How Technology Costs Shape the Emissions Impact of High Data-Center Load

A NEMS Connect sensitivity on data-center electricity growth, zero-carbon technology costs, and power-sector emissions

The question after high load

Data-center electricity demand has become one of the most important uncertainties in U.S. energy outlooks. In the U.S. Energy Information Administration’s Annual Energy Outlook 2026, data-center server electricity use rises substantially across cases, reaching between 446 billion kilowatt-hours (BkWh) and 818 BkWh by 2050. The upper end of that range occurs in EIA’s High Electricity Demand case.

In a previous NEMS Connect sensitivity, we pushed data-center server electricity demand beyond EIA’s High Electricity Demand case to test how the integrated U.S. energy system responds. That analysis showed higher electricity generation, higher capacity additions, and higher power-sector carbon dioxide emissions.

That result raised a second question:

What determines how the system serves that load?

This article explores that second question. We combine the higher server-load stress with EIA’s Low Zero-carbon Technology Cost assumptions to examine how lower-cost zero-carbon technologies change the power-sector result. We also use EIA’s High Zero-carbon Technology Cost assumptions as an illustrative high-cost comparison.

The comparison is not meant to forecast the future. It is meant to demonstrate how a detailed model such as NEMS can be used to learn which assumptions matter when high electricity demand, generation choices, capacity expansion, prices, and emissions interact.

For academic energy researchers, research-program sponsors, policy analysts, and modeling practitioners, the value of this exercise is not simply the numerical result from one sensitivity. The value is the sequence of questions it illustrates: first test the load, then test the technology-cost environment, and then examine how the response moves through generation, capacity, prices, and emissions.

How the sensitivity was structured

The comparison uses four result cases: one official EIA benchmark and three NEMS Connect sensitivities. The NEMS Connect sensitivities use the AEO 2026 framework and apply selected EIA side-case assumptions to a higher server-load stress case.

Table 1. Case names used in this article

EIA uses the Low and High Zero-carbon Technology Cost cases to address uncertainty in future costs for zero-emissions power-generation technologies. These cases should not be read as probability-weighted bounds. They are structured case assumptions that allow analysts to examine how NEMS responds when specified technology-cost assumptions change.

Using EIA’s technology-cost cases as structured assumptions

In EIA’s Low Zero-carbon Technology Cost case, capital and fixed operating costs for a suite of zero-carbon technologies decline more rapidly than in the Counterfactual Baseline, reaching 40 percent below Counterfactual Baseline levels by 2050. The technologies include solar, wind, diurnal storage, hydropower, geothermal, biomass with and without carbon capture and storage, advanced nuclear, small modular reactors, end-use renewable technologies, and hydrogen electrolyzers. EIA’s High Zero-carbon Technology Cost case applies the high-cost side of that structured assumption set.

NEMS Connect applies those EIA case assumptions to a higher server-load sensitivity to ask a targeted modeling question:

Under very high data-center server electricity demand, how much does the technology-cost assumption change generation, capacity,  prices, and emissions?

The three NEMS Connect high-load sensitivities have nearly the same data-center server purchased electricity in 2050, ranging from about 1,086 to 1,089 BkWh. That means the main differences among the NEMS Connect cases are not driven by different server-load assumptions. They are driven by how the system serves that high load under different technology-cost assumptions.

The same load, different power-sector outcomes

The central result is that similar high-load assumptions produce substantially different power-sector outcomes.

Table 2. Selected 2050 results

The key comparison is between the high-load sensitivities. In the Low ZTC sensitivity, natural gas generation falls by about 911 BkWh relative to the HED Stress sensitivity, while renewable generation rises by about 972 BkWh. Electric power-sector CO₂ falls by about 305 million metric tons (MMmt), and the average electricity price falls by about 0.6 cents per kilowatt-hour.

In the High ZTC sensitivity, the response moves in the opposite direction. Natural gas generation rises by about 171 BkWh relative to the HED Stress sensitivity, renewable generation falls by about 184 BkWh, and electric power-sector CO₂ rises by about 56 MMmt.

The broader comparison between the Low ZTC and High ZTC sensitivities is especially informative. By 2050, the difference between those two technology-cost assumptions is about 1,082 BkWh of natural gas generation, 1,156 BkWh of renewable generation, 361 MMmt of power-sector CO₂, and about 0.7 cents per kilowatt-hour in the average electricity price.

Figure 1. Lower zero-carbon technology costs shift generation from gas to renewables

The generation response provides the clearest first result.

In the HED Stress sensitivity, the model serves the higher load with a mix of natural gas and renewables. In the Low ZTC sensitivity, lower zero-carbon technology costs shift the generation mix strongly toward renewables. Natural gas generation falls to about 2,035 BkWh in 2050, while renewable generation rises to about 4,272 BkWh.

The High ZTC sensitivity moves in the opposite direction. Natural gas generation rises to about 3,117 BkWh, while renewable generation falls to about 3,116 BkWh. The result is a near-even split between natural gas and renewables in 2050, rather than the renewable-heavy outcome in the Low ZTC sensitivity.

