Speculative Data Center Development: How to Underwrite Without a Signed Lease
Building ahead of tenants is no longer unusual in data center development. The question is whether the underwriting holds when power and schedule risk are baked in.
The speculative data center build has become a defining feature of the AI infrastructure cycle. Developers are breaking ground on powered shells and campus infrastructure before hyperscale tenants sign. Some are doing it without confirmed grid interconnection. The economics, when they work, are substantial. When they do not, the capital impairment is equally substantial.
Understanding what drives the decision to build spec, how capital is structured around it, and where the models break is now core knowledge for any institutional developer active in digital infrastructure.
Why Developers Build Without Leases
The logic is a function of the supply-demand structure, not recklessness.
Power access has become the real constraint. Average grid interconnection wait times in primary US data center markets now exceed four years, according to JLL's 2026 Global Data Center Outlook. A developer who waits for a signed lease before entering the interconnection queue will be years behind a competitor who started the process speculatively. The lease follows the power, not the other way around.
Hyperscalers want capacity that is ready or nearly ready. Amazon, Microsoft, Google, and Meta together are projected to spend over 00 billion on data center build-out in 2026. These companies prefer to lease powered shells or wholesale capacity where delivery risk is already absorbed. That preference makes it economically rational for developers to build ahead and absorb the risk themselves.
Speed to power is the primary site selection variable. JLL data frames speed to power as the top criterion for hyperscale and AI tenant decision-making in 2026, ahead of construction cost, latency, and proximity to customers. A developer with available capacity on a powered site wins deals that a developer with a better raw site on a four-year interconnection queue cannot.
What the Pipeline Actually Looks Like
Sightline Climate tracks 190 GW across 777 large data center and AI factory projects of 50 MW or more announced since 2024. Of the roughly 16 GW slated for delivery in 2026, only about 5 GW is actually under construction. The remaining 11 GW is at announcement stage despite typical 12 to 18 month build timelines. Based on 2025 slippage patterns, Sightline estimates 30 to 50 percent of the 2026 pipeline will not come online on schedule.
This gap between announced and deliverable is the defining risk of the speculative build model. Capital committed to announced projects that slip runs in the tens of billions.
How Capital Structures Handle Pre-Lease Risk
Lenders and equity providers have adapted their underwriting, but the adjustment is calibrated to sponsor quality and power certainty, not to the absence of a lease per se.
Equity requirements are higher. Shell-and-core construction costs have risen to approximately 1.3 million per MW in 2026, up from .7 million per MW in 2020, according to JLL. At these levels, even a 100 MW speculative phase implies roughly 65 million in shell costs before any tenant fit-out. Lenders require meaningful equity from institutional infrastructure funds, listed data center REITs, or private platforms backed by sovereign or pension capital.
Construction debt is underwritten to power and sponsor strength. Banks and private credit providers will proceed without a signed lease if interconnection is contracted or on-site power is committed, the sponsor has a demonstrable track record of hyperscale leasing, and there is credible line-of-sight to a long-term colocation agreement or PPA.
Mini-perm structures with lease-up covenants are standard. Typical construction debt carries three to seven year tenors with the expectation of refinancing into long-term debt after leasing. Deals often include triggers for additional equity, amortization step-ups, or cash traps if lease-up lags the underwritten schedule.
Private credit is filling the gap. JLL estimates approximately 70 billion of new debt will be required for data center real estate through 2030. Traditional bank construction lending is not the primary source. Infrastructure debt platforms and private credit funds are taking pre-lease risk at higher spreads in exchange for exposure to the structural demand thesis.
Where the Speculative Model Breaks
Three failure modes are well-documented.
Phantom interconnection. Utilities and ISOs report speculative grid reservations from projects that may never build, clogging queues and inflating apparent demand figures. Developers who enter the queue without genuine build-readiness are consuming capacity that constrains legitimate projects and creating a systemic mismatch between reported pipeline and deliverable supply.
Community and regulatory ambush. As of June 2026, 4 billion in US data center projects have been blocked or delayed by community opposition and regulatory intervention since mid-2024. A speculative project caught inside a new state-level moratorium zone, with capital deployed and an interconnection position aging, faces costly options.
Construction cost overrun against fixed lease economics. Shell-and-core costs have risen roughly 40 percent per MW since 2020 and are projected to increase a further six percent in 2026. If lease pricing does not keep pace with cost escalation, the development spread compresses and returns fall below underwriting.
The Projects That Illustrate the Model
Applied Digital's Delta Forge 1 campus in Boyce, Louisiana represents the spec build at scale: .6 billion, 300 acres, announced as AI campus infrastructure ahead of named tenants. CloudBurst Data Centers broke ground on a 1.2 GW campus in Central Texas without disclosed pre-leasing. The Stratos project in Utah, a proposed 9 GW AI campus backed by investors including Kevin O'Leary, is described as one of the most ambitious and scrutinized projects of the current cycle given its grid implications.
On the industrial conversion side, Terawulf's redevelopment of the former Hawesville, Kentucky aluminum smelter illustrates how legacy heavy-industrial sites with robust power infrastructure are being repositioned speculatively, with the power thesis as the primary investment rationale.
The Underwriting Framework
For institutional developers evaluating a speculative build, the key variables in order of priority are:
Power certainty first. Interconnection study progress, substation headroom, on-site generation feasibility, and utility engagement quality are the primary underwriting gates, not the lease pipeline.
Market absorption rate and competitive supply. Speculative builds work in markets where hyperscale demand is documented and competing deliverable supply is constrained. They do not work in markets where both conditions are absent.
Regulatory and community risk at the site and jurisdiction level. This belongs in the first diligence pass. A site in a jurisdiction trending toward moratorium is already impaired.
Sponsor track record with hyperscale tenants. Capital providers price this heavily. A developer with a history of executed hyperscale leases in similar product types commands different terms than a new entrant, regardless of site quality.
Cost-to-deliver versus market rent at delivery. The 2026 construction cost environment makes the spread narrower than it was in 2022 or 2023. Underwriting needs to stress the delivery cost and lease rate assumptions together, not independently.
The speculative data center build is not a new phenomenon. What is different in 2026 is the capital scale, the regulatory environment, and the execution complexity. Developers who underwrite all three rigorously are building positions that are defensible. Those who rely on the demand macro to solve execution problems are taking a different kind of risk.