The renewable rollback looks like a managed slowdown. The build-out data says otherwise.
In late 2025, the United States added more solar capacity to its grid than in any comparable period in history. Utility-scale solar alone accounted for roughly 75% of new generation coming online. Wind additions outpaced natural gas. Battery storage nearly doubled year-over-year. By the time the full-year numbers landed, the US had put up somewhere north of 53 gigawatts of new clean capacity — a record, by a distance.
This was happening while the most aggressively anti-renewable administration in modern history was in office.
The One Big Beautiful Bill had just killed the core clean energy tax credits. Offshore wind permitting was frozen. The EPA Endangerment Finding — the legal backbone of federal climate regulation since 2009 — had been repealed. Solar project costs were up 20–40% from new tariffs. And yet the grid kept filling with solar panels and battery banks, month after month, at a pace that would have seemed implausible three years earlier.
The consensus interpretation of this paradox is reassuring: the transition has momentum that policy can’t easily stop. Renewables are now the cheapest new power source, full stop. The market will find a way. The rollback will slow the transition but not reverse it. Even the most pessimistic credible analysis — Rhodium Group’s modeling of the OBBBA — frames the damage as a 57–62% decline in clean capacity additions over the decade. A slope, not a cliff. A painful unwinding, not a rupture.
This is the working hypothesis I want to challenge.
The working hypothesis: The US grid is heading for a discrete capacity gap beginning in 2028 — not a gradual slowdown, but a measurable step-down concentrated in high-demand regions — as the pre-OBBBA pipeline exhausts and post-OBBBA additions fail to replace it at scale. The planning data is already pricing in the first edge of this cliff. Almost no one is modeling the second.
The Slope That Isn’t One
The Rhodium decade-average is doing a lot of narrative work that the underlying data doesn’t support.
When you break the modeled decline out by year — using BloombergNEF’s deployment curve, which Rhodium’s own scenario analysis is broadly consistent with — the shape stops looking like a slope. What you get instead is a front-loaded pipeline burning down in 2026 and 2027, followed by a hard drop.
The numbers: approximately 86 gigawatts of new capacity planned for 2026 — a record — with a comparable surge in 2027 as developers race to beat credit deadlines. Then 2028: roughly 48 gigawatts. That’s a 41% single-year decline. A step function, not a glide path.
The Rhodium decade-average looks like 57–62% because the math is doing exactly what averages do: smoothing a front-loaded spike and a subsequent collapse into a single, manageable-sounding number. The pipeline isn’t declining gradually. It’s being burned down in two years, after which the replacement rate drops off a ledge.
This is the cliff. And there are actually two edges, not one.
The Two Deadlines Nobody Has Stacked
The clean energy press has covered the July 4, 2026 deadline extensively. To claim the technology-neutral tax credits established by the IRA, wind and solar projects had to begin construction within 12 months of the OBBBA’s enactment. Miss that date and your project has to be placed in service by December 31, 2027 — a timeline that is, for most utility-scale projects, physically impossible. This is the first edge. It’s real, it’s documented, and the industry is in a full sprint to meet it.
The second deadline has received far less attention, and it’s the one that determines what happens after the sprint ends.
Buried in the OBBBA’s Foreign Entity of Concern provisions is a separate construction start date: December 31, 2025. Projects that began construction before that date are grandfathered from the new material sourcing requirements. Projects that began construction after it are not. And those requirements are not trivial.
Under the MACR schedule — the Material Assistance Cost Ratio that the OBBBA introduced — solar and wind projects starting construction in 2026 must source at least 40% of their component value from non-Chinese entities. That threshold rises to 50% in 2027, 60% in 2028, 70% by 2029. Battery storage requirements are even more stringent, starting at 55% in 2026 and climbing to 85%.
Now stack that against the actual supply chain. China produces approximately 91–95% of global polysilicon. It controls 97% of wafer manufacturing. It produces more than 90% of solar cells globally. The threshold for getting a post-2025-start project across the FEOC line isn’t demanding — it’s, for most of the supply chain, functionally impossible without a near-complete restructuring that doesn’t yet exist.
Here is the piece of data that makes this precise: according to Latitude Media’s coverage of LevelTen Energy’s analysis, approximately 76% of solar projects slated for completion by end of 2028 already began construction before 2026. Those projects are grandfathered. They clear the FEOC threshold automatically. Which means the existing pipeline — the one producing the record 2026 and 2027 numbers — is largely fine.
The cliff isn’t the existing pipeline failing. It’s the replacement rate collapsing once that pipeline exhausts. Every project that starts construction after December 31, 2025 has to navigate an escalating compliance gauntlet against a supply chain that is 80–95% controlled by a prohibited foreign entity. The 86-gigawatt 2026 record is the last flare of a pipeline that was locked in before any of this applied.