This is the first major lesson from the analysis. High server load does not by itself determine the emissions result. The emissions result depends on what technologies are available and economic when the system expands to meet that load.

Figure 2. The lower-emissions path requires a different capital stock

The capacity results show the infrastructure side of the same story.

In the Low ZTC sensitivity, total electric power-sector capacity rises to about 2,568 GW by 2050, compared with about 2,268 GW in the HED Stress sensitivity. Much of that increase comes from renewable capacity and diurnal storage. Diurnal storage rises to about 282 GW in the Low ZTC sensitivity, compared with about 105 GW in the HED Stress sensitivity.

In the High ZTC sensitivity, total capacity is lower than in the Low ZTC sensitivity, and storage remains close to the HED Stress level. Combined cycle capacity is higher than in the Low ZTC sensitivity, reflecting a greater reliance on natural gas generation.

This result is important because a lower-emissions outcome is not simply a dispatch result. It requires a different capital stock. Lower zero-carbon technology costs lead the model to build a larger renewable and storage system, while higher zero-carbon technology costs leave the system more dependent on gas-fired resources.

Figure 3. Technology costs change the emissions outcome

The emissions results follow from the generation and capacity response.

In 2050, electric power-sector CO₂ emissions are about 986 MMmt in the HED Stress sensitivity. In the Low ZTC sensitivity, they fall to about 680 MMmt, below both the HED Stress sensitivity and the EIA Counterfactual Baseline. In the High ZTC sensitivity, they rise to about 1,041 MMmt.

This is a central policy-relevant result. In this sensitivity, the higher server-load stress does not have a fixed emissions consequence. Under lower zero-carbon technology costs, the model more than offsets the emissions increase associated with the high-load stress. Under higher zero-carbon technology costs, the system leans more heavily on natural gas and emissions increase.

This does not mean that lower costs are guaranteed, or that the Low ZTC sensitivity is the most likely future. It means that technology costs are a key driver of whether high electricity demand produces a high-emissions or lower-emissions power-sector outcome.

What the sensitivity teaches 

The main lesson is not that one sensitivity is “right” and another is “wrong.” The lesson is that the system response is conditional.

High data-center server load creates the need for additional electricity supply. But the emissions outcome depends on the technologies used to provide that supply. If lower-cost zero-carbon technologies are available, the system that the model represents can meet high load with substantially more renewables and storage, lower gas generation, lower electricity prices, and lower emissions. If those technologies remain more expensive, the model relies more heavily on raising gas-fired generation, raising power-sector CO₂ emissions.

High load creates the challenge, but technology costs shape the outcome

How to read these results

These NEMS Connect sensitivities are not official EIA projections. They are independent model runs using the AEO 2026 public NEMS framework, designed to explore how NEMS responds when selected assumptions are combined or extended beyond EIA’s published side cases.

Official EIA values used for the Counterfactual Baseline and High Electricity Demand case are drawn from EIA’s published AEO 2026 data tables. NEMS Connect values are drawn from NEMS Connect model runs using the AEO 2026 framework.

The NEMS Connect sensitivities are interpreted as structured comparisons against official EIA benchmarks. They are intended to illustrate the direction and approximate scale of the model response under specified assumptions, not to provide probability-weighted forecasts.

The modeling insight from this work

The emissions consequence of data-center load growth is not determined by load alone. It depends on how the energy system responds.

In these NEMS Connect sensitivities, very high server electricity demand produces different outcomes under different zero-carbon technology-cost assumptions. Lower zero-carbon technology costs shift the model toward more renewables and storage, less natural gas generation, lower electricity prices, and lower CO₂ emissions. Higher zero-carbon technology costs move the response back toward greater reliance on gas-fired generation and higher emissions.

More insights from these runs to follow:

The power-sector result is only part of the integrated response. Because NEMS links electricity and fuel markets, the next article follows how these technology-cost assumptions affect natural gas demand, prices, production, and LNG exports.

Sources

U.S. Energy Information Administration, Annual Energy Outlook 2026. Official AEO 2026 case tables and downloadable data. https://www.eia.gov/outlooks/aeo/

U.S. Energy Information Administration, Annual Energy Outlook 2026 Case Descriptions. Defines the Counterfactual Baseline, High Electricity Demand case, and zero-carbon technology cost cases. https://www.eia.gov/outlooks/aeo/assumptions/pdf/case_descriptions.pdf

U.S. Energy Information Administration, “Data center server energy use grows across the United States in all AEO 2026 cases.” Provides EIA’s published 2050 data-center server electricity range, including 446 BkWh to 818 BkWh. https://www.eia.gov/todayinenergy/detail.php?id=67704

NEMS Connect analysis of AEO 2026-based sensitivities. NEMS Connect High Electricity Demand Stress, High Server Load + Low Zero-carbon Technology Cost, and High Server Load + High Zero-carbon Technology Cost sensitivities.  https://nemsconnect.com/data-centers-are-different-what-a-high-demand-stress-case-shows-about-ai-load-growth/