Read the explainer about Data Centers and AI in En-ROADS for an introduction to the topic and how to use the feature in En-ROADS. The FAQs below go deeper: modeling, Baseline assumptions, exploring different scenarios, and what falls outside scope—local and environmental equity impacts, AI-climate solutions, and enabled emissions.


TABLE OF CONTENTS


Model & Scope questions

What's new about how En-ROADS models data centers? 

Previously, energy demand from data centers was folded into the broader Industry sector rather than tracked on its own. As data center growth accelerated—driven largely by AI—and better data on their energy demand and emissions became available, we separated them into their own end-use sector: Industry is now split into Traditional Industry and the Data Center Industry. This lets you see data center energy and emissions explicitly and test assumptions about how fast the sector grows, how efficient it becomes, and how much AI boosts the broader economy.


Does this sector cover AI-focused data centers, or all data centers? 

All data centers, not AI exclusively. The sector includes cloud computing and storage, internet services, web hosting, financial transactions, and other digital infrastructure, alongside AI workloads, which are the fastest-growing driver of demand. 


Where do I find the data center sector in En-ROADS?

Data center and AI-specific graphs can be found under Graphs > Demand Sectors and End Uses. Some graphs, such as the “Total Final Consumption by End Use” show data centers separate from Traditional Industry. In graphs where data centers are not broken out separately, they are included in Industry. The assumption sliders that control data center growth, costs, efficiency, and AI's economic boost are under Simulation > Assumptions > Economy > Data centers including artificial intelligence (AI). 



Baseline questions


Baseline temperature appears unchanged with the AI addition. Does that mean the emissions are insignificant?

No; it means the emissions were already there. This update disaggregates it rather than adding new emissions, so the Baseline temperature doesn't change. Before this release, data center energy demand was included in En-ROADS inside the broader Industry sector. Our projections accounted for AI energy demand to the extent that our calibration sources included it in their electricity growth forecasts. 


What data center Assumptions are more realistic or reasonable? 

There is significant uncertainty across nearly all the key parameters. The "most realistic or reasonable" values depend on how quickly AI capabilities improve, how costs evolve, and how broadly the technology diffuses across the economy. The defaults in the Baseline Scenario are calibrated to current estimates and near-term projections from sources like the IEA, DNV, and Deloitte. Rather than searching for a single right answer, we encourage you to test the ranges. The sliders are deliberately wide because credible published projections for 2030 span more than an order of magnitude. 



Exploring Scenarios


Can I simulate a scenario where AI is restricted and its growth is limited? 

En-ROADS has no explicit "restrict AI" policy lever, but you can produce a reduction in compute capacity by moving the assumption sliders, such as in this scenario. Lower the "Strength of cumulative data center use on compute intensity of GDP" slider (slower diffusion into the economy) and the "Strength of cost compute intensity of GDP" slider (weaker demand response to falling costs). Raising the progress ratios (slower cost and energy intensity declines) also dampens growth. Together these approximate a future where regulation, permitting limits, or market saturation constrain the sector. 


Can AI result in economic damage rather than a boost? 

The AI productivity slider only adds a boost (its minimum is zero), so En-ROADS doesn't directly model AI reducing economic output. To explore that possibility, use the "Economic Growth" slider as a proxy. 


Why does the model assume data centers are powered entirely by electricity? What about on-site natural gas generators or dedicated nuclear/renewables? 

The vast majority of data centers draw their power from the electric grid, so for modeling simplicity En-ROADS assumes the sector runs entirely on grid electricity, with emissions tracking the grid's carbon intensity. Some operators are pursuing on-site power generation—using gas (sometimes with CCS), nuclear, or renewables—but En-ROADS does not yet have a dedicated lever to represent this. As a result, data center emissions in the model respond to every supply-side policy in the simulator, which is why grid decarbonization is a powerful lever on data center emissions. 


How does En-ROADS account for efficiency improvements in chips and data center design over time? 

Through learning curves. The "Progress ratio (Energy intensity)" slider controls how fast the energy used per unit of computation falls with cumulative experience. A progress ratio of 0.8 means a 20% reduction in the energy intensity of new data centers, AI hardware, and algorithms each time capacity doubles. A separate "Progress ratio (Infrastructure)" slider does the same for the non-energy cost of building and running data centers. 


Why is the AI economic productivity boost set to zero in the Baseline? Doesn't AI already affect the economy?