After it burns down, what refills it?
The Physical Work Test Closed the Door
There was a valve that could have softened this. Under the old IRA rules, developers could “begin construction” for tax credit purposes by spending 5% of a project’s total costs — placing equipment orders, signing contracts, committing capital. This let developers nominally start hundreds of gigawatts of projects even without a shovel in the ground, effectively papering the pipeline forward against future deadlines.
The Trump administration closed it.
On July 7, 2025 — three days after signing the OBBBA — the President issued an executive order directing Treasury to tighten the definition of “begin construction” and explicitly prohibit the kind of safe-harbor maneuvering that the 5% test allowed. Treasury’s subsequent guidance, IRS Notice 2025-42, did exactly that: for projects over 1.5 megawatts, the sole valid method for establishing construction commencement is now the Physical Work Test. Significant physical work. On-site or in a factory. Verifiable.
This matters for the cliff thesis in a specific way. The paper-start loophole would have let developers buffer the post-cliff period — banking tax credit eligibility on hundreds of gigawatts of projects that weren’t ready to build. The executive order eliminated that option. The projects that will get built are the projects with actual dirt turned, actual steel in the ground, before July 4, 2026.
The EVA analysis estimates roughly 43 gigawatts of clean capacity with approved interconnection agreements could feasibly meet both OBBBA eligibility thresholds near-term. Another 187 gigawatts has a “possible path.” But the historical solar interconnection completion rate — the share of projects that enter the queue and actually come online — is approximately 14%, per Lawrence Berkeley National Laboratory’s interconnection queue data. Most of that 187 gigawatts will not make it.
The Grid Already Knows
The most important piece of evidence for the cliff thesis is not a forecast. It’s already happened.
In December 2025, PJM Interconnection — the grid operator serving 65 million people across 13 states and the District of Columbia, the largest competitive wholesale electricity market in the world — ran its annual capacity auction. For the first time in PJM’s history, the auction came up short. The 2027/2028 delivery year fell short of reliability requirements.
PJM is not a bearish source. It is not an environmental advocacy group or a clean energy trade association. It is the organization whose job is to keep the lights on for one in five Americans, and it has a structural incentive toward optimism — a shortfall finding is, for a grid operator, an admission of failure. The 2027/2028 shortfall was its first.
This is the first edge of the cliff showing up in planning data — before the OBBBA pipeline has even exhausted, before the physical work test has filtered out the paper projects, before the FEOC thresholds have started biting. The grid’s own accounting is already registering the gap.
PJM is not alone. Morgan Stanley projects a 44 gigawatt US power shortfall through 2028. NERC’s 2025 Long-Term Reliability Assessment warns of capacity shortfalls as early as 2028. Schneider Electric projects peak supply falling short of demand by 2028, with the gap reaching 175 gigawatts by 2033.
These analyses are not coordinated. They were produced independently, using different methodologies, with different political valences. They converge on the same year.
And the demand side is not waiting. Lawrence Berkeley National Lab projects data centers alone could account for 12% of US electricity demand by 2028. Rhodium’s modeling puts data centers at 47–65% of total electricity demand growth in 2030. PJM’s own peak load estimate for 2030 has increased from 158 gigawatts in 2023 to 183 gigawatts in 2026. The loads are materializing faster than the models expected. The supply additions are about to fall off a cliff.
This is not a story about climate. It’s a story about arithmetic.
What the Consensus Is Missing
The mainstream framing of the renewable rollback has two components that are both technically accurate and, together, misleading.
The first: renewables are now cost-competitive without subsidies. Utility-scale solar runs 4–8 cents per kilowatt-hour even without the IRA credits. That’s cheaper than new gas at 14–26 cents. The transition has fundamental economics on its side. True.
The second: the Rhodium decade-average shows a 57–62% decline, which is large but not apocalyptic. The market will build what pencils out. True.
What the consensus framing elides is the timing mismatch. The transition’s economics are correct on a long enough horizon. But the cliff is a 2028 problem. And the only energy technology that could actually close a 2028 gap on a 2028 timeline is utility-scale solar plus battery storage — the technology the administration is actively suppressing, whose supply chain is subject to the FEOC thresholds, and whose construction timeline is the only one short enough to matter.
Gas turbines have a five-year lead time. GE Vernova’s backlog went from 46 gigawatts to 83 gigawatts in 2025 and is projected to hit 100+ gigawatts by end of 2026. Delivery slots are going out to 2030. Nuclear doesn’t exist at relevant scale before 2032 at the earliest. FERC’s “high probability” additions for natural gas through late 2028 total 10–22 gigawatts — against a demand curve growing at 25+ gigawatts per year.
