Trumbull AI Data Center Plan Advances Despite a Two-Year Moratorium
The Trumbull AI data center plan remains active despite a new two-year moratorium because the rule applies only to new facilities. The proposal would modernize an operating data center at 80 Merritt Boulevard instead of constructing another building elsewhere in town.
That distinction has become the center of a local fight. Trumbull has paused the establishment and development of new data centers while allowing an existing facility to pursue an AI-oriented conversion.
365 Data Centers and property owner Aphorio Carter are evaluating the site for high-density computing. Their plans place Trumbull near the front of Connecticut’s emerging debate over artificial intelligence infrastructure.
The companies describe the project as an upgrade inside a decades-old facility. Residents see a more fundamental change, one that could introduce heavier computing equipment and different demands on local infrastructure.
Town officials say the operators have accepted limits covering electricity, water, noise, and the building footprint. However, officials are still determining how those commitments would remain enforceable if the operation later sought more capacity.
That is the central tension. Trumbull has adopted a broad pause for future projects, yet its most immediate AI infrastructure decision sits outside that protection.
What Trumbull Approved and Excluded
Trumbull stopped new data centers for two years, but it did not stop the proposed conversion of its only existing facility.
The Town Council approved the moratorium on September 9, 2026. Resolution TC31-65 covers the “establishment and development” of new data centers within Trumbull.
The language in the official council resolution is short. It does not explicitly regulate modernization, equipment replacement, new tenants, or changes in the workloads processed inside an operating building.
That omission matters because the facility at 80 Merritt Boulevard already functions as a data center. It has operated for more than 50 years and previously belonged to Digital Realty.
Aphorio Carter now owns the property. 365 Data Centers is the prospective long-term operator, according to information released by the town.
The companies announced a broader partnership in May 2026. Their AI infrastructure pipeline targets roughly 200 megawatts across several United States markets.
The initial portfolio includes locations in Colorado, Kentucky, Virginia, Ohio, and Connecticut. The companies said they were evaluating Trumbull as part of that phased expansion.
Their strategy focuses partly on converting existing properties into facilities that can support denser computing systems. This can shorten development because the sites already have buildings, utility connections, and some operating infrastructure.
In Trumbull, that strategy also changes the regulatory question. The proposal does not need to present itself as a new data center when the property has served that function for decades.
The intended workload is still different. 365 Data Centers and Aphorio Carter have described their planned facilities as supporting high-density, AI-ready computing.
AI-ready generally means a facility can supply and cool racks containing more computing hardware than a conventional enterprise deployment. Higher rack density concentrates electrical demand and heat within less physical space.
The companies said their broader portfolio could support cabinet densities ranging from 50 kilowatts to more than 200 kilowatts. Those figures describe the portfolio, not a confirmed specification for the Trumbull location.
No final public design establishes the number of cabinets, their density, or the amount of AI computing planned for Merritt Boulevard. The Trumbull proposal also remains subject to final agreements and approvals.
Connecticut Public requested the letter of intent concerning the site. A company representative declined to release it, calling the document confidential and nonbinding.
The representative also provided no timetable for a fuller public announcement. That leaves residents evaluating a proposed transformation without the complete operating plan.
Trumbull’s moratorium therefore creates a clear boundary but not a complete answer. A new building faces a two-year pause, while a new computing purpose inside an old building does not.
Why the Trumbull AI Data Center Matters Beyond Town
The immediate pressure falls on local officials who must regulate an evolving operation before its final technical scope becomes public.
Trumbull is not choosing between a data center and an empty parcel. Officials are deciding how much operational change can occur before an existing facility effectively becomes a different infrastructure project.
The town describes the proposal as a modernization. First Selectman Vicki Tesoro has emphasized that the building is already zoned, powered, operating, and included in the local tax base.
That position has practical appeal. Reusing an industrial property can avoid clearing a new site, constructing a larger campus, or creating an entirely new utility connection.
The facility also serves existing customers. According to Tesoro, about 36% of its current customers are Connecticut businesses, including financial and telecommunications companies.
A modernization could preserve those services while adapting the building for newer computing needs. The prospective operator has proposed a 15-year commitment to the site, according to the town.
Officials also expect a substantial private investment. Tesoro said the applicant plans to invest more than $100 million and has agreed not to appeal the resulting property assessment.
Those figures come from the town’s account, not a final public contract. They should therefore be treated as proposed commitments rather than completed economic benefits.
Residents face a different calculation. The facility stands near other businesses, a daycare center, and a wooded area, according to local reporting.
For nearby households and organizations, the relevant questions concern noise, generators, cooling systems, water, and electrical infrastructure. Those effects depend on operations, not simply the age of the building.
