Report 156 · Energy Storage
The pitch for DC-coupled solar-plus-storage usually starts with a picture: a solar curve with its peak sliced flat by the inverter, and a battery catching the slice. The picture is real. The slice is small. At the DC-to-AC ratio most US plants are actually built with, the national lab that modelled it puts the recoverable energy at about two tenths of one percent of what the array produces.
Start with why clipping happens at all. A solar array produces direct current. The grid wants alternating current, so an inverter sits between them, and the inverter has a maximum output. Developers deliberately install more panel capacity than inverter capacity, because panels got cheap faster than inverters did and an oversized array keeps the inverter near full output for more of the day. The ratio between the two is the inverter loading ratio, or ILR, sometimes written DC/AC.
The consequence is that on the sunniest hours the array could produce more than the inverter can pass. The excess is clipped: the inverter simply does not take it. In a plant where the battery has its own inverter on the AC side (AC-coupled), that energy is gone. In a plant where the battery sits on the DC side and shares the solar inverter (DC-coupled), the battery can soak up the excess before the inverter ever sees it.
That is the whole mechanism, and it is correct. The question a buyer should ask is how much energy we are talking about.
The most useful public number comes from a 2021 technical report by the National Renewable Energy Laboratory (since renamed the National Laboratory of the Rockies), written to add DC-coupled PV-plus-battery plants to the lab's national capacity-expansion model, ReEDS. To do that, the team had to estimate how much clipped energy a DC-coupled battery could actually recover, and they did it the expensive way: hourly simulations of a single-axis tracking plant with an ILR of 1.3, run across roughly 55,000 grid cells covering the developable land of the contiguous United States, on seven years of weather data from 2007 to 2013.
The result, from the report's appendix:
The maximum recoverable clipped energy for ILR=1.3 is ~2.1% of total DC energy, observed at a small selection of sites in the southwest. The average recoverable clipped energy is 0.2%.
That average became the model input. In the body of the report the authors describe raising the plant's capacity factors by 0.2 percent in daylight hours to represent it, and a footnote on their results is even more direct: the need to recover clipped energy is, in their words, "modest throughout most of the conterminous United States with an ILR of 1.3."
To put 0.2 percent in terms a site owner can feel, here is my own arithmetic, not the report's: two tenths of a percent of a year's output is about 0.73 days of average production. The best southwestern sites, at 2.1 percent, get roughly a week and a half. That is real energy, but it is not the headline benefit of a system that costs as much as a battery does.
A skeptic would say the number is small only because 1.3 is a modest ratio. That is fair, and the NREL authors say the same thing: larger amounts of clipped energy would be available at higher ILRs, and the report notes that DC-coupled plants are expected to eventually be built with relatively large ILRs for exactly that reason.
But notice what that sentence means. Clipping at a high ILR is not free energy you were losing anyway. It is energy from panels you chose to add in order to create the clipping. The recapture is then a return on the extra panels plus the DC-coupled hardware, and it has to be judged as that investment, not as a bonus attached to a plant you would have built regardless.
And 1.3 is where the industry actually sits. The NREL report cites a 2018 median ILR slightly above 1.3 for both tracking and fixed-tilt projects. Lawrence Berkeley National Laboratory's 2025 update of its utility-scale solar dataset, covering 1,726 projects, reports that the average ILR has held steady since 2017. Berkeley Lab also measured what the oversizing buys in practice: across its sample, the highest-ILR quartile of projects averaged about one percentage point higher capacity factor than the lowest.
If clipping recapture were a decisive advantage, you would expect new hybrid plants to be DC-coupled. They mostly are not. The same Berkeley Lab update states that 80 percent of new PV-plus-battery projects used AC-coupling in 2024, and puts it bluntly: despite the theoretical cost savings of DC-coupling, most recent projects use AC-coupling, with Florida the exception.
There are good reasons for that, and the NREL report lays several out. AC-coupled plants are less complex to deploy and can deliver more capacity value, because the solar and the battery each have their own inverter and can push power to the grid at the same time. In a DC-coupled plant they share one inverter, which caps the combined output. The NREL authors flag that their own model probably overstates the DC-coupled battery's arbitrage value for precisely this reason: it does not account for moments when prices are high but the shared inverter is already full.
