Can AI Data Centers Become Better Neighbors? E076

Can AI Data Centers Become Better Neighbors?

Venture Step E076 moved from documenting the costs of AI infrastructure to asking what a better project would require. The answer is not one efficient chip or clean-energy contract. It is coordinated governance across utility rates, siting, power, cooling, public disclosure, and enforceable local obligations.

This article substantially revises the original August 5, 2025 episode page after a July 2026 source review. The recording preserves Dalton Anderson's 2025 viewpoint. Policy status and technology claims below reflect the newer record.

The question after the problem

E075 examined the physical and local costs hidden behind the word "cloud." E076 was the second half of that argument. Dalton asked whether state policy, community participation, better cooling, more efficient computing, cleaner power, and public reporting could produce a different kind of data-center buildout.

That solution turn still holds up. The episode's individual examples needed work.

Some policies were proposals when the recording was made and later changed or became law. Some technology descriptions compressed very different systems. A memorable Microsoft underwater experiment was blended with a later land-based cooling design. The revised story keeps the governing idea and updates the evidence.

The most durable lesson is that the project must be governed as one system.

flowchart TD
    A["Large new computing load"] --> B["Utility rates and grid investment"]
    A --> C["Generation and transmission"]
    A --> D["Site, water, cooling, air, and noise"]
    A --> E["Roads, emergency response, jobs, and public revenue"]
    B --> F["Project agreements and public oversight"]
    C --> F
    D --> F
    E --> F
    F --> G["Measured operation, expansion, and closure"]

Rate design decides who carries the risk

One of E076's strongest ideas was that a large speculative load should not be treated like an ordinary customer. A utility may need to build substations, transmission, generation, and other facilities years before the customer reaches full demand. If the project is delayed, downsized, or canceled, someone remains responsible for those costs.

Ohio now provides a concrete example. The AEP Ohio data-center tariff page says the Public Utilities Commission of Ohio adopted the utility's settlement in July 2025. The current process applies to new data centers and expansions, with added study requirements at 25 megawatts or more. It includes load studies, long-term service agreements, load-ramp rules, buildout reimbursement, collateral, minimum-demand charges, and exit provisions.

The public record also corrects one detail from the episode. The current minimum-demand framework uses an 85 percent ceiling and related formulas, not the 90 percent proposal discussed in the recording.

Oregon took a legislative route. HB 3546 became Chapter 323 in June 2025. It directs the Public Utility Commission to establish a service classification for large energy-use facilities, allocate the costs of serving them, mitigate risk to other retail customers, and require qualifying service contracts.

These policies do not make a facility sustainable. They do something narrower and necessary: make the financial relationship between a very large customer and the public utility more explicit.

State frameworks are becoming more specific

Minnesota's current framework combines several mechanisms. The Minnesota Public Utilities Commission describes cost responsibility, special tariffs, electric-service-agreement review, clean-energy compliance, reliability review, and an annual community-support fee for qualifying large-scale data centers.

Georgia's response has also moved beyond the episode's bill narrative. The Georgia Public Service Commission approved a rule in January 2025 that allows additional contract terms for new customers above 100 megawatts and addresses site and upstream costs. A later commission action associated with Georgia Power's 2025 integrated-resource plan added another layer of claimed ratepayer protection. A separate 2026 House bill was still pending in the official legislative record reviewed for this revision.

The distinction matters. A commission rule, approved tariff, enacted statute, pending bill, utility settlement, and company promise have different authority. A serious public page should name which one it is using.

The policy goal should not be finding one perfect law. It should be closing the gaps between these mechanisms.

Siting is an infrastructure decision

E076 argued that communities need a real say before a site and design become politically difficult to change. Fairfax County, Virginia offers a useful example of measurable land-use controls.

The county's September 2024 zoning amendment requires data-center buildings to sit at least 200 feet from adjacent residential property or districts. Specified backup equipment must be 300 feet away or separated by the data-center building, subject to the ordinance's special-exception path. The amendment also addresses equipment screening, building design, and pre- and post-construction noise studies.

Those distances are not a universal answer. They show what a public rule can make observable. Instead of promising to be a good neighbor, a project has to identify where the building and equipment sit, how noise will be measured, and which review applies if it cannot meet the standard.

The broader site record should cover land authority, grid studies, water rights and capacity, cooling, fiber diversity, flood and fire exposure, air permits, roads, construction, emergency response, public cost, expansion, and closure. [[How to Evaluate a Data Center Site]] turns that list into a constraint-first sequence.

