---

Discussion

Eric Arcan:

The issue is bigger than monthly electricity consumption. A large data center can require new generation, high-voltage transmission, substations, distribution equipment, and reliability arrangements. Utilities may commit those resources before the facility reaches full operation. If the project shrinks or cancels, the equipment remains while the customer commitment disappears. The bill's central idea is that a qualifying large load should cover the incremental costs needed to serve it and protect against stranded assets.

Kate Burvish:

That distinction between energy and infrastructure is the economic center of the debate. If new capacity is spread across the utility territory, households and small businesses may pay even if they never use the data center's services. But assigning every investment automatically to the facility could discourage projects that use existing capacity or help finance infrastructure that later benefits everyone. The key question is which costs would exist without this particular load.

Red Velhouse:

Before deciding who pays, what exactly is being powered? Omar, how do major workloads affect the electricity system differently?

Omar Seidren:

A data center is an information-processing facility surrounded by an electrically intensive support system: servers, storage, networking, uninterruptible-power equipment, cooling, backup generation, and network connections. Conventional cloud and enterprise computing runs databases, business software, websites, and storage. Video streaming adds content delivery: facilities store, cache, encode, and transmit video, often through regional locations close to viewers.

Omar Seidren:

Artificial-intelligence training uses large numbers of graphics-processing units, or GPUs, operating in parallel for long periods to build or refine models. Those accelerators are power-dense, creating concentrated electricity and cooling requirements. Artificial-intelligence inference—generating an answer or prediction from a trained model—can also be intensive, but it often has a latency requirement and may be distributed across several locations so users do not wait for a distant facility.

Eric Arcan:

And a megawatt is not just a megawatt. A steady, predictable load is easier to plan for than one that ramps rapidly. A load that can curtail during a heat wave is different from one that must remain online. Dense artificial-intelligence clusters can also have different cooling, voltage, and power-quality requirements from storage facilities. The North American Electric Reliability Corporation, or NERC, is developing standards for computational loads because behavior during disturbances matters, not merely annual consumption.

Red Velhouse:

So the label hides several kinds of customers. Returning to the legislation: what would H.R. 9340 require, and where would states retain discretion?

Kate Burvish:

It would require state utility regulators and nonregulated utilities to consider a standard under which qualifying large-load customers cover the incremental costs of generation, transmission, and distribution upgrades needed to serve them. It also contemplates financial assurances or contributions before construction. But the bill preserves state authority; states are asked to consider the standard, not adopt identical rules. Its practical effect depends on commissions, utility contracts, collateral requirements, and enforcement.

Eric Arcan:

That discretion matters because the engineering problem varies by site. A project using existing capacity is not the same as one requiring a new substation, transmission line, or generation resource. Regulators need expected hourly demand, ramping, power-quality requirements, cooling needs, and genuine curtailment capability before assigning costs.

Omar Seidren:

The workload should shape the contract too. A hyperscale cloud or real-time inference facility may need firm service. A mining operation or some batch-training schedule may accept managed curtailment. That does not mean every workload moves with a switch; data-transfer costs, security, coordination, and deadlines matter. Applicants should identify what portion of demand is genuinely flexible instead of labeling an entire campus firm or flexible.

Kate Burvish:

That design explains both the bill's appeal and the dispute. Republicans point to the 417-to-3 House vote and say they acted to keep ordinary customers from subsidizing major technology companies. Democrats argue that a standard states merely consider is not enforceable protection. The same flexibility that makes the bill politically saleable can make results uneven.

Red Velhouse:

Is that why the Senate effort stalled even though both parties describe consumer protection as part of the debate?

Kate Burvish:

The mechanism is not settled. Senate Republicans are using the issue to strengthen an affordability and consumer-protection message before the midterms. Democrats can support the goal that large users should pay their share while arguing that the House bill is too weak because implementation is voluntary. Senator Martin Heinrich's GRID Savings Act would require large electricity users to pay for facilities needed to connect them. Republicans respond that a mandatory federal regime could intrude on state authority or slow strategic artificial-intelligence investment. The fight is over substance, implementation, and political credit.

Eric Arcan:

The political timing does not make the physical problem less real. The Energy Information Administration, or EIA, reported average annual electricity-demand growth of about 1.7 percent from 2020 through 2025, compared with 0.1 percent from 2005 through 2019. Data centers are one contributor, alongside manufacturing, electrification, weather, and other changes. If load arrives faster than generation and transmission can be built, the consequences show up regardless of the campaign calendar.

