A Sierra Club analysis, first reported by Bloomberg on October 28, 2026, says AI-driven data-center growth could push US power-sector CO2 emissions to about 2.1 billion metric tons by 2030, roughly 42% above 2025 levels and enough to erase about 15 years of electricity-sector emissions cuts. The figures reached us second-hand and describe a high-impact scenario, not a prediction. We could not open the report, so this post separates what is reported, what the coverage itself flags as uncertain, and what you can check.
TL;DR: the numbers and their status
| Claim (2030) | As reported | Status |
|---|---|---|
| US power-sector CO2 | About 2.1 billion metric tons, about 42% above 2025 | Reported via Bloomberg and aggregators; report not seen |
| "15 years of progress erased" | Framing of the same increase | A characterization of the 42% rise |
| Off-grid data-center generation | More than 460 million metric tons a year | Reported; the "like Mexico's output" comparison applies to off-grid only |
| Grid-connected data-center power | About 154 million tons, a roughly 10% increase | Reported |
| Nature of the estimate | High-impact scenario tied to proposed and pending projects | Stated in coverage |
| Methodology | Not described in any coverage we read | Unknown |
What was actually published
The chain of sourcing matters here. Per the Yahoo write-up by Alex Barrientos, published October 30, the analysis is the Sierra Club's and was first covered by Bloomberg on October 28. A Gadget Review summary says the same, citing Bloomberg with additional reporting from the Detroit News, edie.net and Tom's Hardware. We did not find the Sierra Club's own report or press release in our searches, and the Bloomberg piece sits behind a paywall.
So there are three layers: the Sierra Club's analysis, Bloomberg's report on it, and aggregators summarizing Bloomberg. Every number below comes from the third layer unless noted. When the primary document appears, check the figures against it.
What the coverage says
According to the secondary reports:
- Total. US power-sector emissions could reach 2.1 billion metric tons of CO2 by 2030, about 42% above 2025.
- Off-grid generation. Generators built by or contracted for data-center developers, mostly fossil-fueled, could emit more than 460 million metric tons a year. The coverage compares this to Mexico's annual emissions, a comparison that applies only to the off-grid slice.
- Grid-connected power. Data-center demand served by the grid could add about 154 million metric tons, roughly a 10% increase on its own.
- Earlier estimates. Tom's Hardware, as relayed by Yahoo, says earlier projections ranged from one-quarter to one-half of the Sierra Club's figure.
One apparent mismatch deserves a flag. The off-grid slice (460 million tons) plus the grid slice (154 million tons) is about 614 million tons, which is close to the 42% increase, roughly 618 million tons if 2025 emissions are near 1.5 billion tons. That suggests the two pieces make up the headline increase, but no coverage we read states the relationship explicitly. It is our arithmetic, not the Sierra Club's.
It is a scenario, not a forecast
Coverage is consistent on this point, and it is the most important caveat. The Yahoo article calls the estimate a "high-impact scenario tied to proposed and pending projects, not a guaranteed outcome." Brendan Pierpont, director of electricity at Energy Innovation, told the Detroit News the outcome is "possible but not inevitable," and listed turbine backlogs, interconnection delays and project cancellations as things that could reduce the final number.
That matches what we have reported on the supply side. Transformer lead times of up to 160 weeks limit how fast new capacity can connect, and projects do get dropped, as in Microsoft's exit from the Nscale Loughton site over grid power. Both cut against a build-everything-on-schedule assumption. The Sierra Club's scenario is meaningful as an upper-range warning, not as a base case.
Why off-grid gas drives the number
The largest slice comes from power plants that sit outside the grid. Developers build their own gas generation because grid interconnection for a large load can take years, and speed to power decides who wins capacity. The coverage names this "behind-the-meter" generation as the main driver. It also explains why this scenario is so much higher than earlier ones: older estimates leaned on grid-connected demand.
We have seen what such sites look like on the ground. In our SpaceX Colossus 2 claimed-vs-verified piece, a Reuters report cited by AI Weekly said dozens of gas turbines were installed at a Memphis-area site without federal clean-air permits while a larger permanent plant was planned. Whether that pattern is typical is exactly the kind of thing a scenario like this tries to scale up.