Better cooling begins with the measurement boundary

The episode looked to liquid cooling, immersion, reclaimed water, underwater experiments, and more efficient hardware for relief. Each can matter. None removes the need to follow heat and resource use through the complete system.

The Department of Energy explains how computing heat can move through room air, chilled-water systems, liquid loops, cooling towers, dry coolers, and hybrids. Direct-to-chip cooling can reduce fan and chiller demand at high rack density, but the facility still needs a final way to reject heat.

Microsoft's examples need to be separated. Project Natick was an underwater data-center experiment off Scotland. Microsoft's later zero-water cooling design is a land-based, closed-loop, chip-level system introduced in 2024 that the company says avoids water evaporation for cooling during operation.

The later design was not achieved by putting conventional data centers underwater. It was achieved by changing the cooling architecture and heat-rejection boundary.

Reclaimed water adds another option. The EPA's Quincy, Washington case documents an industrial water-reuse system associated with data-center operations. That approach can preserve potable supply, but it still needs treatment, pipes, energy, contracts, backup, and a local water analysis.

[[Data Center Cooling Options and Their Water Tradeoffs]] compares the architectures without naming one universal winner.

Clean power has to arrive in the same world as the load

E076 correctly treated additional power as a central part of the solution. The difficult part is defining "additional."

A data center can sign a contract for enough renewable electricity to match annual consumption. It will still draw from the connected grid every hour. Solar production will not match a flat load at night. A distant project may face transmission limits. A proposed resource may spend years in development and interconnection.

The International Energy Agency analyzes the physical generation serving data centers rather than operators' contractual portfolios. Its base case expects renewables to supply a large share of added demand while fossil generation remains important in the near term.

A credible plan therefore needs new or newly available energy, a real interconnection and delivery path, hourly accounting, firm capacity, transmission, storage or flexible load, backup, and an interim plan if infrastructure is late. [[What Additional Clean Power Means for Data Center Growth]] explains how to test each part.

Community benefits need authority and remedies

The episode argued that host communities should receive more than disruption and generalized economic-development promises. That principle becomes useful only when a commitment can be governed.

A community benefits agreement should identify legitimate parties, a verified project baseline, measurable obligations, milestones, public reports, enforcement, remedies, change control, successor duties, expansion, and closure. It should distinguish construction work from long-term jobs and a public fee from a project-specific obligation.

Massachusetts' 2026 community-benefit guidance emphasizes early participation, community-defined priorities, transparency, and measurable benefits. The state's later responsible-data-center expectations point developers toward that framework. It is useful evidence, not a universal data-center contract.

[[How Community Benefits Agreements Can Govern Data Center Projects]] provides the governance anatomy. It remains under a professional-review hold because authority and enforceability depend on local law and the parties. No generic clause can substitute for representative participation and qualified counsel.

Transparency is the operating layer

Project approval begins with forecasts. Accountability begins when metered results replace them.

A useful operating record should show load, electricity, supply method, backup generation, water withdrawal and consumption, discharge, facility efficiency, air and noise compliance, incidents, jobs, public payments, community obligations, and project changes. Security and commercial confidentiality can be protected without converting all material outcomes into secrets.

Expansion should reopen the relevant parts of the decision. A first phase that fits the grid and watershed does not prove that later phases fit. A new owner, cooling system, water source, generator, or workload can change the answer.

That is the missing connection between many of E076's ideas. Rates, siting, cooling, energy, and community benefits are not separate acts of corporate responsibility. They are linked controls that need shared project data.

A better neighbor is a governed project

The phrase "sustainable data center" can become too broad to be useful. E076's updated contribution is more concrete.

A better project defines its workload and phases, chooses a site that can support them, pays the attributable infrastructure cost, adds credible power, fits the watershed, controls air and noise, gives affected communities meaningful authority, reports actual operation, and carries its duties through expansion and closure.

Use [[Can AI Data Centers Be Sustainable]] as the full decision framework. If the site's basic fit is still unresolved, start with [[How to Evaluate a Data Center Site]]. E075 remains the companion conversation about the costs that this solution package is designed to confront.

The original E076 recording is available on Spotify and YouTube. It is worth hearing as a dated attempt to move beyond criticism and toward design. The revised pages do the slower work of updating what happened next.

This article was freshly written from the preserved E076 transcript and current official records reviewed on July 27, 2026. It distinguishes Dalton's August 2025 viewpoint from later policy status and corrects the Microsoft cooling example. AI assistance was used for research organization, drafting, and validation. Publication remains unauthorized.