The Cliff That Moved: America Safe-Harbored 216-240 GW of Solar on Paper, and at Its Forecast Pace More Than a Third Strands

Yesterday, July 4, 2026, was the last day a new American solar project could lock in the federal government's 30 percent tax credit. The One Big Beautiful Bill Act, signed exactly one year earlier, ended the clean-electricity credits in Sections 45Y and 48ESections 45Y and 48EThe parts of the US tax code that house the current clean-electricity production and investment credits. of the tax code for solar and wind plants placed in serviceplaced in serviceThe moment a plant is finished and delivering power, the point at which the tax credit is actually claimed. after 2027, with one exception. Any project that 'began construction' within twelve months of enactment keeps its credit, provided it is finished within a set window. That twelve-month clock ran out at midnight.

The response was one of the largest paper stampedes in the history of American energy policy. Wood Mackenzie, the energy research firm that publishes the industry's benchmark market data with the Solar Energy Industries Association, estimates that developers safe-harbored between 216 and 240 gigawatts of solar capacity between mid-2024 and the deadline. A December 2025 survey by investment bank Roth Capital, covering roughly a quarter of the utility-scale market, extrapolates to about 250 gigawatts. Two independent rulers, one answer: the pile is enormous.

How enormous? The United States has installed 279 gigawatts of solar in its entire history through the end of 2025. The best year it has ever had, 2024, added just under 50. The stock of projects that now claim to be under construction equals about 80 percent of everything ever built, or more than four record years' worth of work, all of it, on paper, started in the last two years.

That claim is the subject of this article. The credit is not banked when construction begins; it is claimed when a plant is 'placed in service,' meaning finished and delivering power. And the safe harborsafe harborAn IRS rule that lets a project count as 'under construction' for tax purposes once the developer takes a defined minimum step, even if no ground has been broken. comes with a fuse: finish within four calendar years of the year you started, or lose the automatic protection. So the narrow question that matters is physical, not legal: can 216-240 gigawatts of solar actually get built by the time the windows close in 2028, 2029, and 2030? Cumulant's arithmetic says no, not close.

In everyday English, beginning construction means breaking ground. In tax law it means passing one of two tests. The first is the 5 percent safe harbor5 percent safe harborQualifying a project by spending at least 5 percent of its total cost on equipment before the deadline.: spend at least 5 percent of the project's total cost on equipment, panels, trackers, transformers, before the deadline. The second is the physical work testphysical work testThe alternative qualification route: starting real, significant construction work on-site or at a factory building custom components. It has no minimum spending requirement.: start 'physical work of a significant nature,' either on the site or at a factory making custom components for the project. The second test has no minimum dollar amount. Think of it as a restaurant reservation system where a deposit holds your table: the deposit is real money, but it is a small fraction of the meal, and you can still decide not to show up.

The past year turned those definitions into a legal battlefield. In August 2025, the IRS issued Notice 2025-42, which stripped the 5 percent spending route from any solar project larger than 1.5 megawatts beginning construction after September 2, 2025, leaving only the physical work test. Then, on June 6, 2026, the US District Court for the District of Columbia vacatedvacatedWhen a court strikes down a government rule, erasing it as if it had never been issued. the notice in Oregon Environmental Council v. IRS, ruling that the agency had failed to reasonably explain why it singled out wind and solar, the 'arbitrary and capriciousarbitrary and capriciousThe legal standard for striking down an agency rule that was not reasonably explained or justified.' standard. That restored the easier 5 percent route for the final month of the window. The government is expected to appeal, and the court itself noted that appellate review will likely outlast the deadline, so a slice of the pile sits on contested legal ground.

Whichever door a project came through, the same fuse is now lit. Under the IRS's continuity safe harborcontinuity safe harborThe IRS condition that a safe-harbored project finished within four calendar years of the year it started automatically keeps its credit; beyond that, the owner must prove continuous work under a case-by-case test., a project finished within four calendar years of the year it began construction keeps its credit automatically. Beyond that, the owner must prove 'continuous' progress under a facts-and-circumstances test that conservative tax-equity investors, the banks that actually monetize these credits, generally refuse to finance. In practice, four years is the window. Projects that started in 2024 must finish by the end of 2028, the 2025 cohort by end-2029, and the final rush by December 31, 2030. The cliff Congress scheduled for 2026 did not disappear. It moved.

Here is the whole analysis in one step. Take Wood Mackenzie's 216-240 gigawatt stock. Give it every benefit of the doubt: assume all of it has until the end of 2030, ignoring the earlier 2028 and 2029 cohortcohortAll the projects that began construction in the same calendar year, which therefore share the same completion deadline. deadlines. That leaves the four years 2027 through 2030 to build it, which requires 54 to 60 gigawatts of new solar every year, four years running.

Now compare that to what America has ever done. The all-time record, set in 2024, is just under 50 gigawatts, and that counts every rooftop panel in the country. The utility-scale record, the segment where nearly all of this pile lives, is 41.4 gigawatts. Actual installations in 2025 fell back to 43 gigawatts across all segments. And the current run rate is going the wrong way: the first quarter of 2026 added 7.8 gigawatts, down 27 percent from a year earlier, an annualizedannualizedA short period's number scaled up to a full-year rate, one quarter's installations multiplied by four. pace of about 31 gigawatts. Wood Mackenzie's own base-case forecast for 2026-2031 averages roughly 43 gigawatts a year across all segments, about 35 of it utility-scale.