A facility can retain its footprint while changing the intensity of activity inside it. New servers can draw more power per rack and produce more heat without requiring a larger exterior structure.
That makes the Trumbull dispute relevant to other municipalities. Many zoning codes classify buildings by broad uses rather than the density of computing inside them.
An older colocation facility, where customers rent space for their own equipment, can gradually adopt hardware suited to AI training or inference. Inference is the process of running a trained model to produce an answer or prediction.
That transition can happen through equipment and cooling upgrades rather than a new campus application. A rule limited to new construction might never address the operational change directly.
Connecticut communities are already testing different versions of temporary restrictions. Morris adopted a two-year pause, while Groton and West Haven adopted one-year moratoriums, according to the original report.
Those pauses give towns time to examine zoning, utility capacity, emergency planning, noise, and environmental oversight. They do not automatically resolve how existing data centers should be treated.
Trumbull is now the clearest example of that gap. It has both an active modernization proposal and a moratorium that excludes the property most relevant to the debate.
The project also arrives as Connecticut reconsiders its earlier strategy toward the industry. State lawmakers enacted broad data center incentives in 2021 to attract large investments and financial computing operations.
That policy offered long-term tax exemptions to qualifying projects. By 2026, lawmakers were considering legislation that would eliminate parts of those incentives.
The change in political attention reflects a broader shift. Data centers are no longer discussed only as economic development projects with large capital budgets.
They are increasingly evaluated as major electricity customers with local environmental and infrastructure consequences. AI demand has intensified that scrutiny because AI processors can concentrate much more computing into each rack.
For developers, reusing an existing site offers speed. For municipalities, the same advantage reduces the time available to create rules before modernization begins.
The Moratorium Stops New Sites, Not a New Workload
Trumbull’s policy controls where a new facility can appear, while the current conflict concerns what an existing facility can become.
This is a tradeoff between asset reuse and regulatory visibility. Each side of that tradeoff rests on a reasonable argument.
Aphorio Carter specializes in acquiring and modernizing existing data center properties. 365 Data Centers supplies operating experience, connectivity, cloud services, and colocation capacity.
Their model can return underused infrastructure to productive operation. It may also deliver computing capacity faster than building a hyperscale campus from the ground up.
The Trumbull facility is not expected to become a hyperscale data center. Hyperscale sites are very large campuses designed to support extensive cloud or AI operations across multiple buildings.
Kenneth Gillingham, a Yale environmental and energy economics professor, estimated the existing Trumbull capacity at 15 megawatts. He compared that level with the combined demand of three large hospitals.
A 15-megawatt facility is small beside the largest AI campuses. It is still a substantial and continuous industrial load within one municipality.
The companies have not publicly confirmed the final operating capacity for the converted site. They have also not disclosed a tenant list for future AI workloads.
That distinction limits what can be concluded today. “AI-ready” describes infrastructure capability, but it does not prove that a specific AI company has committed to occupy the facility.
The broader partnership announcement describes liquid-to-chip infrastructure. This cooling approach moves liquid close to high-temperature processors rather than relying entirely on room-level air cooling.
Liquid cooling can support denser hardware and transfer heat more efficiently. Its presence would signal that the facility is being designed for equipment beyond many traditional enterprise racks.
However, the companies have not published a final cooling design for Trumbull. Town officials instead say the property will move from open-loop evaporative cooling to a closed-loop system.
A closed-loop system recirculates cooling fluid rather than continually consuming an equivalent volume of fresh water. That can reduce operational water demand, depending on the final design.
The town says modernization will not increase water consumption. It also says no additional utility capacity, new substation, or greater grid draw is being requested.
Those conditions help distinguish the proposal from a large greenfield AI campus. A greenfield project places new construction on previously undeveloped or newly prepared land.
Yet the proposed limits also show why the workload matters. Officials would not need detailed assurances about power, cooling, and sound if the modernization were merely cosmetic.
The equipment inside a data center determines much of its environmental impact. A zoning label alone cannot reveal rack density, peak demand, backup generation, or cooling requirements.
This is where the moratorium’s wording becomes important. It freezes a category of new development without creating performance standards for existing operations.
A performance standard would regulate measurable outcomes, such as electrical demand, water consumption, sound levels, generator schedules, or building expansion.
Such standards can apply even when a property’s legal use does not change. They can also establish what happens if an operator exceeds a defined limit.
Trumbull has not publicly completed that enforcement framework. Officials are still examining how to make the four commitments binding if the facility later seeks to expand.
Residents therefore question whether the moratorium gives the town meaningful time to study AI infrastructure. The pause covers future applicants, but the current applicant can keep moving.
