO valuation. Public market investors buy shares in a company whose biggest customer was paid to stay. The compute gets built. The models get trained. The infrastructure is real. But the demand signal that underwrites the IPO was partly manufactured by the entity selling the shares.
I want to be precise about what I am not saying. I am not saying the demand for AI compute is fake. OpenAI's inference costs are enormous and growing. The Stargate campuses will house real GPUs running real workloads. The 10-gigawatt Ohio facility that OpenAI signed a 20-year lease for is not a Potemkin village. What I am saying is that the financing structure around that demand is self-referential in a way that should make public market investors uncomfortable, because the warrants create a feedback loop where the tenant's incentive to keep leasing is denominated in the landlord's equity rather than in the tenant's actual compute needs.
The counter-case is real, and it deserves a fair hearing. First, circular financing is not inherently fraudulent. Airlines buy planes from manufacturers who finance the purchases through leasing arms that sell the leases to investors. The auto industry runs on captive finance companies. The structure works when the underlying asset — the plane, the car, the data center — produces real economic value independent of the financing. If OpenAI's models generate enough revenue to justify the compute spend, the warrants are just a retention device, and the IPO investors are buying into a genuine infrastructure play.
Second, SoftBank's track record with this kind of structure is not uniformly bad. The Vision Fund's early bets on Uber, DoorDash, and ARM all involved circular or quasi-circular financing, and two of those three produced real returns. Masayoshi Son has been wrong about timing and valuation, but he has not been wrong about the direction of compute demand. The thesis — own the model layer, own the physical layer, make the two unable to walk away from each other — is coherent even if the execution is aggressive.
Third, the alternative is worse. If the only way to fund a 10-gigawatt data center campus is pure debt or pure equity from parties with no strategic interest, the financing either does not happen or happens at terms so punitive that the compute never reaches the labs that need it. The AI build-out requires capital at a scale and speed that traditional infrastructure financing cannot match. Circular deals are the lubricant. Without them, the timeline for frontier model training stretches by years.
I hear all of that, and the take still holds. The reason is not that circular financing is always bad. The reason is that this particular loop is about to be tested by public market investors who are being asked to price SB Energy on the strength of demand that was partly manufactured with equity. The IPO prospectus will show revenue growth driven by OpenAI's lease payments. It will show a pipeline of future campuses justified by OpenAI's expansion plans. And it will show, in a footnote, that the biggest tenant received $5.5 billion in warrants to stay. A sophisticated investor can read that footnote and still buy the story. But the footnote is the story, and the fact that it is buried rather than highlighted tells you what the underwriters think about how the market will read it.
The broader pattern is what matters. Nvidia's GPU value guarantees, Nvidia's halved backstop, SoftBank's warrants, the SB Energy IPO, the Stargate lease structures — these are all pieces of the same financing architecture, and each piece creates an incentive for the next piece to exist. Nvidia guarantees GPU value to lock in demand. SoftBank builds campuses to capture that demand. OpenAI signs leases to access the compute. The warrants keep OpenAI in the lease. The IPO monetizes the lease. And the cycle starts again with the next campus, the next lease, the next set of warrants. The system is not a Ponzi scheme — real value is being created — but it is a system where the participants are each other's customers, each other's investors, and each other's reasons to keep building.
What would change my mind is straightforward. If SB Energy's IPO prices at the low end of its range and trades flat or down in the first six months, the public market is telling the circularity story back to the participants, and the loop tightens. If OpenAI's revenue growth continues to outpace its compute costs — if the models actually pay for the infrastructure without the financing sweeteners — then the warrants are just smart retention and the circularity is cosmetic. And if a major investor in the IPO discloses that it modeled the warrants as a material risk and bought anyway, that is a market signal worth more than any editorial take.
Today, the signal is the opposite. Nvidia's own shareholders capped the company's OpenAI exposure at half the headline number. The debt markets balked at the original Stargate financing structure. And now SoftBank is sweetening a lease with warrants that it will not need if the demand is as real as the IPO prospectus claims. The infrastructure is real. The financing is circular. And the public markets are about to be asked to pretend those two facts are unrelated.
When a landlord pays its tenant to stay, who is the real customer — and who is the product? Tell us in the comments.
Sources: Reuters · Wall Street Journal · Nvidia backs $500B data center deal with GPU value guarantee · Nvidia halves OpenAI's $250B data center backstop to $120B · Techmeme
