The Project Is Bigger Than a Building
A conventional construction project can be understood primarily as a building.
An AI data center is different.
It is better understood as an energy-and-computing ecosystem.
The building is only one component.
A hyperscale AI campus requires:
- electrical generation and transmission
- substations
- high-voltage distribution
- backup power
- cooling infrastructure
- water management
- fiber connectivity
- structural systems
- mechanical systems
- fire protection
- security
- roads and logistics
- stormwater management
- maintenance infrastructure
- enormous amounts of computing equipment
Project Washington's proposed 1.2 GW electrical load illustrates the scale of the challenge. Delaware's Public Service Commission has noted that a 1,200-MW facility would represent almost half of the state's peak summer electricity demand, which the PSC described as approximately 2,700 MW.
At that scale, the question is no longer simply:
"Can Delaware build the building?"
The question becomes:
"Can Delaware build the entire infrastructure ecosystem responsibly?"
The AI Data Center Is Also a Construction Project
This is where the conversation should become particularly interesting to the construction industry.
AI may be developed in software, but AI infrastructure is constructed with physical materials and skilled labor.
A large AI campus can require:
- concrete foundations
- structural steel
- roofing systems
- massive electrical distribution
- switchgear
- transformers
- generators
- mechanical piping
- cooling equipment
- fire suppression
- controls
- security systems
- fiber infrastructure
- site utilities
- roads
- drainage
- landscaping
- commissioning
That means the AI boom is also creating a new category of construction demand.
For contractors, the question is not simply whether an AI data center is good or bad for Delaware.
The question is:
What construction opportunities will this infrastructure create, and what standards should those projects meet?
The Problem: Scale Changes Everything
A 10-MW facility and a 1.2-GW campus should not be treated as the same type of development.
The scale changes the engineering.
It changes the grid impact.
It changes the water requirements.
It changes the backup-power requirements.
It changes the environmental footprint.
It changes the construction logistics.
And it changes the consequences of failure.
The Delaware PSC has already recognized this distinction by treating facilities of 25 MW or more as "large load" facilities in the context of its regulatory discussion. The commission opened a docket for a large-load tariff intended to prevent data centers and other large loads from shifting infrastructure costs to other ratepayers.
In July 2026, Delaware lawmakers also advanced legislation aimed at establishing additional standards for large energy users such as data centers.
The policy question is therefore evolving from:
Should Delaware have data centers?
to:
Under what conditions should Delaware allow them?
That is a much more useful question.
What If We Designed the AI Data Center From the Ground Up?
Here is where the conversation becomes more interesting.
Instead of treating environmental and infrastructure concerns as obstacles that must be mitigated after the design is complete, what if they became design requirements from day one?
Imagine an AI data center designed around six principles:
1. Minimize surface impact
2. Minimize water consumption
3. Minimize grid stress
4. Capture and reuse waste heat
5. Reduce fossil-fuel dependence
6. Make the facility economically beneficial to the surrounding community
Some of these concepts already exist independently.
The opportunity is to combine them.
Concept One: Build Part of the AI Data Center Underground
This sounds radical.
It may also be worth investigating.
An underground or partially earth-sheltered AI campus could potentially reduce certain surface impacts:
- visual mass
- surface building footprint
- noise propagation
- exposure to extreme weather
- some surface heat effects
- potentially the amount of land that must remain permanently occupied by above-ground structures
But underground construction does not automatically equal environmental protection.
Excavation creates its own impacts.
A large underground facility would require:
- massive excavation
- groundwater analysis
- dewatering strategy
- waterproofing
- structural retaining systems
- ventilation
- emergency egress
- fire/life-safety systems
- material handling
- underground utility distribution
- long-term access and maintenance systems
And there is another problem.
Heat.
A very large AI facility produces an enormous amount of heat.
Putting the computers underground doesn't make that heat disappear.
It simply changes the thermal environment in which the heat must be managed.