Sherry Boyd, a Trumbull resident of 33 years, summarized that concern during the council meeting. She said the two-year learning period does not protect residents from what is happening now.
Her objection is not simply opposition to technology. It concerns the timing and coverage of the policy response.
If the existing site completes its conversion before new standards arrive, it may become the precedent those standards were supposed to prepare for. That would reverse the intended order of review and development.
Four Promises Still Need Enforceable Limits
The project’s credibility now depends less on its stated limits than on who measures them, records them, and responds when they are exceeded.
Tesoro has identified four commitments from the applicant. They cover electricity, noise, water use, and the property’s physical footprint.
The first commitment says electrical draw will not rise beyond the site’s existing capacity. The town says existing transformers will be reconfigured, with no new substation or additional grid draw.
That claim needs a clear baseline. “Existing capacity” can describe a utility connection, transformer rating, historical peak, contracted allocation, or another technical measurement.
Those values are not necessarily identical. A facility operating below its connection limit could increase actual consumption without exceeding its theoretical capacity.
Public reporting does not establish which baseline Trumbull will use. It also does not identify how often demand would be measured or whether peak and average consumption receive separate treatment.
The second commitment covers noise. The town says backup generators will run briefly once each month and for two hours once annually.
Generator testing is only one potential source. Cooling equipment, fans, pumps, transformers, and loading operations can also affect nearby sound levels.
Tesoro has directed town staff to verify noise independently as the design becomes final. That verification will matter only if it uses a documented baseline and repeatable measurement method.
The third promise covers water. The proposed closed-loop cooling system is intended to prevent consumption from increasing during modernization.
That is a useful design commitment, but it still requires a defined accounting boundary. The public needs to know whether the comparison includes cooling, maintenance, domestic use, and unusual operating conditions.
The fourth promise prevents expansion of the building footprint. One of three buildings on the property is already being demolished, according to the town.
Tesoro says the completed work will reduce the total building area by 23%. That reduction does not necessarily reduce computing density inside the remaining structures.
A smaller building can host a more concentrated electrical load. Floor area therefore offers an incomplete proxy for operational impact.
These issues do not mean the commitments are false. They show why voluntary promises and enforceable conditions serve different purposes.
A voluntary commitment communicates present intent. An enforceable condition defines a measurable duty, assigns oversight, and creates consequences for noncompliance.
The enforcement gap becomes more important if demand grows. Gillingham argued that a data center owner would have a financial incentive to seek additional capacity later.
He described a possible progression from 15 megawatts to 20 or 30 megawatts. That was a hypothetical scenario, not a disclosed expansion plan from the companies.
Still, the scenario identifies the decision Trumbull must prepare for. A future capacity request could arrive after the town has accepted the initial conversion.
If protections depend only on the present ownership group’s statements, they may not survive a tenant change, property sale, or revised business plan.
A durable agreement would attach obligations to the facility or approval, not only to current executives. It would also specify access to operating data.
The confidential letter of intent adds uncertainty. Residents cannot compare the commercial plan with the public assurances because the underlying document remains unavailable.
A letter of intent is often preliminary and nonbinding. Its confidentiality does not by itself suggest misconduct, but it limits public understanding of the proposed arrangement.
Sarah Beck, another resident, said the town had not gathered enough information. She also noted the difficulty of identifying unknown risks around a relatively new class of infrastructure.
That skepticism is reasonable without assuming a worst-case outcome. Neither supporters nor critics possess a final public engineering design.
The town’s own project statement provides more detail than the company has released about Trumbull. It remains an official summary rather than a binding technical agreement.
The decisive document will establish baselines, reporting duties, inspection rights, and remedies. Until that exists, the four promises describe a negotiating position rather than a complete protection system.
A Small AI Site Can Still Matter to the Grid
Trumbull does not need to host a hyperscale campus for its experience to influence Connecticut’s wider AI infrastructure policy.
The facility’s local significance comes from concentration. A steady industrial load can affect planning even when it represents a small share of statewide electricity use.
AI computing raises the stakes because accelerators, specialized processors used for model workloads, demand more power per rack than conventional servers.
More concentrated power creates more concentrated heat. Operators then need cooling systems capable of removing that heat without disrupting hardware performance.
This relationship explains why electricity and water concerns often appear together. The exact balance varies according to cooling design, local climate, and hardware density.
Gillingham said a converted Trumbull facility would probably use more electricity than its previous configuration. He also warned that additional demand adds pressure during peak periods.
That does not establish that this single project will directly increase consumer bills. Utility rates depend on generation, transmission, distribution, contracts, regulation, and demand across the regional system.
It does show why capacity commitments deserve precise definitions. A stable 15-megawatt ceiling creates a different planning problem from an operation allowed to grow toward 30 megawatts.
