What actually kills a deal in diligence, and how early you can see it

Four hundred sixty-seven acres of Franklin Township farmland, a confidential land assembly, a rezoning petition that cleared the Metropolitan Development Commission 8-1, and then nothing. Deep Meadow Ventures LLC, which everyone eventually learned was Google, withdrew the roughly $1 billion Franklin Township data center petition on September 22, 2025, minutes before a City-County Council vote it was not going to survive.

Nobody involved in that project woke up on September 22 and discovered the opposition. Protect Franklin Township had been filing remonstrance since spring. The district councilor had been on record against the rezoning for months. A statewide advocacy group had already made Indiana utility bills into the opposition's talking point. The failure was not sudden. It was legible, in the public record, for most of a year before it killed the deal.

That is the pattern worth naming plainly: entitlement failure is not random noise. It clusters into a small number of repeatable categories, and every category we track leaves a tell in the public record well before the hearing that ends the project. I want to walk through the four that show up most often in our case files, in the order a screening team should actually check for them, because the tell for each one arrives on a different clock.

Policy risk arrives before the ordinance does

The first category is the one people underwrite the least: the jurisdiction itself changing the rules. On March 18, 2025, the Loudoun County Board of Supervisors eliminated by-right development for data centers, the zoning classification that had made Loudoun the densest data-center market in North America for two decades. New applications now require a special-exception hearing with mandatory public findings. Existing pipeline was grandfathered; anything filed after March 18 was not.

The tell here was not subtle. Virginia's Joint Legislative Audit and Review Commission published its data-center report in December 2024, examining tax benefits against energy, water, and infrastructure tradeoffs, and finding that localities had lost effective authority over aspects of data-center regulation. That report became, in the words of the trade press covering it, the most-cited document in the reform debate that followed. A team reading JLARC in December had three months of runway before the ordinance vote. A team that read the vote itself, in March, had none.

Fauquier County shows the same mechanism running to completion. The county adopted a formal data-center policy on December 14, 2023, one Virginia land-use counsel called among the strictest in Northern Virginia at the time. Eighteen months later, on June 18, 2025, the Planning Commission recommended denying the Gigaland campus 4-1, the first fully contested application reviewed under that policy. Nobody in that room should have been surprised. It was the policy doing exactly what the county said it would do in 2023, on a case that showed up in 2025.

Policy risk is not zoning risk in the ordinary sense of "is the use permitted." It is a bet on whether the rules will still say yes by the time you need them to. That bet has a visible price: read the board's own stated intentions, not just its current code.

Opposition risk is the one that compounds

The second category is the one Franklin Township illustrates directly, and it behaves differently from policy risk because it builds inside a single case rather than across a jurisdiction's history. Our own read on that site moved from 25 out of 100 at filing in March 2025, structural risk only, no organized opposition yet visible, to 2 out of 100 by August, once remonstrance was filed, a coalition had formed, and the district councilor had committed to blocking it. Same parcel. Same code. Five months apart. The delta is entirely the public record catching up to what the neighborhood already knew.

Organized opposition rarely announces itself at a podium for the first time on hearing night. It shows up first as a Facebook group, then as a named coalition, then as remonstrance filings with specific factual objections, then as councilor lobbying. Each of those stages is dated and, in most jurisdictions, public. A team that only checks the hearing agenda finds out at the last stage. A team that checks the record finds out at the first.

Utility and environmental risk share a mechanism, and it is not a hearing

The third and fourth categories are the ones a pure zoning read misses entirely, because the fight is not at the planning commission. It is at a state agency, and the opponent is not a neighborhood coalition. It is an organization with legal standing and a staff attorney.

The fight simply moves out of the room where your team is watching.

Hyundai's Metaplant in Bryan County, Georgia depends on a Georgia EPD permit authorizing withdrawal of up to roughly 6.6 million gallons a day from the Floridan Aquifer for the site, with the Metaplant's own expected draw closer to 4 million gallons a day. The Ogeechee Riverkeeper, a long-established Savannah-area nonprofit with its own counsel, appealed that permit in 2023. The appeal did not go to a jury. It went to a settlement, signed January 2025 and executed that April, that compressed the timeline for bringing an alternate, non-aquifer water source online from 25 years to 15, plus new monitoring obligations. EPD then fined the operator $30,000 for separate wastewater violations that same spring. None of this touched the zoning entitlement. All of it changed what the plant was allowed to do with water for the next decade and a half.

The Amazon Morrow County nitrate case is the same mechanism running on a longer clock. Oregon officials had acknowledged the Lower Umatilla Basin needed cleanup for more than 30 years before the Oregon Health Authority formally identified the crisis in 2024: at least 634 domestic wells testing above the federal nitrate limit, some close to ten times over. Amazon settled the resulting suit for $20.5 million in March 2026, denying it caused the contamination and saying it wanted to spend its time on the community rather than the litigation. Sixteen other defendants, mostly agricultural operators with decades of history in the basin, are still in the case. Hermiston annexed 800 acres to attract more data-center investment in 2025, in the same basin, with the litigation already pending. That decision was made with the risk fully on the public record, not hidden from anyone who checked.

Here is my bet, stated plainly enough to be wrong: utility and environmental risk will overtake opposition risk as the leading cause of data-center entitlement failure within three years, because the zoning fights are becoming better understood and the water and power fights are not. Every team I have watched screen a site checks the zoning code first. Almost none of them pull the state environmental agency's enforcement history or an NGO's litigation docket before they file. That gap is where the next Franklin Township is sitting right now.

The four categories, and the one thing they share

Policy risk, opposition risk, utility risk, and environmental risk look like different problems because they show up in different venues: a board meeting, a remonstrance filing, an agency docket, a state health department report. They are the same problem. Each one leaves a public antecedent that predates the failure by months or years, and each one is legible to anyone willing to read it before they file rather than after they are denied.

A cited site memo does not eliminate any of these four risks. What it does is put the record in front of a development team before the money is committed, so the read at month one looks like the read at month eight, not a surprise that arrives after the land is assembled and the attorneys are billed. Franklin Township's record was there in March. Loudoun's was there in December. Bryan County's was there in 2023. Morrow County's was there for thirty years.

The question for any team about to file is not whether one of these four will show up. One of them almost always does. The question is whether you read for it before the hearing or after the withdrawal.