Concept Two: Treat the Earth as a Thermal Resource
This is where an old principle becomes interesting in a very modern application.
For centuries, people have understood that the ground provides a relatively stable thermal environment compared with outdoor air.
Modern geothermal heat-pump systems deliberately use that characteristic.
The U.S. Department of Energy explains that shallow underground temperatures are relatively stable and can act as a heat sink during cooling operations. Ground-source systems circulate fluid through underground heat exchangers to transfer heat between buildings and the ground.
So what if an underground AI campus incorporated a large-scale ground-coupled thermal system?
Not a few pipes.
Not a residential geothermal loop.
A purpose-engineered thermal infrastructure system.
That distinction matters.
A 500-MW Problem Cannot Be Solved With a 6-Inch Pipe
This may sound obvious, but it illustrates one of the fundamental engineering principles that should govern AI infrastructure:
Thermal infrastructure must be proportional to computational load.
As a simplified engineering concept, electricity consumed by computing ultimately becomes heat.
So if a hypothetical AI campus were consuming 500 MW continuously, the thermal-management system would ultimately need to deal with approximately 500 MW of heat, subject to the facility's actual efficiency and auxiliary loads.
That is not an HVAC problem in the conventional building sense.
It is a thermal infrastructure problem.
For perspective:
100 MW for one hour = 100 MWh of energy
At 100 MW of continuous IT-related heat, a full day represents:
2,400 MWh of thermal energy.
The numbers quickly become enormous.
That is why an underground AI facility should not be conceptualized as:
Data center + geothermal HVAC.
It should be conceptualized as:
Computing + liquid cooling + heat exchangers + thermal storage + heat rejection + possible heat reuse.
That is a completely different engineering problem.
Concept Three: Liquid Cooling Instead of Treating the Building Like a Giant Air Conditioner
AI computing is increasingly pushing toward higher-density computing architectures.
That makes direct liquid cooling particularly interesting.
The National Renewable Energy Laboratory has demonstrated high-performance computing facilities using direct liquid cooling, including systems designed to capture heat directly from computing equipment and reuse it. NREL's work emphasizes three connected objectives: efficient liquid cooling, waste-heat recovery, and minimizing water use.
This suggests an important conceptual shift:
Instead of:
GPU → air → room → air conditioner → heat rejection
a more efficient architecture can be:
GPU → liquid → heat exchanger → thermal system
The closer the cooling medium is to the heat source, the more directly the system can manage the thermal load.
NREL has demonstrated warm-water liquid cooling in a high-performance computing environment, with return-water temperatures around 95°F–104°F in one system and recovered heat used to support surrounding buildings.
That is a real engineering precedent—not a science-fiction concept.
The question is whether the concept can be scaled and economically engineered for much larger AI campuses.
Concept Four: Don't Throw the Heat Away
This may be one of the biggest opportunities.
The conventional mindset is:
Computers generate heat. Cooling systems remove heat.
But heat is energy.
The better question may be:
Who can use the heat?
A future AI campus could potentially integrate with nearby thermal users such as:
- commercial buildings
- industrial facilities
- greenhouses
- agricultural operations
- district heating systems
- water-treatment processes
- other thermal users
Not every waste-heat application will be economically viable.
Temperature matters.
Distance matters.
Seasonality matters.
Infrastructure costs matter.
But the principle is already being demonstrated.
NREL has specifically researched data-center waste-heat recovery and reuse, including using recovered computing heat for building heating.
That changes the equation from:
Electricity → computation → waste heat → atmosphere
to:
Electricity → computation → useful thermal energy → secondary economic activity
Concept Five: The Underground Facility Could Become a Thermal Battery
This is where the concept gets more speculative.
Imagine an underground AI campus with a large thermal network.
The ground is not treated as an infinite heat sink.
Instead, it becomes one component of a larger system.
