At 4 p.m. on a Tuesday in August, the electricity flowing out of a Fresno wall socket costs roughly seven times what the same neighborhood's rooftops sold back to the grid at noon. The electrons are identical; only the price changed, because the utility's clock said so.
This is the daily rhythm of California's Net Billing Tariff, the policy everyone still calls NEM 3.0. Since April 2023, the state's big three utilities have paid solar owners for exported power based on its hourly value to the grid, a figure the California Public Utilities Commission calls the avoided cost. It averages about 8 cents a kilowatt-hour, roughly a 75 percent cut from the old one-to-one credit of around 30 cents, according to the CPUC's own analysis. Meanwhile the power you buy back between 4 and 9 p.m. runs about 55 cents on PG&E's E-TOU-C rate. Sell low, buy high: Wall Street did not invent this trade, but Sacramento now enforces it by tariff.
A $13,500 Machine That Eats the Spread
A home battery is the obvious answer, and the industry knows it. Tesla's Powerwall 3, the default choice, stores 13.5 kilowatt-hours and installs for about $13,473 on average, per EnergySage's 2026 pricing data, or roughly $998 per kilowatt-hour. Enphase, FranklinWH, and a half-dozen others sell variations on the same box. Charge it with cheap noon solar, drain it through the expensive evening window, and the 47-cent spread between an 8-cent export and a 55-cent import becomes money.
Vendors dress this up as intelligence: Tesla's Time-Based Control studies your rate plan and usage patterns to pick charge and discharge times, Enphase markets something it calls AI Optimization, and the brochure language implies the algorithm is doing the earning, that a smarter battery pays back faster than a dumber one.
Show Your Work: The Payback Math
Here is the payback math for a Powerwall 3 at $13,473 installed, with every assumption labeled so you can argue with it. Assume it cycles 10 kilowatt-hours a day through the 4-to-9 p.m. window: 13.5 kWh of usable capacity minus a backup reserve and round-trip losses. Each stored kilowatt-hour displaces a 55-cent import instead of earning an 8-cent export, a net of 47 cents. Ten kilowatt-hours times 365 days times 47 cents is $1,716 a year. Divide $13,473 by $1,716 and you get 7.9 years.
Two add-ons change the picture, the first being California's Emergency Load Reduction Program, which reopened to new battery enrollments in March 2026 and pays $2 for every kilowatt-hour dispatched during grid-stress events, seven or more of them each summer; PG&E Powerwall owners have historically earned $200 to $600 a summer from it. At the $400 midpoint, payback drops to about 6.4 years. The federal math moved the other way: for homeowner-owned systems, the 30 percent residential clean energy credit expired for systems placed in service after December 31, 2025, a change confirmed by the IRS, which added roughly four years to every payback computed in 2024. Every "five-year payback" quote from last year is now stale. Check the date on any number a salesperson shows you.
What the Algorithm Actually Adds
Now the uncomfortable question for the AI industry: what does the algorithm actually add? A fixed timer that discharges the battery from 4 to 9 p.m. captures the time-of-use spread by construction. Smart dispatch's marginal contributions are real but narrow: pre-charging ahead of a forecasted storm, timing discharges around VPP events, refusing to drain the battery the evening before a cloudy day. Nobody publishes the uplift of smart dispatch over a dumb timer, which tells you roughly how large the industry believes it is, because a number that flattered the software would already be on every spec sheet.
Be careful with the studies vendors cite. A Pacific Northwest National Laboratory simulation found smart-controller batteries saved homes 13 to 26 percent on daily electricity costs through time-of-use load shifting, in minute-by-minute modeling of a ten-home neighborhood. That comparison was against homes with no battery at all, not against homes with a battery on a timer. Conflating those two baselines is how a modest software gain gets sold as the whole investment thesis. The $13,500 battery earns its keep; the intelligence is garnish.
The Case Against
Against buying, the calendar argues loudest. With the 30 percent credit dead, you are paying full sticker in 2026, and batteries degrade: Tesla warranties 70 percent capacity at year ten, and real-world Powerwalls lose something like 1 to 2 percent a year, which means that $1,716 in annual value quietly shrinks. My model ignores degradation, flattering the payback by roughly a year. Say it plainly: 7.9 years is the optimistic case.
Second, the no-solar household. Without cheap noon power to store, the battery charges from the grid off-peak at around 35 cents and discharges at the 55-cent peak, netting maybe 18 cents a kilowatt-hour after efficiency losses. That is $657 a year against a $13,473 box: a twenty-year payback on a ten-year warranty. A battery without solar is a backup purchase, not an investment, and it should be priced against a $500 portable power station and an electrician's transfer-switch quote before anyone talks about arbitrage.
Third, the counterparty. NEM 2.0 became NEM 3.0 by commission vote, and nothing stops a future commission from rewriting the spread again. You are buying a ten-year bet that the 47-cent gap persists. California's rates are the highest in the continental US for structural reasons, averaging 34.7 cents across all hours per the EIA's June 2026 data, so the bet is reasonable. Reasonable is not guaranteed.
Buy It or Skip It
If you are on PG&E, SCE, or SDG&E, have rooftop solar, and sit on a time-of-use rate, the battery pencils at roughly six to eight years with VPP income, which makes it one of the rare home improvements that pays for itself faster than the solar array it pairs with. Buy it if you will own the home for a decade, check your SGIP incentive step before signing, and ignore any AI-upsell tier if one is offered. The timer is free.
No solar on the roof? Skip the battery as an investment. Buy backup if outages scare you, but do the portable-station math first.
Outside California's time-of-use territory, run your own spread: peak rate minus off-peak rate, times ten kilowatt-hours, times 365. If that annual number does not clear $1,200, the battery does not pay. Most of the country fails this test, which is why the sales pitch travels with the rate schedule.
Limitations
That 55-cent peak figure is approximate; PG&E's bundled totals move with advice-letter filings, though the conclusion survives a 10-cent swing either way (payback lands between 6.5 and 10 years). No vendor publishes smart-versus-dumb dispatch uplift, so the "garnish" claim is inference from rate-structure logic, not a measured A/B test; NREL's dispatch research optimizes against cost and battery aging jointly but does not isolate the software's marginal value either. My model assumes 365 days of full cycling; cloudy stretches and conservative backup reserves reduce it. It ignores SGIP incentives, which would shorten payback for qualifying households, and degradation, which lengthens it. And this is a California analysis: where peak-minus-off-peak spreads run under 20 cents, the math usually fails.