The Baseline GDP projections already reflect historical productivity trends, which include computing to date. The slider represents an additional, AI-specific boost to economy-wide productivity, and estimates of that boost vary significantly, with little consensus yet on its size or timing. Because it is so uncertain, it defaults to zero, and you can use the "Maximum economic productivity boost from AI" and "Delay in economic productivity boost from AI" sliders to test how a productivity surge would ripple through energy demand and emissions. 


If I decarbonize the grid, does that automatically reduce data center emissions too? 

Yes. En-ROADS assumes data centers draw from the grid, so their emissions carry the grid's carbon intensity directly. For example, apply a carbon price, clean electricity standard, or renewable and nuclear subsidies, and data center emissions intensity will fall even while their electricity demand keeps rising. 


Data center emissions look small compared to other sectors. Does that mean we should not be concerned about them?

Small shares of a very large total still matter when every gigaton counts—2.44 GtCO₂ cumulatively through 2025 in the Baseline—nearly 1% of the remaining 1.5°C carbon budget in 2025—and they are one of the fastest-growing sources of electricity demand globally. Other reasons for attention: their emissions are highly solvable (data centers run on electricity, so grid decarbonization directly cuts them); their indirect effect could be larger than their direct one (if AI boosts productivity across a carbon-intensive economy, the resulting economy-wide emissions increase can exceed anything happening inside the data centers themselves); and there are important social and environmental considerations, such as water consumption and energy security. 


Why CO₂ emissions from Energy for Data Centers in the Baseline grow faster in the short term, then stabilize, and keep rising at a faster pace later in the century?

In the short term, compute intensity of GDP grows rapidly as AI and scaled computing diffuse across sectors of the economy, and the grid remains largely carbon-intensive, so emissions rise steeply. By mid-century, two forces catch up: the energy intensity of compute capacity and the carbon intensity of electricity both decline, flattening emissions growth. After 2050, emissions climb again as rising GDP per capita and population growth drive up compute demand in absolute terms, while the pace of grid decarbonization slows. 



Out of scope from En-ROADS


What are the local equity considerations of AI and data centers? 

En-ROADS is a global model and doesn't capture local impacts, but they are significant. Data centers cluster geographically, so a single cluster can stress a regional grid and raise electricity prices, disproportionately affecting low-income households. When demand is met by fossil fuels or backup diesel generators, air pollution affects communities nearby. Data centers also consume large amounts of water, an issue in water-stressed areas. You can read more equity considerations in the Data Center and AI in En-ROADS Explainer. 


What are the policies to control or drive the growth of AI and data centers? 

En-ROADS doesn't include AI-specific policy levers, but real-world policies map onto its assumptions: efficiency standards and R&D support correspond to the progress-ratio sliders; permitting limits, siting restrictions, or compute governance correspond to lowering the growth-strength sliders; and clean-power procurement requirements act like grid decarbonization policies, which the model handles directly. Use the assumption sliders as proxies to represent the outcome of such policies. 


How do I model a scenario where AI accelerates the adoption of climate solutions?

En-ROADS doesn't model this directly yet, but you can test the idea with existing sliders: pair an AI economic productivity boost with accelerated progress elsewhere—for example, stronger energy efficiency, a technological breakthrough in a new zero-carbon energy source, or cheaper renewables—to represent AI speeding up climate solutions. Comparing that scenario against the boost alone shows how large the AI-for-climate effect would need to be to offset the added growth. 


How do I model a scenario where AI improves efficiency of the oil&gas sector and finds new reserves?

This mechanism, sometimes called "enabled emissions," is not modeled in En-ROADS yet. As a proxy, you can lower the cost of fossil fuels (for example, subsidizing oil and gas) to represent AI making extraction cheaper and expanding recoverable reserves. For background, see Enabled Emissions and Turliuk & Sterman (2026), which covers this pathway alongside the ones En-ROADS currently models. 


Doesn't the geography of where data centers are placed determine their emissions impact? How do you account for that? What about local electricity price spikes?

En-ROADS is a global model, so it doesn't represent where data centers are sited, regional grid constraints, or local electricity prices—an acknowledged limitation. In reality, siting matters a great deal: a data center on a coal-heavy regional grid has a much larger footprint than one on a clean grid, and concentrated demand can spike local prices (aggregated at the global level in the Price of Electricity graph in En-ROADS). In the model, all data centers share the global grid mix, which responds to the energy supply policies you apply. For a local-scale story, you can read this WRI article.