During favorable conditions:
Heat → thermal storage / ground system
During unfavorable conditions:
Heat → engineered heat rejection system
At the same time:
Higher-temperature waste heat → potential heat reuse
The control system continuously determines the most efficient path.
That creates something closer to a thermal dispatch system.
The facility could theoretically decide:
Store.
Reject.
Reuse.
Recover.
Shift load.
The objective would be to minimize total energy and water consumption while maintaining the required computing temperature.
Concept Six: Build the Energy System With the Data Center
Another major issue is electricity.
A 1.2-GW data center cannot simply be viewed as another commercial customer plugging into an existing electrical network.
Delaware's PSC has explicitly highlighted the extraordinary scale of proposed data-center loads and the potential impact on electrical infrastructure and ratepayers.
That suggests another conceptual approach:
Build the energy infrastructure as part of the project.
Instead of:
Data Center → Existing Grid
consider:
Data Center + Grid + Dedicated Generation + Storage + Transmission
as one integrated infrastructure project.
Possible components could include:
- renewable generation
- battery energy storage
- long-term power purchase agreements
- dedicated substations
- grid-support systems
- demand-response capability
- potentially other generation technologies subject to environmental and regulatory requirements
The objective would be to reduce the amount of infrastructure burden transferred to the surrounding system.
Concept Seven: Make the Data Center Pay for the Infrastructure It Requires
This is less futuristic and perhaps more immediately practical.
If a project requires:
- new substations
- transmission upgrades
- distribution upgrades
- new generation
- water infrastructure
- roads
- emergency services
then those costs should be transparently allocated.
Delaware's PSC opened its large-load tariff proceeding specifically around the concern that large facilities should not shift their infrastructure costs to other ratepayers.
That principle could become part of a broader Responsible AI Infrastructure Standard.
Concept Eight: Build on Previously Disturbed Land
The question should not simply be:
"Where is there enough land?"
It should be:
"Where can Delaware accommodate the infrastructure with the least additional environmental and community impact?"
That could prioritize:
- existing industrial areas
- brownfields
- previously disturbed land
- existing electrical corridors
- existing fiber infrastructure
- existing water infrastructure
- locations with appropriate zoning
- locations away from sensitive environmental resources
This could reduce the need to convert large areas of relatively undisturbed land.
It could also make the project easier to integrate with existing infrastructure.
Concept Nine: Modular AI Infrastructure
Why build the entire campus before the demand exists?
A different approach could be:
100 MW
→ commission
→ evaluate
→ expand to 250 MW
→ evaluate
→ expand to 500 MW
→ continue only when infrastructure and demand justify it.
This is essentially phased infrastructure scaling.
It could reduce:
- stranded capital
- unnecessary construction
- premature grid investment
- environmental exposure
- uncertainty
It also gives regulators and communities measurable data before the next expansion.
Concept Ten: A Data Center That Produces More Than Computing
This is perhaps the most provocative concept.
Imagine an AI campus that is not simply an electricity consumer.
It becomes an infrastructure hub.
ENERGY
↓
┌───────────────┐
│ AI COMPUTING │
└───────┬───────┘
│
┌───────┼────────┐
↓ ↓ ↓
HEAT DATA JOBS
↓ ↓ ↓
INDUSTRY AI TRAINING
│ │
└───────┬────────┘
↓
COMMUNITY
The project could potentially provide:
- construction employment
- permanent technical employment
- apprenticeship programs
- local contractor opportunities
- tax revenue
- workforce training
- energy infrastructure
- heat reuse
- technology partnerships
That creates a different definition of economic development.
What About the Water Problem?
Water deserves its own discussion.
Project Washington has generated significant attention over its projected water needs. Public reporting based on project documents has cited annual water-use estimates in the range of roughly 10–20 million gallons, while the developer has described approximately 12.7 million gallons of annual use and a one-time fill requirement for its closed-loop cooling system.
The precise figure matters less than the principle:
Cooling design is now part of water policy.
