Anchor tenants split the digital infrastructure stack

A six-year GPU capacity lease at Maysville, Georgia, valued at $13.7 billion, includes warrants for up to 50.81 million shares at $0.01 each. That is the price an unnamed cloud customer demanded for taking buildout risk on a data center campus. The warrant detail is the clearest sign yet that the digital infrastructure capital stack has split along anchor-tenant lines: customers with enough demand to anchor a campus can now extract equity, while platforms without such an anchor must settle for debt, minority capital, or both.

PWD's tracking shows the two poles forming in the same quarter. At one end, Anthropic's reported $45 billion, six-year, 460-megawatt lease with Nscale prices a $51 billion contracted-revenue backlog before the neocloud's September IPO. At the other, Eurofiber locks in €2.2 billion of sustainability-linked debt, replacing a €1.5 billion platform and testing whether open digital infrastructure can hold infrastructure pricing without a hyperscaler anchor. The gap between those two transactions is not just size; it is the price of having a named tenant.

Rum's lease terms are the most explicit form of anchor-tenant pricing. The six-year deal for GPU capacity at Maysville includes warrants for up to 50.81 million shares at one cent, which means the customer is being compensated in equity for committing to a facility that still has to be built and powered. No underwriter would propose such a structure to a diversified customer base; it only works when one counterparty is large enough to make or break the project. The unnamed customer has effectively taken an option on Rum's equity as the price of providing demand certainty, and that option sits at the top of the capital stack.

The one-cent warrant is not new in venture capital, but it is new in infrastructure capital. Data center leases have traditionally been fixed-price contracts with rent escalators; now the anchor tenant can demand a share of the developer's equity upside as a condition of signing. That transfers value from the equity stack to the tenant at the exact moment the developer needs to raise construction debt, and it means the developer's cost of capital includes a dilution component that never appears in the loan documents.

Nscale's reported lease with Anthropic is the opposite side of the same coin. A single tenant signing for 460 megawatts over six years gives Nscale a contracted-revenue backlog of $51 billion, a number that will anchor its September IPO. The lease does not just fill capacity; it re-rates the developer from a speculative builder to a contracted infrastructure owner. EdgeConneX's $4.2 billion of Texas data center filings show the same dynamic at work. The construction plans are credible because CoreWeave is set to lease at least part of the campus, which means the capex has an offtake before the concrete is poured. In both cases, the tenant is the underwriting asset; the building is secondary.

This is not to say anchored assets are risk-free. A $51 billion backlog concentrated in one tenant is a credit concentration risk of the first order, and a lease with a single AI lab does not guarantee that the tenant's own business model survives the next six years. But capital markets are clearly treating such concentration as a feature, not a bug. The Nscale IPO will test whether public investors accept a backlog anchored by one customer, and whether the infrastructure multiple survives the loss of diversification.

Eurofiber's €2.2 billion sustainability-linked debt is the test from the other side. The larger facility replaces a €1.5 billion platform and asks whether open digital infrastructure—fiber networks with no single hyperscaler customer—can still hold infrastructure pricing without an anchor. That the debt is sustainability-linked rather than asset-backed suggests lenders are pricing Eurofiber on its corporate ESG targets and cash flows, not on a contracted tenant lease. The debt is larger, but the pricing signal is softer: the lender is not underwriting a tenant, it is underwriting a business model.

MTN's minority stake in Africa Data Hub Holding Limited tells the same story in equity form. The UAE-backed platform targeting up to 150 megawatts in South Africa and Nigeria sells a slice of itself to a strategic partner because it cannot yet point to a contracted offtaker. The deal settles the equity question but leaves power procurement unresolved, which means the asset is still a development option rather than operating infrastructure. MTN has bought a minority stake in the option, not a controlling interest in a cash-flowing data center. That distinction is the whole story.

Even mid-market construction finance now leans on anchor-tenant track records. E.Sun Bank closed a construction loan for Empyrion's 7-megawatt Taipei colocation building, citing the lender's role in AirTrunk's $1.2 billion Tokyo financing. The 7MW deal is small, but the underwriting logic is large: the bank is lending against a sponsor that has financed hyperscale-tenanted assets before, not against the Neihu building's own lease book. The loan would not have been written on the strength of a 7MW colocation market; it was written on the memory of a $1.2 billion hyperscale deal.

The same split is now visible inside a single operator's corporate structure. SK Telecom has moved eight operating South Korean data centers into SK Horizon, a company 49 percent owned by KKR and IMM, while gigawatt-scale development stays inside SK Hyper. That is a clean separation: stabilized, operating assets can be financed as infrastructure, while greenfield AI capacity remains a development risk. The private equity money went to the stabilized side, not the gigawatt pipeline, which tells you where infrastructure capital will and will not go today.

The Africa Data Hub Holding situation is instructive. The platform targets up to 150 megawatts, but the equity stake does not solve the hardest problem: where the power comes from. South Africa and Nigeria both have grid constraints and unreliable utility supply, so the data center's real value depends on securing power, not on closing an equity round. The minority stake is a placeholder; the anchor tenant, if one ever appears, will demand the same equity kicker Rum gave away, and the current investors will be diluted again.

The consequences for the capital stack are already visible. A named anchor tenant such as Anthropic or CoreWeave turns a data center into a contracted infrastructure asset, priced at the low end of the cost of capital. A platform without an anchor—Eurofiber, MTN, Empyrion's smaller Neihu project—must pay for debt with ESG targets or a bank's memory, or sell minority stakes to strategic partners. And an anchor tenant with enough demand can demand warrants at one cent. The split is not a temporary dislocation; it is the new underwriting standard.

For family offices and private wealth allocators, the implication is that digital infrastructure funds are no longer a homogeneous asset class. A fund holding Nscale-type leases is exposed to hyperscaler credit risk and construction risk; a fund holding Eurofiber-style fiber is exposed to ESG-linked covenants and retail cash flows. The split means due diligence must start with the tenant ledger, not the megawatts. A 150-megawatt campus in Africa with no anchor is not the same risk as a 460-megawatt campus leased to Anthropic, and the price should not pretend otherwise.

The market will eventually resolve the split, but for now it is widening. Anchor tenants are learning that they can extract equity for their demand certainty, which raises the cost of capital for every platform without one. The only way non-anchored platforms close the gap is to find an anchor, merge into an anchored platform, or accept that they are no longer infrastructure—they are development companies with infrastructure aspirations.