Kate Burvish:

But we should not turn that national trend into a universal claim about household bills. PJM Interconnection, commonly called PJM, reports major capacity-price increases associated with load growth, while the broader research says the relationship between data-center expansion and household electricity prices remains geographically and methodologically contested. Those findings are not automatically contradictory: a constrained region may face pressure while a national estimate over a different period shows a different effect. Location, market structure, timing, and method matter.

Red Velhouse:

That brings us to location. If the same computing service can be placed in different regions, where should facilities go? Is proximity to a power plant the obvious answer?

Eric Arcan:

It is too simple. The first test is deliverable grid capacity: an actual interconnection path, adequate transmission and substation capability, manageable congestion, protection equipment, and a credible upgrade plan. A plant may be nearby while the network cannot deliver its output to the campus. Continuous workloads also need firm service and power quality. A renewable power-purchase agreement or nearby wind and solar project does not eliminate the need for interconnection, backup, or balancing. The engineering question is what electricity is available during difficult hours.

Omar Seidren:

The second test is network topology. Streaming, interactive cloud applications, gaming, financial services, and real-time artificial-intelligence inference benefit from being near users, fiber routes, internet exchanges, and regional network hubs. That can mean distributed facilities rather than one giant remote campus. Training, backups, archival storage, background analytics, and some scientific-computing jobs are more geographically flexible. They can sometimes be scheduled during surplus, shifted among facilities, or curtailed under an agreement. The cloud is not weightless; it has a physical address.

Kate Burvish:

Flexibility has economic value, but it should not become a slogan. A company may promise emergency curtailment while reserving firm capacity all year. Regulators need hourly load forecasts, ramp-rate information, curtailment terms, and financial responsibility for reserved infrastructure. A lower-cost site is not actually lower cost if it lacks transmission, fiber, water, permits, or a customer willing to stand behind upgrades.

Red Velhouse:

Water is another reason a power-rich site may not be low-impact. What should planners evaluate?

Eric Arcan:

They should evaluate local water availability and competing community demands, including peak summer and drought conditions—not rely on a national average. Evaporative cooling can reduce electricity use but consume water, while air cooling may reduce direct water use but require more electricity in hot climates. Liquid cooling and closed-loop systems can support dense artificial-intelligence hardware while reducing or eliminating routine potable-water use, depending on the heat-rejection design.

Kate Burvish:

Reclaimed wastewater or industrial process water can be practical, but it requires treatment, pipes, permits, energy, residuals management, and long-term agreements. Quincy, Washington, offers a useful example: Microsoft and the city developed a closed-loop reuse system that saves an estimated 138 million gallons a year. It required roughly $31 million in capital financing and more than a decade of planning and construction. Other facilities farther away found connection costs prohibitive and continued using potable groundwater as of early 2023.

Red Velhouse:

We have three tests: deliverable power, workload-specific network needs, and local water feasibility. Should Congress impose a mandatory national rule, establish a federal baseline with state discretion, or leave this primarily to the Federal Energy Regulatory Commission, regional operators, and state commissions?

Eric Arcan:

I favor a firm baseline for cost responsibility, reliability, and site disclosure, with regional implementation. The customer that causes a dedicated upgrade should stand behind it financially. Applicants should disclose hourly demand, ramping, power-quality needs, cooling requirements, and genuine curtailment capability. But rules should recognize shared infrastructure and avoid blocking projects that use existing capacity or absorb surplus power.

Omar Seidren:

I would add a workload-specific operating plan. A facility should explain which services require low latency and continuous operation, which jobs can be delayed or relocated, and how redundancy protects users during curtailment. That creates an incentive to design software and contracts for flexibility instead of asking the grid to treat every workload as firm demand. Artificial-intelligence training, video delivery, cloud applications, and mining do not have identical needs.

Kate Burvish:

I would preserve substantial state and regional discretion, but require transparent accounting, credible financial assurances, and public justification for exceptions. Markets work best when participants face the costs they impose. They fail when optimistic projections are socialized and losses are left to captive ratepayers. The evidence does not show that every data center raises every household's bill, but it does support guarding against identifiable incremental costs and evaluating water and infrastructure locally.

Red Velhouse:

The unresolved issue is not whether data centers need enormous amounts of electricity. It is what they are doing, which workloads require firm and nearby service, which can shift toward grid surplus, and who bears the risk when power, water, and transmission systems expand for them. Responsible siting looks for deliverable grid capacity, reliable power, adequate transmission, water or reuse options, strong network connections where latency demands them, and enforceable financial responsibility. Proximity to a power plant alone is not enough, and no single location is best for every workload. Watch Senate procedure, efforts to make cost recovery mandatory, the Federal Energy Regulatory Commission's large-load reforms, reliability standards, regional capacity prices, and state decisions such as Virginia's. Those developments will show whether this becomes enforceable ratepayer protection or remains a campaign principle. Sources and references for this discussion are available with the episode at Factolio.com.