A future AI campus should be evaluated not simply on how much water it uses, but on:
Water Use Effectiveness
and the source of that water.
Potential strategies include:
- closed-loop cooling
- reclaimed water
- treated wastewater
- rainwater capture where practical
- dry cooling where economically and technically appropriate
- liquid cooling
- heat recovery
- hybrid cooling systems
New Castle County has also considered data-center rules involving closed-loop cooling systems.
The Backup Power Question
This may be one of the most difficult issues.
Project Washington's proposal included 516 backup generators and a proposed fuel-storage system exceeding 2.5 million gallons of diesel. Those elements were central to DNREC's February 2026 Coastal Zone Act decision.
The engineering reason for backup generation is understandable.
AI infrastructure cannot simply shut down every time the grid experiences an interruption.
But the question is:
Does the backup architecture have to look like a traditional hyperscale facility's backup system?
Future systems could investigate combinations of:
- battery energy storage
- multiple power feeds
- microgrids
- cleaner backup generation
- diversified generation
- energy storage
- demand management
- advanced controls
No single technology automatically solves the problem.
But the design requirement should be:
Mission-critical reliability without unnecessarily creating another environmental liability.
Could an AI Data Center Be Underground?
Possibly.
Would it automatically be better?
No.
The correct answer is more interesting.
Potential benefits
- smaller visible surface footprint
- potential noise attenuation
- protection from some weather conditions
- stable subsurface temperatures
- potential integration with thermal storage
- potential restoration of portions of the surface
Potential disadvantages
- enormous excavation
- groundwater management
- waterproofing
- emergency access
- fire protection
- ventilation
- construction logistics
- heat rejection
- higher capital costs
- complicated maintenance
The concept therefore deserves engineering analysis, not dismissal.
The real question is:
Could a partially underground AI campus reduce the project's net environmental footprint when the entire life cycle—including excavation, construction, cooling, energy, and operations—is considered?
That is a question worth modeling.
The Most Interesting Concept: An Integrated AI Infrastructure Campus
Perhaps the future facility shouldn't be called simply a data center.
Maybe it should be designed as an:
AI Infrastructure Campus
A conceptual Delaware facility could combine:
Computing
High-density AI/GPU systems.
Electrical infrastructure
Dedicated substations, storage, grid interconnection and power-management systems.
Thermal infrastructure
Direct liquid cooling, heat exchangers, thermal storage and heat recovery.
Water infrastructure
Closed-loop cooling, reclaimed water and water-efficiency systems.
Construction
Modular, phased buildings designed for expansion.
Land strategy
Brownfield or previously disturbed sites where practical.
Community strategy
Local workforce, apprenticeships, contractor participation and transparent economic commitments.
That is a fundamentally different model from simply building the largest possible collection of server buildings.
Delaware Could Turn the Controversy Into a Design Challenge
Project Washington has already forced Delaware to confront questions that many communities around the country are now facing.
How much electricity should a single private project consume?
Who pays for the infrastructure?
How much water should it use?
Where should it be located?
How should backup power be handled?
How much noise is acceptable?
How should communities benefit?
How should environmental risks be evaluated?
And perhaps the most important question:
What should responsible AI infrastructure look like?
Delaware's updated Climate Action Plan emphasizes resilience, infrastructure, energy, land use, and community impacts, while the state continues developing policies around large energy users and data centers.
That means Delaware is not simply deciding the fate of one project.
It is beginning to establish a framework for an industry that could continue evolving.
BRO's Proposal: Build the Next Data Center Differently
BRO Builder believes the discussion should move beyond a simple choice between:
Build it
or
Don't build it.
There is a third possibility:
Design it better.
A future Delaware AI infrastructure project could be required to demonstrate:
1. Energy responsibility
The project should transparently account for the incremental electrical infrastructure it requires.
2. Water responsibility
Cooling systems should minimize freshwater consumption and maximize reuse where technically feasible.