Could the industry simply accelerate? The binding constraints are not ambition. High-voltage transformers carry lead times of two to four years, interconnection queues, the waiting lists to plug into the grid, routinely run longer than the credit windows themselves, and the labor force and tax-equity market were both sized for a 40-gigawatt industry, not a 60-gigawatt one. Those are precisely the frictions Wood Mackenzie cites in explaining why its forecast stays flat despite what it calls a robust safe-harbored pipeline. The gap between 43 and 57 is not a rounding error. It is the difference between a backlog and a bet.

Why would developers file paperwork for far more than they can build? Because it was cheap, and the alternative was irreversible. A safe harbor is an optionoptionA small payment made now to preserve the right, but not the obligation, to do something bigger later, the economic structure of a safe-harbor purchase.: a small payment today that preserves the right, but not the obligation, to claim a 30 percent subsidy later. Missing the deadline forecloses the subsidy forever; over-filing costs only the deposit. Any rational developer facing that asymmetry files for everything in the drawer, the speculative site, the project stuck in the queue, the one whose power contract fell through.

The procurement data shows exactly this shape. Camelot Energy Group, a consultancy that tracks safe-harbor equipment purchases, estimates developers locked up more than $3.8 billion of hardware between July 4, 2025 and the deadline, against up to $200 billion of intended project investment. That is roughly $53 of stated intent riding on each dollar actually spent. The 5 percent spending test caps that leverage at 20 to 1 by definition, so a ratio near 53 to 1 tells you the pile leans heavily on the physical work test, the route with no spending floor at all.

The pile's own measurer does not believe it: the firm that counted 216-240 gigawatts of safe-harbored solar still forecasts about 43 a year.

That is the strongest single tell in the data. Wood Mackenzie sized the stockpile and, in the same breath, kept its base-case forecast at roughly 43 gigawatts a year, a number that implies the firm expects at least a third of its own measured pile to expire unused. When the entity with the best view of the order book prices in massive attrition, the order book is an options ledger, not a construction schedule.

Run the scenarios against the 228 gigawatt midpoint of Wood Mackenzie's range, with every window generously held open through 2030.

At the utility-scale forecast pace of about 35 gigawatts a year, 87 gigawatts, 38 percent of the pile, miss their windows. Across the full 216-240 range, that is 35 to 41 percent. Even if the industry sustained its all-time utility-scale record for four straight years, something it has never done for two, 27 percent strands. The only row that gets stranding down near 12 percent requires holding the all-segments record of 50 gigawatts permanently, and that row cheats, because it counts rooftop installations against a pile that is overwhelmingly utility-scale. Under any assumption the industry itself considers realistic, more than a third of the paper never becomes steel in time.

Stranding is not cancellation. A project that misses its window loses the 30 percent credit, not the right to exist; some will be rebid as unsubsidized projects, and surging data-center demand for power will rescue a share of them at higher contract prices. But a 30 percent cost swing is usually the difference between a power contract that clears and one that does not, which is why the credit existed in the first place. For electricity buyers, the practical translation is that the subsidy embedded in contract prices starts draining out of new deals years before anyone announces a cliff.

There is a precedent for what a federal tax-credit deadline does to a build schedule, and it is not subtle. In 2012, the wind industry faced the scheduled expiry of its production tax credit at year-end. Developers raced a record 13,131 megawatts into service. The following year, installations collapsed 92 percent, to 1,087 megawatts, the deadline had not created demand, it had vacuumed the future into the present.

The analogy needs handling with care, in both directions. The 2012 deadline was a placed-in-service deadline with no multi-year grace period, so the air pocket arrived immediately. The 2026 deadline is a begin-construction deadline with four-year windows, so the pull-forwardpull-forwardWhen a deadline causes activity that would have happened later to happen earlier, borrowing from the future rather than creating new demand. plays out in slow motion: the boom is spread over 2027-2030 as safe-harbored projects race their fuses, and the analogous collapse arrives around 2029-2031, when the last windows shut and the unsubsidized era begins. The wind chart is not a forecast of next year. It is a forecast of what the far side of 2030 looks like if nothing replaces the credit, and a reminder that deadline-driven booms measure urgency, not demand.

Everything above describes yesterday's deadline as a market reaction: a legal and financial event, measured in filings and hardware deposits. The economic effect, electrons, jobs, steel in the ground, resolves over the next eighteen months, and there is a clean public instrument for watching it. EIA Form 860MEIA Form 860MA monthly federal survey where power-plant owners report the capacity and expected start dates of projects they are actually planning to build. is the federal government's monthly census of planned power plants; developers report projects there when they are actually being built, with interconnection agreements and delivery dates, because there is no tax advantage to padding it.

Our read of the evidence is the second branch. The required pace has never been achieved, the current pace is falling, the leverage ratio says the pile was assembled as options rather than commitments, and the firm that measured it forecasts as if a third of it will lapse. Congress and the industry spent a year fighting over where the door would close, and in the end 216-240 gigawatts squeezed through it on paper. The arithmetic says the fight was less decisive than it looked. The cliff was not cancelled on July 4, 2026. It was rescheduled, to the quiet expiry of credit windows in 2028, 2029, and 2030, one missed deadline at a time.