Home Battery Emissions: Is Your Powerwall Fighting Peakers or Storing Coal?

There's a specific moment I look forward to on summer evenings. It's 4:02 PM, the Powerwall flips from idle to discharging at about 4 kW, and the house stops pulling anything from the grid for the next three hours.

I like it because of money. SRP's E-27 summer on-peak window runs 2 to 8 PM on weekdays, but the single 30-minute average that sets my demand charge almost always lands between 4 and 7, so that's the stretch I bother to defend. The bill is set by the worst half-hour of the month, nothing else. Every kW the battery shaves off that one interval, above the 10 kW tier, is $36.13 off the July or August bill under SRP's E-27 ratebook effective November 2025. That's the reason the thing is bolted to my garage wall.

The same moment does something else: it takes three hours of my house's load off a gas peaker. Most of what gets written about home battery emissions is either vague ("go green!") or wrong in a specific way that's easy to correct.

Peaker plants are the dirtiest thing on the grid

When your house pulls 8 kW at 5:30 PM in August, that power doesn't come from the grid's average generator. It comes from the last generator dispatched, and at 5:30 PM in August that's usually a simple-cycle gas turbine.

These are peaker plants. They exist to run a few hundred hours a year, they're built for fast starts rather than efficiency, and the difference shows up in the heat rate. Per EIA's 2024 figures, a combined-cycle gas plant burns about 7,550 Btu to make a kWh. A simple-cycle gas turbine burns about 11,000. Same fuel, same combustion chemistry, roughly 45% more CO2 per kWh delivered, because the extra fuel goes out the stack as heat instead of into a second turbine.

Peakers also tend to be old, and they're sited in load pockets close to the people they serve. PSE Healthy Energy counted 17 gas and oil peakers in Arizona, 12 of them more than 45 years old, with six units sitting inside federal ozone non-attainment areas around Phoenix and Yuma. NOx is what makes ozone. Carbon is only part of what comes out of the stack.

This is why marginal emissions and average emissions are different numbers, and why the average one is misleading. The US grid averaged about 0.35 to 0.37 kg CO2 per kWh in 2023, depending on whether you take EPA's eGRID or EIA's accounting. Arizona's gas peakers, in that same PSE dataset, report 0.54 to 0.73 kg per kWh. Anywhere gas sets the margin, which is the Desert Southwest and California on a hot evening, the power your battery displaces at 6 PM is half again to nearly double the grid average. (The direction flips in coal-heavy regions, and I'll get to that.)

Here's the part that makes this easy: utilities already priced it in. The 4–7 PM window on APS, the 2–8 PM window on SRP, the 4–9 PM window on PG&E and SCE. Those windows exist because that's when the grid is most expensive to serve. Expensive to serve and dirty to serve are the same hours, for the same reason. (If you're in the Phoenix metro, APS and SRP are the whole conversation.) Nobody has to choose between shaving demand charges and cutting emissions. The optimal schedule for your bill is roughly the optimal schedule for the grid.

Curtailment is the other half of the story

The evening ramp gets the attention. The midday problem is bigger.

CAISO's net load curve drops through the floor around 11 AM as solar comes on, then climbs steeply when the sun goes down. When CAISO first published the duck curve a decade ago, it projected a 13,000 MW three-hour ramp. The record now stands above 23,000 MW, set in January 2025. Grid operators have to hold conventional generation online through the midday trough just so it's available for the ramp, which means solar gets curtailed. California threw away 3.4 million MWh of solar and wind in 2024, up 29% from the year before, and 93% of it was solar. Not because it wasn't generating. Because there was nowhere to put it.

Every kWh of storage that charges during the trough is a kWh of renewable generation that doesn't get curtailed and doesn't get replaced by gas four hours later. This is the same physics whether the storage is a 400 MWh utility installation in Kern County or your 13.5 kWh Powerwall. Storage lets the generators you already built run at full output. That's why solar plus storage is what keeps showing up as the working mechanism of grid decarbonization: IEEFA's projection has wind, solar and hydro covering more than all of US demand growth through 2027, with coal and gas losing share even where their output doesn't fall.

Your own system does a small version of this every day. If your Powerwall charges from midday solar that would otherwise have been curtailed upstream or exported at the three or four cents a kWh California's net billing tariff pays during the spring surplus, and then discharges at 6 PM against a gas peaker, the round trip is close to free from an emissions standpoint. Powerwall 3 is DC-coupled, so Tesla rates the solar-to-home path at 97.5%. Charge it from the grid instead and you're on the 89% AC round-trip figure. Either way, you spend a little energy to make the rest of it land in a much dirtier hour.