3. Thermal responsibility
Waste heat should be evaluated as a potential resource rather than automatically treated as waste.
4. Land responsibility
Previously disturbed or industrial land should be considered before undeveloped land where appropriate.
5. Construction responsibility
Local contractors and workers should have meaningful opportunities to participate.
6. Community responsibility
Economic benefits should be measurable and transparent.
7. Infrastructure responsibility
The project should demonstrate that roads, utilities, emergency response and electrical systems can support its operation.
8. Resilience
The facility should be designed for extreme weather, grid disruptions and other foreseeable risks.
9. Transparency
Communities should have access to meaningful information before major commitments are made.
10. Innovation
Developers should be encouraged—or required—to evaluate emerging solutions rather than simply reproduce conventional hyperscale designs.
The Bigger Opportunity for Delaware
There is a temptation to think of AI data centers as something that happens somewhere else.
But AI infrastructure is rapidly becoming a physical infrastructure industry.
And physical infrastructure creates opportunities for:
- electricians
- plumbers
- HVAC contractors
- mechanical contractors
- concrete contractors
- framers
- steel contractors
- equipment operators
- civil contractors
- surveyors
- engineers
- commissioning specialists
- technology contractors
- maintenance companies
The AI economy may be digital.
Its construction bill is not.
That distinction matters to Delaware.
What Happens to Project Washington?
At the time of publication, Project Washington remains a proposed project rather than an operating data center. DNREC determined in February 2026 that the project was prohibited from development in the Coastal Zone, and the Coastal Zone Industrial Control Board issued its April 17, 2026 final order following Starwood's appeal. The legal and regulatory questions surrounding the project are therefore not simply a matter of construction scheduling; they are part of a broader debate about Delaware's treatment of large-scale AI infrastructure.
Whether Project Washington ultimately moves forward, changes form, relocates, or does not get built, the underlying demand for AI infrastructure is unlikely to disappear.
That means Delaware will probably have another opportunity to answer the same fundamental question.
Perhaps the First AI Data Center Should Be the First AI Infrastructure Experiment
There is an opportunity here that goes beyond Project Washington.
What if Delaware became known not as the state that either accepted or rejected AI data centers, but as the state that developed a better model for them?
Imagine a facility where:
- computing is optimized for energy efficiency
- heat is captured rather than simply rejected
- water is reused
- electrical infrastructure is planned with the project
- energy storage supports resilience
- construction is phased
- previously disturbed land is prioritized
- local workers participate
- contractors have access to procurement opportunities
- communities receive measurable benefits
- environmental impacts are quantified
- new technology is tested rather than dismissed
That would be a different kind of infrastructure project.
It would be an experiment in how the physical world adapts to artificial intelligence.
The AI Revolution Will Need Builders
Artificial intelligence is often discussed as if it exists entirely inside software.
It doesn't.
Behind every AI model are buildings.
Behind those buildings are electrical systems.
Behind those electrical systems are substations, transformers, transmission lines and generation.
Behind the servers are cooling systems.
Behind the cooling systems are pumps, pipes, heat exchangers and thermal infrastructure.
And behind all of it are engineers, contractors, operators and construction workers.
AI needs builders.
Project Washington has forced Delaware to confront that reality earlier than many states.
The question now is not simply whether Delaware will build an AI data center.
The more important question may be:
If Delaware builds the next generation of AI infrastructure, can it build it smarter?
Perhaps the answer begins underground.
Perhaps it begins with liquid cooling.
Perhaps it begins with thermal storage.
Perhaps it begins with reclaimed water.
Perhaps it begins with a different energy model.
Or perhaps it begins with a simple engineering principle:
Don't treat the data center as a building. Treat it as an integrated infrastructure system.
The first major AI data center proposed for Delaware may ultimately be remembered not only for its size or controversy, but for the questions it forced the state to ask.
And those questions are only beginning.
Share Your Experience