Inverter efficiency: why your home battery misses its 96% at night

The datasheet of a home battery often gives an efficiency of about 96%. At night it does not reach that. When discharging, the Conow Atlas 6000 AC reaches 94.8% ± 7.6 at 500 W, and 78.8% ± 3.2 at 100 W. own measurementour techbench, the device's energy counter; ladder-v1 and ladder-low-v2, 19 to 21 September, with an error bar per step · 3000 W setting · 21 September 2026 It is the same device, and nothing is broken.

The difference is in one item: what the inverter costs to be on. Draw the loss in watts instead of the efficiency in percent, and you see that item at once. This article belongs to the video about inverter efficiency on Jay's Desk. The video gives the numbers. Here you also find the formulas, the measurements per step, the definitions behind the count on the CEC list, a worked example and the sources. The video is in Dutch.

Key takeaways

• Efficiency is a curve, not a number. When discharging, the Conow reaches 94.8% ± 7.6 at 500 W and 78.8% ± 3.2 at 100 W. Below about 300 W its efficiency drops steeply. own measurementour techbench, the device's energy counter; ladder-v1 and ladder-low-v2, 19 to 21 September, with an error bar per step · 3000 W setting · 21 September 2026
• At a small load you mostly pay the fixed loss. At 97 W out, the Conow loses 26 W ± 5. A fit through all the steps gives a fixed loss of about 25 W, in both directions. derivedfrom the Conow's efficiency per step, with a weighted fit loss = p0 + p1·P + p2·P²
• The peak sits in the middle, not at full load. In the model loss = p0 + p1·P + p2·P², the peak sits where the fixed loss p0 equals the quadratic part p2·P². On the CEC list, the median peak sits at 51.7% of rated power. Of the 2,343 solar inverters, 3.9% have their peak at full load. derivedfrom the Sandia parameters of 2,343 PV inverters in the CEC list of NREL SAM, downloaded on 9 November 2026
• At night a home battery sits at the bottom of the curve. In the measurements at 28 single-family homes, on average 72% of the power flows at night are below 300 W. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026, with the ISFH load profiles from the Stromspeicher-Inspektion 2024 · 10 November 2026
• The design decides more than the size. Four 10 kW hybrid inverters lose between 17 and 84 W at 150 W out, in the same class and for the same work. third partyHTW Berlin, in pv magazine, June 2024 · 10 November 2026
• The 96% on the datasheet is the top of the curve. According to HTW Berlin, the efficiency on a datasheet usually belongs to the most efficient path: from the panels to the grid, without the battery. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026 · 10 November 2026

How it was measured and calculated

Three plug-in batteries supply numbers for this piece. They are examples of a property, not a choice. This article is not a review and not a ranking.

• The Conow Atlas 6000 AC. The efficiency per step, charging and discharging, from 100 to 3000 W. Measured on our techbench from 19 to 21 September 2026, in the 3000 W setting. The efficiency comes from the device's energy counter (steps of 10 Wh) against the socket, 300 seconds per step. Every step carries its error bar. These are the numbers from the Conow review.
• A second home battery. The idle draw from the grid: 31.9 W, measured with an MID-certified kWh meter over a little more than 12 hours, on 22 October 2026. And the manufacturer's own statement: 30 to 40 W fixed loss. This unit is no longer with us.
• A third home battery, of a little over 2 kW. Only the shape of its curve. The DC side comes from the battery voltage times the battery current that the unit reports itself, against the measuring plug, one reading per second. The ladder ran on 7 November 2026; the point for charging at full load comes from a different run on 6 November.

The second and the third appear here without brand and without type number. Their numbers serve as an example of a property, not as a verdict on a product.

Two limits belong with this:

• A scale error in the counter. The Conow's energy counter sits on the battery side. A fixed scale error k then makes charging higher and discharging lower, by the same factor. Because both directions come close to 1 at 1500 W, "efficiency ≤ 1" limits that error to 0.980 ≤ k ≤ 1.025, on the central values. The loss at 97 W out then becomes 23 to 29 W. That falls within the 26 W ± 5. derivedfrom the central values of charging and discharging at 1500 W (0.976 and 0.980) and the 100 W step Note: the error bars at 1500 W are ± 8 percentage points, so at the edges that limit is looser.
• Why the third only supplies a shape. For the third, a scale error in the current reading can lie between 0.955 and 1.031. Such an error shifts every efficiency as a standalone number, and above all the loss in watts at the top. The order of the steps stays, and so does the efficiency as a share of its own peak. That is why this article gives no efficiency and no loss as a number for the third, only the shape. One exception: at about 100 W out, something in the order of 20 W goes to heat, as a rough order, without an error bar.

We did the calculations on the CEC list on 7 November 2026 and repeated them on 9 November, on the list as NREL published it then. They are solar inverters, not battery inverters.