None of this makes DC-coupling a bad design. It makes clipping recapture a weak reason to choose it. The NREL report's sensitivity runs show where the case actually rests. Deployment of DC-coupled hybrids in the model scaled with how much capital cost the shared hardware saves (the default assumption was 5 percent cheaper than separate solar and battery plants), and it was highly sensitive to whether the battery could earn money by charging from the grid and discharging later.
A 2025 NREL report on battery dispatch software for a DC-coupled plant, written up from a research agreement with Southern Company, gives a small illustration. Its controller is told to always charge from clipping when clipping occurs and to reserve room for it. In the example days the report walks through, the battery still:
charges minimally from the PV system during peak operation to eliminate inverter clipping. The battery primarily charges from the grid in the middle of the night when electricity prices are low
That is a single hypothetical example, not a survey of plants, and I would not lean on it for more than it shows. But it shows the shape: clipping is the first rule in the controller and a minor share of the battery's actual work.
I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link. Full policy here.
The reason this topic matters to me is that the decision in the dispatch example above (charge from clipping, from the panels, or from the grid, and when) is exactly the kind of decision cycling software makes all day. When a system is sold on one benefit that turns out to be a fraction of a percent, the owner ends up judging the battery against the wrong yardstick. The right yardstick is the one the NREL model landed on: what the battery earns and protects when it can charge and discharge on its own schedule. Nothing in this report is about that company's products or any other specific product.
These are modelled numbers, not metered ones. The 0.2 percent average and 2.1 percent maximum come from simulation using NREL's System Advisor Model, not from measured plant data. Berkeley Lab's 2025 update adds a relevant caution in a different context: it suggests that newer plants' underperformance against models may partly reflect higher DC/AC ratios, because sub-hourly clipping is not captured well by hourly modelling. My reading of why: brief spikes above the inverter limit can average out inside an hourly timestep, so an hourly simulation may understate clipped energy somewhat. I found no public dataset quantifying by how much.
A small internal inconsistency in the NREL report. The Figure 5 caption says clipping losses at ILR 1.3 are always less than 3 percent of DC energy and describes first-year values, while Appendix B describes the same analysis as averaged over 2007 to 2013 with a maximum of about 2.1 percent. The two statements are compatible (2.1 is less than 3), and I have used the appendix figures throughout.
This is utility-scale evidence. The NREL model and the Berkeley Lab dataset cover ground-mounted plants. Home hybrid inverters are also sold on clipping capture, and the same physics applies, but I did not find a comparable national dataset for residential DC/AC ratios or clipping, so I am not putting a number on homes. The practical takeaway still transfers: ask what the DC/AC ratio is before valuing clipping capture at all.
Age of the core source. The NREL modelling report is from April 2021, and its median-ILR figure is for 2018. The Berkeley Lab update (October 2025) is the reason I am comfortable that the ratio has not shifted much since.
Clipping recapture is real, correctly explained in most marketing, and small. At the DC/AC ratios US plants are actually built with, NREL's national model puts the recoverable energy at about 0.2 percent of the array's output on average, and a little over 2 percent at the best desert sites. Bigger numbers require deliberately oversizing the array, which turns the recapture into a return on panels you added, not a free gift.
If you are evaluating a DC-coupled system, ask for three things: the DC/AC ratio, the modelled clipped energy at that ratio for your site, and how the battery earns its keep the other 99.8 percent of the time. The last answer is the one that decides whether the system is worth buying.
Scope note: this report explains how much energy DC-coupled storage can recover from inverter clipping, using published national-laboratory modelling and data. The 0.73-day and week-and-a-half conversions are the author's arithmetic from the NREL percentages, not figures from the sources. No plant was measured and no product is evaluated or recommended. Disclosure: the author helps design AI battery-cycling systems for a veteran-owned energy-storage integrator linked in the body; he does not own it and receives nothing for the link.
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.