The administration has, in effect, suppressed the only technology that could respond in time while betting on technologies that cannot respond at all by the relevant deadline.
What Would Change My Mind
I want to be explicit about what would invalidate this thesis, because a thesis that can’t be lost isn’t a thesis.
One: Treasury’s December 2026 FEOC safe harbor tables are permissive enough that post-2025-start projects can clear compliance at scale. The February 2026 interim guidance was received as “more workable than feared” — a meaningful data point in the optimistic direction. The formal tables, due by end of 2026, are the real test. If they provide workable certification pathways given the actual supply chain, the replacement pipeline could refill faster than the cliff thesis assumes. I’d be watching for developer announcements of post-2025-start projects moving to financial close after the tables are published.
Two: Gas permitting accelerates materially and 10+ gigawatts of new combined-cycle gas comes online in 2027. Current EIA projections put planned additions at 3–6 gigawatts. Meeting the 10-gigawatt threshold would require a permitting and construction acceleration with no recent precedent, against a turbine backlog that is already booking into 2030. This falsification condition is available in theory. I put low probability on it.
Three: AI efficiency gains reduce data center load forecasts by 30% or more before 2028, softening the demand side enough that the supply gap doesn’t produce visible reliability stress. DeepSeek’s architecture reportedly uses 10–40 times less energy than comparable US systems. If that efficiency curve compounds faster than AI adoption expands, total data center demand could grow more slowly than current projections. I consider this the most genuinely uncertain of the three conditions. The Jevons Paradox cuts against it — cheaper inference tends to expand the market rather than shrink consumption — but the efficiency gains are real and the trajectory is hard to model.
The scorecard test for this piece: if PJM’s 2028/2029 capacity auctions clear without reliability events, and clean capacity additions in 2028 come in above 60 gigawatts, the slope-not-cliff thesis was correct and this one was wrong.
The Second Edge
There is a version of this story where the 2028 gap forces exactly the policy reversal it takes to avert the worst of the 2029–2030 drop. Reliability events have a way of concentrating minds. A summer of rolling brownouts in mid-Atlantic data center corridors would create pressure — from exactly the industries the administration is trying to cultivate — that is harder to ignore than any climate argument.
That would be an ironic resolution. The anti-renewable rollback fails to stop renewables in 2025–2027 because the pipeline is already there. Then it produces a reliability crisis in 2028 because the replacement pipeline isn’t. The crisis generates political pressure for emergency permitting of the fastest-to-build technology available. Which is solar plus storage.
The transition resumes. Later, more expensively, with the manufacturing base partially ceded to China. But it resumes.
That is not a prediction. It is the least bad version of events that the current data supports. The worse version is that the gap persists long enough to constrain the AI buildout, slow data center construction, and push the investment that would have gone into US grid capacity into jurisdictions with cleaner policy signals.
The 81-to-48 gigawatt drop is not a forecast. It is already baked into the pipeline. What happens after it lands is still an open question.
But the cliff is not.
Working Hypothesis tracks its theses publicly. This piece’s call: discrete US grid capacity gap beginning 2028, visible in reliability planning data before then. Confidence: Medium-High. Falsification window: 24 months.
Sources
- Rhodium Group — Assessing the Impacts of the Final One Big Beautiful Bill (2025)
- Rhodium Group — Taking Stock 2025: US Energy and Emissions Outlook
- Lawrence Berkeley National Laboratory — Queued Up: 2025 Edition (interconnection queue data through end of 2024)
- IRS — Notice 2025-42 (Physical Work Test; eliminates 5% safe harbor for projects >1.5 MW, effective September 2, 2025)
- IRS — Notice 2026-15 (interim FEOC Material Assistance Cost Ratio guidance, February 2026)
- NERC — 2025 Long-Term Reliability Assessment
- PJM — 2027/2028 Base Residual Auction Results (December 17, 2025)
- Bipartisan Policy Center — Unpacking the FEOC Provisions in the One Big Beautiful Bill Act
- EIA 2026 capacity additions via PV Magazine — Solar and Storage to Lead Record-Breaking 86 GW of New US Capacity in 2026
- BloombergNEF post-2027 deployment cliff via PV Magazine — The Rise and Fall of Clean Energy Deployment Following OBBB
- LevelTen Energy / Latitude Media — Most solar and wind projects are safe harbored (76%/86% grandfathering figures; Latitude Media subscription may be required)
- Wood Mackenzie and BloombergNEF deployment data — paywalled; cited via secondary sources above