One device, two numbers

What does a battery reach when it covers the house load at night? For the Conow, every step was measured, in both directions.

The error bar is in percentage points. Read the table like this:

• Below about 300 W the efficiency drops steeply. That statement holds within the error bars.
• Between 300 and 1000 W no order can be read. The error bars overlap there. So the 500 W step is not a peak.
• The 1500 W step comes out high within its error bar. It is not a peak.
• Charging at 200 W has no number. On that step, the counter did not get enough counts.

At 0 W out, the efficiency is 0%

Efficiency is what comes out, divided by what goes in. If nothing comes out, the efficiency is exactly 0%. So every efficiency curve starts at zero.

An inverter also costs something just to be on. HTW Berlin names control and communication as fixed losses. You pay for them even when nothing comes out. In this article that part is called the fixed loss. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026, p. 18–19 · 10 November 2026

How heavy is that fixed loss? HTW gives an example from KIT's lab data. A 10 kW hybrid inverter in the test, system B4, has a no-load loss of 50 W. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026, p. 18–19, lab data KIT · 10 November 2026

• At 100 W out it draws 150 W on the battery side. The efficiency is 66%.
• At 500 W out the efficiency is 90%.
• At 10 kW out the efficiency is 97.8%, with a loss of 228 W.

The manufacturer of the second home battery names its own fixed loss: 30 to 40 W. manufacturerstatement by the manufacturer of the second home battery, in writing · 23 October 2026 Idle, that battery drew 31.9 W from the grid. own measurementMID-certified kWh meter, averaged over a little more than 12 hours · 22 October 2026 That falls within the statement.

From 66% to 90%, it looks as if an inverter gets better the harder it works. That holds up to a point. Beyond it, working harder costs more than it yields.

Three buckets of loss

HTW Berlin splits the loss of an inverter into three parts. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026, p. 18–19 · 10 November 2026

1. Fixed: control and communication. You pay the fixed loss even at 0 W out.
2. Proportional: the switching losses. They grow in step with the power.
3. Quadratic: the ohmic losses in coils, cables and connectors, and the conduction losses in the semiconductors. They grow with the square of the current.

The quadratic part follows from P = I²R. Send twice the current through the same resistance, and you get four times the heat. You get twice the output, and you pay four times as much of that one part. The fixed and the proportional parts do not grow four times. Warm copper also conducts worse: around room temperature, the resistance of copper rises by 0.39% per degree.

Low voltage costs too. At the same power, half the voltage gives twice the current, and so four times the ohmic loss. A plug-in battery sits low: sixteen LFP cells in series, 51.2 V nominal and 58.4 V full. Below 60 V DC, the touch-safe class ES1 from IEC 62368-1 applies, and that is why you can install it yourself. How that works is in How does a home battery work?

The loss model, and where the peak sits

Put the three buckets in one formula, with P the power that comes out:

loss       = p0 + p1·P + p2·P²
efficiency = P / (P + loss)

• p0 is the fixed loss, in watts.
• p1·P is the proportional part.
• p2·P² is the quadratic part.

Efficiency is highest where the loss per watt of output is lowest. The loss per watt of output is:

loss / P = p0/P + p1 + p2·P

The derivative with respect to P is −p0/P² + p2. It is zero where p0 = p2·P², so at P = √(p0/p2). The peak sits where the fixed loss equals the quadratic part.

Two consequences:

• The proportional part p1 does not move the peak. In the formula, p1 stands alone, without P. It pulls the whole curve down, by the same amount at every power.
• Below the peak the fixed loss wins, above it the quadratic part. The harder the inverter works, the less the fixed loss weighs. The quadratic part catches up somewhere.

Does the model match the measurement? A weighted fit through the Conow's steps gives p0 = 25.1 W discharging and 26.0 W charging. At 97 W out, the model then gives a loss of about 26.4 W and an efficiency of 78.7%. The measurement is 78.8% ± 3.2. derivedfrom a weighted fit (1/σ²) of loss = p0 + p1·P + p2·P² through the Conow's steps; discharging p0 = 25.1 W, charging p0 = 26.0 W

Where the Conow's peak sits, I do not say. Its upper end is too loose for that: at 3000 W the loss is 249 W ± 123. A band like that proves no peak position.

Flip the graph

Put the loss in watts on the axis, instead of the efficiency. For the Conow, discharging, it looks like this:

At the bottom the loss barely moves, around 25 W. So the bad curve at the bottom is a fixed loss, divided by a small output. At the top, the bands are too wide to prove a shape. The top needs a different measurement.

