Analysis. On July 14, Governor Kathy Hochul signed Executive Order 62, freezing state-level environmental permitting for large data centers in New York. Fifteen days later, Baird downgraded Caterpillar and cut its price target by a quarter, citing that order by name.
That sequence — a state environmental action showing up in an equipment analyst’s model within a fortnight — is the clearest evidence yet that the binding constraint on the AI infrastructure buildout is not chips, and increasingly not even power. It is permission.
What the Order Actually Does
The scope is narrower than “New York banned data centers,” and the details matter.
EO 62 directs the Department of Environmental Conservation to hold in abeyance all applications for any discretionary permit, approval, licence or similar permission for the construction or expansion of a data center — specifically, applications that were pending and had not been deemed complete before the order’s date. The threshold is data centers consuming 50 megawatts or more. Local government permits are explicitly excluded, so a municipality can still approve what the state has paused.
The freeze runs until the Department of Public Service submits a final Generic Environmental Impact Statement and associated findings. DPS must run a formal public process to produce it, assessing energy demand, water use and quality, air quality, disproportionate impacts on disadvantaged communities, and noise. There is no fixed end date — the moratorium lasts as long as the study does.
The order cites the pressure behind it: nearly 12 gigawatts of data center load requests sitting in the New York Independent System Operator interconnection queue, with more than eight gigawatts entering in 2025 alone.
The Fifteen-Day Transmission
On July 29, Baird cut Caterpillar from Outperform to Neutral and reduced its price target from $1,200 to $900. CNBC’s headline that day: “Caterpillar is downgraded on growing data center buildout backlash. The shares are falling.”
Baird’s reasoning, as reported, is careful and worth reproducing accurately, because it is not a bearish call on AI. The firm expects strong orders, backlog and earnings momentum through the second half of 2026, and explicitly does not forecast an “AI bust.” The concern is 2027 and 2028: that tighter regulation, permitting hurdles, reduced site availability and higher development costs temper hyperscaler capital spending growth, and with it demand for Caterpillar’s power generation equipment.
The scale of the regulatory shift is the part most people have missed. Per Baird’s count, more than 40 states considered over 260 data-center-related bills in 2025 — and more than 300 further proposals were filed across 30 states in the first quarter of 2026 alone. The policy posture has flipped from competing to attract data centers with incentives to constraining them with restrictions, taxes and oversight. New York is not an outlier. It is the most legible instance of a trend already well underway.
Why This Lands on Construction
Three consequences follow, and they are more concrete than the stock move.
The pipeline is a permitting queue, not a demand queue. For contractors whose backlogs have been built on data center work, the risk has changed shape. Demand is not softening; the ability to break ground on it is being rationed by state agencies conducting environmental reviews with no statutory deadline. That is a fundamentally different planning problem from a demand slowdown, and it rewards different behaviour — optionality across geographies rather than capacity concentration.
The 50 MW threshold shapes what gets built. A permitting regime with a bright line at 50 MW creates an obvious incentive to design beneath it, or to distribute load across sites. Whether that produces genuinely smaller facilities or merely creative parcelling is a question for regulators, but it will show up in project mix.
Ground truth about site conditions becomes commercially valuable. When permission is scarce and reviews assess water, air, noise and community impact, the ability to document actual site conditions rigorously stops being a compliance chore and becomes part of getting permission at all. That is the quiet upside for the reality-capture and monitoring category — the same category Procore just paid $845 million to own, and the one OpenSpace’s 1,000-data-center milestone mapped.
The Uncomfortable Reading
There is a version of this story that is genuinely bad for construction technology, and it deserves stating rather than burying.
A great deal of the sector’s recent growth has been indexed to a single end market. Data centers have been the demand story behind the equipment autonomy companies, the reality-capture platforms and a good share of the specialty trade backlog. A regulatory brake applied simultaneously in dozens of states is a correlated risk across all of it — one that does not care how good the software is.
The counter-argument is that scarcity of permission raises the value of executing flawlessly on the projects that do get approved, which is precisely what the tooling sells. Both readings can be true, and which dominates depends on something no vendor controls: how long DPS takes.
What to Watch
The GEIS timeline is the whole ballgame in New York, and it has no deadline. A study that concludes in months is a speed bump; one that runs past 2027 is a redirection of several gigawatts of demand to other states.
Nationally, watch whether the Q1 2026 wave of 300-plus proposals converts into enacted law or dies in committee — considered bills and passed bills are very different animals, and Baird is pricing a trend line off the former. And watch Caterpillar’s own guidance rather than the analyst note: the company reports Q2 on August 4, which is the first chance to hear whether the order book is actually reflecting any of this yet.