What we cannot verify
- The report itself. Not found; no direct quotes from Sierra Club staff in the coverage we read.
- The model. Which projects count as "proposed and pending," how likely each is to be built, what emissions factors apply, and what 2025 baseline is used.
- The "15 years" framing. It is arithmetic on the 42% rise against the historical decline in power-sector emissions, and depends on the baseline.
- Any reduction scenario. Coverage describes a high-impact case; we did not see the low or central cases.
What people are asking
Does this mean AI itself is the cause? The scenario is about data centers in general, of which AI is the fastest-growing driver. The coverage ties the growth to AI demand, but it does not split the load between AI training, AI inference and ordinary cloud computing, so the figure cannot be read as "AI alone."
Is 42% by 2030 consistent with other estimates? Not with the older ones. Tom's Hardware, as relayed by Yahoo, says earlier projections were one-quarter to one-half of this figure. That gap is itself the story: the range of credible estimates is wide, and it moves quickly as developers announce new off-grid plants. A single number from a single advocacy group is one data point in that range.
Who produced it, and does that matter? The Sierra Club is an environmental advocacy group, and its analysis is built to inform a policy argument. That does not make the numbers wrong, but it means the choice of a high-impact scenario is a framing decision. Read it next to estimates from grid operators, utilities and industry groups, who often model lower outcomes for their own reasons.
What would lower the number? The expert quoted in the coverage points to delivery delays for turbines, slow interconnection approvals and cancelled projects. Efficiency gains in chips and software could also cut the electricity needed per unit of AI work, though the coverage does not model that. Faster renewable build-out and storage would change the mix, but nothing in what we read says by how much.
What would raise it? Faster approvals for gas plants, more projects choosing off-grid power to skip the interconnection queue, and demand growing faster than the planned projects imply.
How this fits the politics
The numbers arrive as electricity costs and data centers are a live campaign issue. Our update on the midterm ad war over data centers covers the political side, and Hochul and AOC's New York moratorium push shows where this turns into policy. A study that says data centers raise emissions gives moratorium backers evidence and gives industry a target to rebut. Expect both sides to cite the same number with different caveats.
The same coverage mentions Ireland's rule requiring 80% renewable sourcing for new data centers, with a six-year phase-in, as an example of a policy response, and questions whether the timeline keeps pace with construction. That is secondhand through Tom's Hardware and EirGrid; check the EirGrid policy for the exact terms.
What this means for people who build with AI
If you run workloads, nothing here changes your bill this week. It does point at three things worth watching:
- Siting and power cost. Capacity is going where power can be had fastest, often off-grid gas. That affects where regions get built and how reliable they are.
- Customer scrutiny. Enterprises with emissions targets increasingly ask cloud providers for location-based figures. Gas-backed capacity makes those answers harder.
- Regulation risk. Moratoriums and permit fights can delay capacity, which feeds back into availability and price. Our climate and energy overview covers the efficiency side.
How to check the claim
- Find the Sierra Club's original report or press release and compare the 2.1 billion, 460 million and 154 million figures.
- Read Bloomberg's October 28 article, or a licensed summary, for the methodology Bloomberg saw.
- Compare against independent projections from grid operators and the federal energy agency, using the same baseline year.
- Track project cancellations and turbine delivery dates; they decide how much of the scenario materializes.
Bottom line
The reported headline is large and plausible as an upper-range scenario: 2.1 billion tons by 2030, about 42% above 2025, driven by off-grid gas. It is also unverified by us, built on proposed projects that may not be built, and relayed through at least two layers of coverage. Cite it as "a Sierra Club scenario, as reported by Bloomberg," not as a forecast, and revisit when the original report is public.
Related reading
- Transformer lead times of 160 weeks and AI data centers
- Microsoft pulls out of Nscale Loughton over grid power
- SpaceX Colossus 2: 1.2 million GPUs, claimed vs verified
- Data-center backlash in the 2026 midterms
- Hochul and AOC on a New York data-center moratorium
- AI, climate change and energy
Figures are as reported on October 28 to 30, 2026 and have not been checked against the original report.