The third home battery supplies that top, as a shape. Its efficiency rises steeply, flattens in the middle and drops again at full load. own measurementbattery voltage × current from the unit, against the measuring plug, one reading per second; shape only · 7 November 2026

• The peak is a region, not a point. It lies between the 750 and 1500 W steps, about 31 to 62% of its full load. The steps in that region overlap each other. The peak position from a fit depends on the scale error k, so I do not give it.
• At full load it reaches less than at 750 and at 1500 W. When discharging, those bands do not overlap. Against 1000 W they do overlap, so I do not compare it with that step.
• For charging, one point at full load closes the curve. It comes from a different run, with a cool battery. It does not prove that charging drops at full load: the point overlaps with the best charging steps.

So the sentence "an inverter is most efficient near full load" is wrong. What does hold: the harder an inverter works, the more it loses in watts. That sentence is about the loss in watts, not about the efficiency.

Where is the peak for other inverters?

For a wider picture, we calculated the peak for every inverter on the CEC list. That list is in NREL SAM, with the parameters of the Sandia inverter model for each inverter. They are 2,343 solar inverters, not battery inverters. derivedfrom the Sandia parameters of 2,343 PV inverters in the CEC list of NREL SAM, downloaded on 9 November 2026

The definition. The Sandia model gives the power on the AC side as a function of the power on the DC side, at the nominal DC voltage. For each inverter we find the power at which the efficiency is highest. The peak position is the power that then comes out, as a share of the rated AC power (Paco). If the curve has no peak within the range, the peak sits at full load.

Between 91 and 89 sit two inverters with a peak just below full load. The median is the middle value of the row: half of the inverters have their peak lower, half higher.

The peak of the third home battery, about 31 to 62% of its full load, overlaps with the middle half on the CEC list.

Where a home battery sits at night

At night, home batteries mostly discharge between 100 and 300 W, HTW Berlin writes. In the measurements at 28 single-family homes, on average 72% of the power flows at night are below 300 W. The load profiles come from ISFH, the Institut für Solarenergieforschung in Hameln. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026, with the analysis from the Stromspeicher-Inspektion 2024 · 10 November 2026 These are averages. Your house can differ.

Lay that band over the Conow's table, and you sit exactly in the part where its efficiency drops steeply: 78.8% ± 3.2 at 100 W, 85.2% ± 6.8 at 200 W, 92.1% ± 7.7 at 300 W.

In 2024, HTW measured four 10 kW hybrid inverters. The worst of the four only reaches an efficiency of 90% when discharging from 750 W up. In the night measurements, the power is lower 96% of the time. third partyHTW Berlin, in pv magazine, June 2024 · 10 November 2026 Also a 96, but of the time.

What it costs

HTW Berlin works out an example. If you buy in the cheapest hours for 25 cents per kWh and use it later at 35 cents, your system must reach at least 71% round trip: 25 divided by 35. Whoever forgets the low efficiency at a small load overestimates what he saves. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026 · 10 November 2026 That sum holds only the price per kWh. Wear and fixed costs are not in it.

A worked example (an example, not a measurement). What does the fixed loss cost per day? A simple estimate:

fixed loss per day ≈ p0 × hours discharging
                   + p0 × hours charging
                   + idle draw × the remaining hours

With these assumptions:

• A fixed loss of 25 W discharging and 26 W charging, the order of the Conow.
• 8 hours a day discharging, 3 hours charging, and 13 hours idle.
• 5 W idle. That number is an assumption, not a measurement of a device. Put in your own number.
• A price of €0.25 per kWh: the purchase price from HTW's example, not a supplier's tariff.

Per year that comes to about 125 kWh. At €0.25 per kWh it costs about €0.09 per day, or €31 per year. derivedfrom the Conow's fixed loss (25.1 W discharging, 26.0 W charging), rounded, with the assumptions above

Three caveats:

• The real loss is larger. The sum counts only the fixed part. The proportional and the quadratic parts come on top, above all when charging at high power.
• The idle draw differs per device. The second home battery drew 31.9 W from the grid when idle, more than six times the assumption above.
• The hours are an assumption. How long your battery charges and discharges depends on your consumption, your panels and your contract.

Bigger costs something, but the design decides more

Does a bigger inverter always cost more at a small load? On the CEC list, the fixed loss in watts grows with the class. derivedfrom the median of Pso per class of rated AC power, CEC list of NREL SAM, downloaded on 9 November 2026

The fixed loss here is the parameter Pso from the Sandia model: the DC-side power the inverter uses before anything comes out. These are solar inverters as well.

So bigger costs something. Within one class, the difference is larger. In 2024, HTW Berlin measured four 10 kW hybrid inverters, all the same class. third partyHTW Berlin, in pv magazine, June 2024 · 10 November 2026

The difference of 17 against 84 W within one size is larger than the step between the median of 1 to 5 kW and that of 5 to 20 kW on the CEC list (21.5 against 38.4 W). Note: the two sources measure different inverters. HTW measured hybrid inverters with a battery; the CEC list holds solar inverters.

