Jim Costello's 4.5 percent line: what a cost-of-capital inflection means for a data-center pro forma

Four and a half percent.

Jim Costello, chief economist at MSCI Real Assets and before that at Real Capital Analytics, put that number at the center of a 2025 research note titled "Real Assets in Focus: The Impact of Crossing the 4.5% Treasury Threshold." His argument, compressed: roughly four decades of cap-rate compression were powered by progressively cheaper capital, borrowed money subsidized returns whether or not the property earned them, and once the 10-year Treasury holds above 4.5 percent that subsidy is gone. What remains is the asset. Costello's prescription is a return to fundamentals: active asset management, disciplined capex, smart leasing, and a cost of capital priced off the market that exists rather than the Treasury-spread habit everyone formed during the cheap decade.

Most people read that as advice about office towers.

I think its sharpest application is the data-center pro forma, because in this vertical "fundamentals" does not primarily mean the roof and the rent roll. The fundamentals of a data-center site are settled before schematic design: the power path, the water design, the approval mechanics, and the local political posture. Two RealClear case files, same product type, overlapping years, sit 23 points apart on exactly those variables.

Abilene bought its fundamentals up front

The Stargate flagship campus in Abilene, Texas is what fundamentals-first underwriting looks like when it is done on purpose. Crusoe, Blue Owl Capital, and Primary Digital Infrastructure closed an $11.6 billion financing package on May 22, 2025, the second phase of a $15 billion joint venture, for a 1.2 GW campus across eight buildings. Construction had begun in June 2024. Phase 1 went live on Oracle Cloud Infrastructure on September 30, 2025.

Look at what was de-risked structurally rather than argued politically. The site sits on by-right Taylor County industrial land inside a Lancium Clean Campus, so the entitlement question that consumes other projects never opened. The cooling is closed-loop, direct-to-chip liquid cooling with zero water evaporation, which removed the water argument that, per the case file, killed data-center projects in San Marcos, Tucson, and Chesapeake. West Texas generates more wind power than the region consumes, so 1.2 GW of load landed next to surplus generation instead of at the back of a transmission queue. The Abilene Industrial Foundation was on record supporting data-center siting before anyone showed up with a term sheet.

Then came the test. In early 2026, Oracle and OpenAI dropped plans to expand the flagship site. Under old-regime math that is the nightmare scenario for a single-purpose asset. What actually happened: Microsoft picked up two adjacent buildings and an on-site power plant, and Meta entered talks, with NVIDIA participation, to absorb additional capacity. The demand stack reshuffled and the asset held, because the thing being underwritten was never one tenant's forecast. It was a power-matched and politically clean piece of infrastructure, fully financed, that any hyperscaler could use.

RealClear's cited read on the site: 82/100 before the first pad was poured.

Saline paid for its fundamentals in court

Now the other file. In Saline Township, Michigan, Related Digital filed to rezone roughly 575 acres of A-1 agricultural farmland for a Stargate campus. The township board denied the request 4-1 on September 10, 2025. Two days later, RD Michigan Property Owner I LLC and landowners sued in Washtenaw County Circuit Court. By October 1 the board had voted 4-1 to pursue settlement, and on October 15 a consent judgment authorized the project subject to a $14 million community package ($7 million for the Saline Area Fire Department, a $4 million farmland trust, a $2 million community fund, $500,000 each for two neighboring fire departments), a 55-decibel noise limit, Michigan Avenue access only, no evaporative cooling, no material expansion, and decommissioning security.

That resolved the zoning. It did not resolve the deal. The power ran on its own clock: the Michigan Public Service Commission approved DTE's special contracts 3-0 on December 18, 2025, at 1,383 MW, with a 19-year minimum term, an 80 percent minimum billing demand, customer-funded storage, and quarterly reporting. Attorney General Dana Nessel moved to reopen that approval on February 5, 2026. A resident's post-settlement intervention was denied on February 20. Township updates showed site work underway by March 18, 2026, with the Compute 1 pylons finished and roughly 60 percent of gravel hauling complete.

Six months from lawsuit to dirt-moving is fast. It is also expensive in exactly the way Costello's framework prices. The case file's decision framework says to budget 0.15 to 0.25 percent of project value for community concessions on this pattern; against the $7 billion project value that is a band of roughly $10.5 million to $17.5 million, and the actual package landed at $14 million, close to dead center. The concessions were never optional. They were a knowable cost, discovered through litigation instead of priced at the LOI.

What 4.5 percent does to that spread

Under the old math, Saline's path was survivable bordering on invisible. When debt was cheap and exit cap rates compressed on schedule, a $14 million package on a $7 billion project was roughly 20 basis points, and six months of litigation was a footnote the refinancing forgave. The cheap decade did not reward teams for telling the difference between Abilene and Saline. It paid both the same.

Above 4.5 percent, the two files stop rhyming. Every month between land control and energization is carried at a real cost of capital, so the six-to-twelve-month litigation detour the Saline file describes becomes a priced line item instead of a war story. The concession band becomes a required reserve, modeled on day one. The parallel utility docket, with an attorney general willing to reopen it, becomes schedule variance a lender will ask about by name. Costello's point, translated into this vertical: when capital no longer subsidizes the deal, the pro forma has to be honest about what the capital was hiding, and in data centers what it was hiding was mostly entitlement drag.

Cost-of-capital diligence and entitlement diligence used to be separate workstreams.

They are now the same exercise. The discount rate tells you what a month costs. The public record tells you how many months the jurisdiction will take, and both numbers were knowable in advance in both files. Abilene's by-right posture and water design were public before financial close. Saline's A-1 zoning, its five-member board, its rural-preservation politics, and its separate MPSC track were all visible before the first filing.

My read is that data-center land in litigation-path jurisdictions is still being bid as if the cheap-decade math applied, and that the spread between an Abilene-grade site and a Saline-grade site is wider than current pricing admits. If the 10-year holds above Costello's line through 2027, I expect that spread to show up in land basis, publicly and rudely, within two acquisition cycles. Check the pro forma on your desk right now. If entitlement drag appears nowhere in it as a carried cost, you are still underwriting with 2019 money.