The Trumbull case also complicates the common comparison between data centers and traditional job-producing developments. Data centers can require substantial investment while supporting relatively small permanent workforces.
Their local value therefore depends heavily on property revenue, service reliability, infrastructure effects, and the terms negotiated with the municipality.
Tesoro argues that the proposed investment will create a dependable tax base because the new owner has agreed not to appeal its assessment.
Residents want more information about whether those benefits justify the operational risks. Both positions depend on agreements that have not been fully published.
Connecticut has been wrestling with this balance since adopting its 2021 incentive law. That legislation sought to attract investment by offering tax benefits to qualifying facilities.
The policy was designed partly around financial services and the possibility that trading infrastructure might move from neighboring states. The largest anticipated relocation did not materialize.
Five years later, AI has changed the demand story. Developers now seek sites for denser computing, while lawmakers and municipalities focus more closely on grid and environmental effects.
A 2026 legislative proposal sought to remove certain state data center incentives. Testimony also urged stronger transparency and protections against shifting infrastructure costs to residents.
Trumbull sits between those two policy eras. Its data center reflects Connecticut’s earlier connection to financial and telecommunications customers, yet its proposed future depends on AI-ready infrastructure.
The site could therefore become a practical test of adaptive reuse. If it modernizes within firm resource limits, supporters can point to an alternative to building larger campuses.
If its demand, noise, or footprint later expands, critics can argue that an existing-site exemption enabled the outcome the moratorium intended to delay.
Comparable towns will watch that distinction. An outright ban on new facilities offers little guidance when an old data center changes its equipment, cooling, ownership, or customers.
Municipalities may respond with rules based on operational thresholds. Those could include megawatts, rack density, generator capacity, cooling technology, or verified water use.
They may also require periodic reporting rather than a one-time zoning decision. AI hardware changes faster than most local land-use rules.
This creates pressure on developers as well. Clear standards can reduce uncertainty by defining acceptable operating conditions before a company commits capital.
The alternative is case-by-case negotiation, where each project produces different promises and residents struggle to compare outcomes.
Trumbull has started that negotiation publicly. Its next task is translating broad assurances into limits that remain meaningful after the modernization is complete.
Three Signals That Decide What Happens Next
The project’s future will become clearer through an enforceable agreement, a detailed technical design, and the residents’ referendum effort.
The first signal is the legal form of the four operating limits. Trumbull must decide how to bind the commitments on electricity, water, noise, and physical expansion.
A host municipality agreement, permit condition, covenant, or another enforceable instrument could provide that structure. The correct mechanism depends on local authority and the approvals required.
Whatever form officials choose, the text should identify measurable baselines. It should also clarify reporting frequency, independent verification, and consequences for a violation.
This signal would strengthen the town’s position if the conditions follow the property and remain enforceable after an ownership or tenant change.
It would weaken the town’s case if the commitments remain informal. General assurances cannot answer disputes over measurement or future expansion.
The second signal is the final technical plan. Important missing details include confirmed power demand, rack density, cooling equipment, generator capacity, and the modernization schedule.
The town previously said demolition and renovation would take about six months. The companies have not released a complete public timetable for AI-oriented operations.
A detailed design that remains within the existing utility allocation would support the argument that this is a bounded reuse project.
A request for new grid capacity, additional generators, or materially different cooling infrastructure would challenge that characterization.
The companies’ own language also deserves attention. Their broader portfolio targets high-density computing and liquid-to-chip cooling across multiple markets.
Trumbull’s final specifications will show whether the location resembles those broader ambitions or serves a narrower group of enterprise customers.
The third signal is organized resident action. Opponents have discussed collecting signatures to force the council to accept proposed changes or hold a referendum.
Resident Richard White estimated that the effort would require approximately 2,400 signatures. Whether organizers reach that threshold will measure the depth of local opposition.
A successful petition would move the dispute from council negotiations to a broader public decision. It could also pressure officials to disclose more information before renovations advance.
Failure to gather enough signatures would not resolve the technical questions. It would, however, reduce the immediate political pressure for expanding the moratorium.
These signals matter beyond Trumbull. Other towns need to know whether a temporary ban can keep pace with changes inside existing infrastructure.
Developers also need a predictable path for modernizing facilities without turning every equipment upgrade into a new land-use battle.
The Trumbull AI data center proposal can establish a workable middle ground only if its limits are transparent, measurable, and durable.
Readers should watch the documents, not the labels. “Existing,” “small,” and “AI-ready” each describe one part of the project, but none defines its full impact.
The decisive question is whether Trumbull can regulate actual operations before the conversion becomes Connecticut’s first major precedent for AI-focused data center reuse.