---

Sources and References

These sources supported the factual material used in this discussion. Factolio’s panel discussion is AI-generated from researched evidence and is written in original language.

1. U.S. House Clerk — U.S. House of Representatives Roll Call Votes: H.R. 9340, Ratepayer Protection Act (PRIMARY)
2. Associated Press — House passes bill aimed at addressing impact of data centers on energy costs (NEWS)
3. U.S. House of Representatives — Ratepayer Protection Act (PRIMARY)
4. Congressional Research and legislative record — H.R. 9340 / Ratepayer Protection Act legislative text and status (PRIMARY)
5. Congressman Brian Fitzpatrick — Fitzpatrick-Led Effort to Shield Ratepayers from Data Center Costs Passes House (PRIMARY)
6. Senator Jon Husted — Husted leads bill to protect Americans from footing the bill for new data centers (PRIMARY)
7. Reuters — Hurt by Trump, Senate Republicans plan to bolster affordability credentials before midterms (NEWS)
8. Senator Jon Husted — Husted bill to protect Americans from footing the bill for new data centers blocked from passage (PRIMARY)
9. U.S. Senate Committee on Energy and Natural Resources — Heinrich Introduces Legislation to Ensure Large Users of Electricity Pay for Grid Upgrades (PRIMARY)
10. Senator Martin Heinrich — Heinrich Offers His GRID Savings Act to Force AI Data Centers to Pay for Grid Upgrades (PRIMARY)
11. U.S. Department of Energy — DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers (PRIMARY)
12. Lawrence Berkeley National Laboratory / U.S. Department of Energy — 2024 United States Data Center Energy Usage Report (DATA)
13. U.S. Energy Information Administration — Fossil generation could rise with faster-than-expected growth in data center power demand (DATA)
14. Federal Energy Regulatory Commission — FERC Launches Aggressive Targeted Action to Speed Large Load Integration (PRIMARY)
15. Federal Energy Regulatory Commission — Commissioner Rosner's Remarks on the Large Load Show Cause Orders (PRIMARY)
16. North American Electric Reliability Corporation — Computational Loads Standards Advance Following Initial Ballot (PRIMARY)
17. North American Electric Reliability Corporation — Large Loads Action Plan (PRIMARY)
18. Federal Energy Regulatory Commission / NERC — 2026 FERC Orders and Rules: Computational Load Integration (PRIMARY)
19. PJM Interconnection — PJM Annual Report 2025: Markets (DATA)
20. International Energy Agency — Energy and AI (ANALYSIS)
21. International Energy Agency — Key Questions on Energy and AI (ANALYSIS)
22. Lawrence Berkeley National Laboratory — 2024 United States Data Center Energy Usage Report (DATA)
23. U.S. Department of Energy — Powering AI and Data Center Infrastructure Recommendations (PRIMARY)
24. U.S. Department of Energy — Clean Energy Resources to Meet Data Center Electricity Demand (PRIMARY)
25. U.S. Environmental Protection Agency — Water Reuse Case Study: Quincy, Washington (PRIMARY)
26. U.S. Environmental Protection Agency — Water Reuse for Industrial Applications Resources (PRIMARY)
27. U.S. Department of Energy — National Transmission Needs Study (PRIMARY)
28. North American Electric Reliability Corporation — Characteristics and Risks of Emerging Large Loads (PRIMARY)
29. U.S. Energy Information Administration — Data centers and cryptocurrency mining in Texas drive strong power demand growth (DATA)
30. International Telecommunication Union — ITU-T Recommendation F.743.6: Service Requirements for Next Generation Content Delivery Networks (PRIMARY)
31. U.S. Department of Energy — Electricity Rate Designs for Large Loads: Evolving Practices and Opportunities (PRIMARY)
32. Wiley / International Journal of Energy Research — Carbon-Aware Dispatch and Resilient Scheduling of Data Centers With Spatiotemporal Load Shifting on Illustrative Renewable-Powered Microgrids With Temporal Offset (ANALYSIS)
33. U.S. Department of Energy — Op Ed: The Truth About AI Data Centers (PRIMARY)
34. International Energy Agency — Electricity 2026: Flexibility (ANALYSIS)