The fixed loss as a share of rated power. The percentage below is a loss share, not an efficiency:

Two caveats. The 17 W of the best hybrid is its whole loss at 150 W out, so its fixed loss is at most that. And for the second home battery it is a manufacturer's statement, not a fit: we did not measure that unit in steps at the low end. So two of our plug-in batteries sit at 0.8 to 1.1%, but only for the Conow does that come from our own measurement.

What designers do about it

How large the fixed loss is, the designer decides. The sources name four techniques.

• A minimum output power. In the 2026 test, two systems deliver nothing below 25 W and 50 W when discharging. HTW measures their curve only from there. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026 · 10 November 2026 What your house asks below that comes from the grid. That follows from the connection; HTW does not write it.
• Multilevel, without a separate inverter. One system in the 2026 test builds the 230 V sine from stacked cell voltages. It has the highest system score (SPI) in the class up to 5 kW: 93.2%. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026 · 10 November 2026
• Switching bridges off and a burst mode. In Fraunhofer ISE's HyBaG project, the inverter switches bridges on or off per load range: one third, two thirds or full load. At very small powers it runs a burst mode on one phase, in which it is active only 10% of the time. third partyFraunhofer ISE, press release on the HyBaG project · 23 July 2020
• Silicon carbide. In the same project, SiC transistors lower the drive losses and the no-load losses. third partyFraunhofer ISE, press release on the HyBaG project · 23 July 2020

What you can do yourself

A little.

• Set a threshold, if your system allows it. Below that threshold the inverter switches off. The SigenStor from Sigenergy, for example, has such an adjustable threshold.
• Do not spread a small load over several batteries. Each unit pays its own fixed loss. Calculated with the Conow's discharging steps: 300 W on one unit reaches 92.1% ± 7.7. The same 300 W over three units of 100 W each reaches 78.8% ± 3.2. The error bars do not overlap. derivedfrom the Conow's measured 300 W and 100 W discharging steps, for three identical units Calculated for three identical units, not measured with three real units.
• Charge in blocks during the cheapest hours, instead of trickling. That follows from the model: at a higher power, the fixed loss weighs less. It is a consequence of the model, not measured.

What the 96% on the datasheet is

The efficiency on a datasheet is one point on the curve: the peak. According to HTW Berlin, that number usually belongs to the most efficient path, from the panels to the grid, without the battery in between. How large the loss is when discharging is often not stated. According to HTW, a reliable comparison based on datasheets is not possible. HTW writes that about the systems in its test. third partyHTW Berlin and aquu, Stromspeicher-Inspektion 2026 · 10 November 2026

Does it say 96% for the whole round trip, charging and discharging together? A round trip is the product of two sides. Two sides of 96% each give about 92% together, before any loss in the battery itself. derivedfrom 0.96 × 0.96 = 0.92 So a 96% round trip is almost always the efficiency of one side of the inverter.

The European efficiency is a weighted average of the efficiency at six load points. The weights are in the standard IEC 62891:2020, Annex D, Table D.1 (and in EN 50530). We have only seen the standard's table of contents; the values come from Penn State's course material (EME 812, §6.5, equation 11.4).

The percentages in the bottom row are weights, not efficiencies. That weighting describes a solar inverter on a sunny day. If a 3,000 W inverter delivers 100 to 300 W at night, it sits at 3 to 10% of its power. That range gets 9% weight in total. derivedfrom 100 and 300 W divided by 3000 W; the weights at 5 and 10% load added up

The box of a home battery gives the inverter's peak power, in watts. The datasheet gives the top of the efficiency curve. Both describe the best moment of the device. At night the battery sits at the bottom of the curve, where the fixed loss weighs most.

What this article does not say

• No verdict on a product. The Conow, the second and the third are examples of a property. There is no best among them, and no order between the three.
• No brand names for the HTW systems. Those are third-party measurements in a lab test.
• Not "bigger is always worse at a small load". The fixed loss grows with the class on average, but within one class the design decides more.
• No measurement with three real units. The example of one against three units is calculated.

Frequently asked questions

Related articles

• The Conow Atlas 6000 AC review for every step, the closed cycle and the measurement setup
• How does a home battery work? for the route from the roof to the kettle, and why a plug-in battery sits at 48 V
• Your home battery is too big for the size of a battery against your consumption

Transparency

No brand paid for this video or this article, and no brand saw them in advance. I got the Conow free of charge from Conow for the review, and I may keep it. No money was paid. This article contains no affiliate links and no discount code.

The article explains a property of inverters and reviews no product. The second and the third home battery appear without brand and type number.

Sources