The EIA’s Annual Energy Outlook 2025 provides a solid baseline for understanding U.S. energy trends, but it’s built on a business-as-usual framework that filters out geopolitical, financial, governance, and ecological risks. It’s a conversation starter—not a roadmap. In the real world, shocks and disruptions are standard fare. The future will likely be more chaotic than the AEO implies.
Models are useful because they anchor discussion in current assumptions. “Model Land” has its flaws, but they’re more manageable than the chaos of “Opinion Land.”
One of the key updates in this year’s AEO is the outlook for U.S. crude and condensate production. The 2023 report projected a long, flat “peak plateau” (red line in Figure 1) around 13 million b/d through 2050. AEO 2025 shifts to a sharper peak at 13.9 million b/d in 2027 followed by a gradual decline to 11.2 million b/d by 2050. Even so, 2050 output still exceeds pre-2019 levels.
A bigger change is on the demand side (Figure 2). The EIA now sees U.S. energy consumption peaking in 2026 (blue curve) and declining 8% by 2040—a reversal from the upward trajectory in its previous report (red curve) .
Oil consumption drops 20% by 2050, with natural gas taking the top energy spot after 2042 (Figure 3). Wind and solar rise to 16% of total energy use. Coal collapses. Nuclear drops 7%, while geothermal and hydrogen barely register. Hydropower remains unchanged.
Electric power forecasts are particularly dramatic. Wind and solar generation are expected to increase 3.5 times from 2024 levels (Figure 4). Coal is eliminated. Natural gas drops 39%. Nuclear declines 5%. But these forecasts don’t align with growing demand from data centers that require round-the-clock power—not intermittent renewables.
This and other inconsistencies led me to rework the EIA’s assumptions, relying strongly on current law rather than EPA and state regulations. My projections show wind and solar 23% lower than the EIA’s forecast. Nuclear remains flat. Coal use remains much higher post-2030 (Figure 5).
The EIA projects a 30% cut in carbon emissions by 2050, largely due to reduced coal use (Figure 6).
But applying less aggressive assumptions cuts emissions to less than 10% from present levels—still progress, but far short of the AEO’s optimistic tone (Figure 7).
EIA provides 11 additional—or “side”—projection cases per energy source in addition to its reference case. These focus narrowly on economic factors like oil prices and tech costs. Systemic risks are ignored or only vaguely implied. This undermines their usefulness in an unpredictable world.
EIA’s side cases for coal mostly assume steep declines (Figure 8). Only the “alternative electricity” case—where the EPA’s 2024 carbon rules don’t take effect—breaks from the view that coal is dead. That outcome seems plausible enough to justify a broader range of scenarios. My most-likely case is shown as “Coal AEB” in red.
Another major shortcoming in AEO 2025’s analysis—one with big implications for carbon emissions—is its assumption about gasoline use. The outlook expects gasoline demand to fall 45% by 2050, with 2030 usage below 1999 levels and 2040 near 1971 levels (Figure 9), mainly due to rising EV adoption.
Gasoline projections suffer from the same bias as coal: steep declines in most cases, inconsistent with history—except for the one scenario tied strictly to existing law–“alternative transportation” (Figure 10).
The report assumes electric vehicles will rise from 3.4% of cars in 2024 to 15% by 2030 and 50% by 2050 (Figure 11). Conventional cars drop from 109 million to 87 million by 2030, and to only 38 million by 2050 (where do all of those cars go?). Transportation emissions fall 555 million tons, or 20% of all U.S. reductions.
EIA doesn’t model alternative scenarios for vehicle stocks, so I created my own. Less aggressive early adoption assumptions reduce EV totals by 7 million (15%) in the P50 case and 14 million (30%) in the P75 case by 2050 (Figure 12).
AEO 2025 gives us something concrete on which to anchor discussions about the energy future. But its strength as a baseline hides a critical flaw: it assumes the future will broadly resemble the recent past—a risky bet in a world facing rising instability. It also seems to overlook the growing economic stress on American households, the fading enthusiasm for renewables and EVs, and the current administration’s rollback of many of the policies and subsidies that once supported them.
I understand how hard it is to build scenarios that capture these risks—they’re messy, nonlinear, and unpredictable. But avoiding them by leaning on business-as-usual assumptions doesn’t make them disappear. The further out we look—especially to 2050—the more dangerous it is to rely on models that sidestep geopolitical shocks, financial instability, ecological limits, or governance breakdowns.
What AEO 2025 gives us is a progress narrative: a long arc of decarbonization, technological adoption, and economic stability. But that arc depends on a world order that is rapidly disintegrating. In that context, the probability that any baseline—let alone a best-case scenario—will resemble the next five years, much less the next 25, is extremely low.
This doesn’t mean we discard modeling or abandon scenario work. On the contrary, it means we need more of it—and it must be more plural, more grounded in legal and institutional constraints, and more willing to include scenarios shaped by disorder, discontinuity, and failure. If we want to plan for the future instead of being surprised by it, we have to stop treating volatility as an outlier. It’s the main feature of the terrain now.
AEO 2025 is a useful tool, but it’s not a roadmap. It’s a map of a world that assumes order will hold. We’d be wise to sketch a few other maps too.