What the battery cost to build

Honest accounting means counting the manufacturing.

Lithium iron phosphate cells, which is what's in the Powerwall 3, run roughly 60 to 110 kg CO2e per kWh of capacity to produce, depending mostly on how clean the grid was where the cells were made. That range is IVL's 2019 estimate for lithium-ion generally; a 2024 review in Nature Communications put LFP specifically at 54 to 69. Call it 1,000 kg CO2e for a 13.5 kWh unit, plus the enclosure, the inverter, the shipping, and the installer's truck. A tonne, give or take, before it moves a single electron.

Now run the payback. A Powerwall doing one meaningful cycle a day moves something like 3,600 kWh a year. If each of those kWh displaces peaker generation at 0.6 kg CO2/kWh (the arithmetic on an 11,000 Btu/kWh turbine burning natural gas at 53 kg per million Btu) and gets charged from surplus solar, you're avoiding close to 2 tonnes a year. The manufacturing debt clears in well under a year. The warranty runs ten, with unlimited cycles in self-consumption and time-based control.

One case breaks this, and a lot of green-tech writing won't say it out loud. If you charge from the grid in a coal-heavy region and discharge into hours that would have been served by gas, your battery can increase emissions. You're taking 100 kWh of coal-fired power, losing 11 of it to conversion, and using the remaining 89 to avoid burning gas. That's a worse trade on carbon, even when it's a fine trade on your bill.

So the emissions math on a home battery depends almost entirely on two things: where the charge comes from, and which hours the discharge replaces. Solar-charged and peak-discharged is unambiguously good. Grid-charged in MISO or the Ohio Valley is a genuinely mixed result. Grid-charged from midday CAISO surplus is good again, because that surplus was headed for curtailment.

This is also the argument for actually controlling the thing rather than leaving it on defaults. A Powerwall sitting in Self-Powered mode with a 100% backup reserve is a very expensive UPS.

One battery does nothing. Ten thousand is a power plant.

The honest version of individual impact: your 4 kW of evening discharge, by itself, changes nothing measurable about grid emissions.

The interesting version: Tesla's California virtual power plant, aggregated out of residential Powerwalls responding to the same signal at the same time, delivered more than 100 MW during a PG&E Emergency Load Reduction Program event in July 2024. That's peaker-plant scale, assembled from hardware people bought for their own reasons and sited exactly where the load is, with no line losses on the way to the load and no permitting fight.

Aggregation is what turns a distributed asset into a dispatchable one, and utilities in California and Arizona now pay for it. The behavior that gets you paid in a VPP event and the behavior that shaves your demand charge are close to identical, which means the fleet is already mostly doing the right thing on its own, without a coordinating signal. Every Powerwall that shifts its evening load flattens the ramp a little. Flatter ramps mean fewer peakers held online in spinning reserve, which is emissions you never see on any chart because they're the emissions that didn't happen.

The setting that decides which side you're on

The setting is charge source. Where your battery's charge comes from, and when, decides whether it's fighting peakers or storing coal. Everything else is detail.

Grid Getter exists because I got tired of managing that by hand. DemandGuard watches your live grid draw through the peak window and, when a spike approaches your target, either holds the battery at its current charge or flips the Powerwall's operating mode, so the battery covers the spike instead of sitting on charge it never uses. Turn on Demand Floor and it can dip below your normal reserve during the window and restore it afterward; Weather Guard stops that extra drop while Storm Watch is active or severe weather is forecast, so you're not trading outage protection for a few dollars of demand charge. The other automations cover the rest: a Battery Threshold rule grid-charges back to a target when the battery falls below a floor you set, and a Custom Schedule rule flips the operating mode or Storm Watch on a calendar. The free plan covers live monitoring and one custom automation. DemandGuard sits on the Pro tier. SRP and APS peak windows come preconfigured, and other utilities get added as we go.

Here's the thing to go check tonight, whether or not you sign up. Open the Tesla app, pull up yesterday's energy graph, and find the Powerwall's charge breakdown, the split between solar and grid. Tesla moves this around between app versions, but it's there. Mostly solar and your battery is running clean. Mostly grid, and the next question is when. Grid charging at 1 PM in California is soaking up surplus that would have been curtailed. Grid charging at 7 PM is buying peaker output and storing it. Same setting, opposite outcome, and the app won't tell you which one you picked.