Commercial Battery Storage Sizing: How Do You Choose Power and Energy Capacity

You choose the two numbers from two different inputs. Power comes off the load profile: annual peak demand minus the peak you are willing to keep buying. Energy comes off the clock: that power multiplied by the hours the load sits above the line, then by 1.8 to 1.9 to cover losses, end-of-life fade and the reserve window.

Peak height and peak length move independently. Shaving 900 kW for 15 minutes takes 225 kWh; holding the same 900 kW for 3 hours takes 2,700 kWh. Identical power, 12 times the energy.

The 1.8 comes from three discounts: 85% round-trip efficiency, 80% end-of-life capacity, 80% usable state-of-charge window. Multiply them and you get 0.54, so whatever the meter says you need, the nameplate runs about 1.8 times that.

Which of the two you push depends on the bill. Demand charges run 30% to 70% of a US commercial bill^[1], so power is what pays. Shandong prices the same kilowatt-hour at 0.2 in the deep valley and 1.2 at the sharp peak^[2], which makes energy the number that earns.

Power and energy do different jobs

Say a plant's highest 15-minute average demand over the year is 1,200 kW. With a battery on site, the grid side never goes above 800 kW. The 400 kW in between is the discharge power the system has to deliver.

What that 400 kW is worth depends on your basic capacity charge. At 30 to 45 yuan per kW per month, you're looking at 12,000 to 18,000 yuan a month, or 144,000 to 216,000 a year.

Packaged units usually ship as fixed pairs, like 100 kW with 215 kWh and 125 kW with 261 kWh. Both work out to roughly 2 hours, about 0.5C. A 4-hour block means adding another battery string.

Take that same 400 kW of shaving to a quote sheet. About $590,000 on a 2-hour unit, $750,000 on a 4-hour unit. Same inverter both times, and the extra $160,000 is all cells^[3].

How long the overage lasts decides how much energy sits behind it. An hour of overage a day, 400 kW needs 400 kWh of energy; two hours, 800 kWh; five hours across an afternoon, 2,000 kWh.

Get the power wrong and the peak never moves, so the capacity charge keeps coming in full. Under-size the energy and the battery holds for 30 minutes, the grid takes over, and the monthly peak resets anyway.

Both 1 MW / 1 MWh and 1 MW / 4 MWh get sold as "a 1 MW battery." The price difference between them comes to more than double.

Reading your meter data

Start with the meter. The transformer nameplate gives you the ceiling. It says nothing about what you actually pulled.

Pull at least 12 months at 15-minute resolution. That's 35,040 points a year, and most demand charges are calculated on exactly that window^[1]. Utilities generally hand over 12 to 24 months for free; going further back takes a request and 1 to 2 weeks.

You only need four things out of those 35,040 points. When the annual peak hit. How many metering intervals each overage lasted. What load you can't cut when the grid drops. And how much spare transformer capacity you have, plus the longest charging window in your tariff.

A peak that shows up once a year isn't worth chasing. Covering it with an extra 300 kW costs 108,000 yuan a year for one day of benefit. Peaks that show up every week are a different conversation.

If you run welders, crushers, big motor starts or injection molders, ask for 1-minute data. That's 525,600 points a year, but 15-minute averaging can hide more than 20% of the spike.

Sizing the power

The 400 kW is a battery-side figure. The inverter nameplate has to cover four things on top of it: ambient temperature derating, altitude derating, reactive power output and station auxiliary load.

Common practice is 10% to 20% headroom, so 480 kW at the 20% end. That lands on the 500 kW frame out of the standard 100, 125, 250 and 500 kW sizes. Working through those four items usually takes 3 to 5 days.

An inverter at 40 °C derates 5% to 10%. Above 2,000 m altitude, another 5%. Together they chew through most of a 10% margin before you've even started.

Size to the single worst hour of the year and you'll pay 1.5 to 2 times what the 20th-highest peak would need, for a benefit that lands once. Most shaving projects size somewhere between the 10th and 20th highest peak instead.

Keep the target line between 70% and 90% of transformer rating. Push it lower and the battery ends up carrying the transformer for hours, the discharge window stretches past 5 hours, and the energy bill climbs fast.

The battery also has to sit behind the meter. On the utility side of it, that 400 kW belongs to someone else's bill and your own peak is still 1,200 kW.

Sizing the energy

400 kW over the line for 2 hours is 800 kWh, and for 4 hours 1,600 kWh. Those are load-side numbers, and turning them into a nameplate takes three discounts.

AC round-trip efficiency is the first cut, and it covers two losses. The inverter itself runs 98% to 99%, the battery side 95% to 97%. NREL uses 85% for lithium systems^[3], and mainstream products land between 85% and 92%.

End-of-life capacity bites harder. The system still has to do the job on the last day of warranty, and warranties typically promise 70% to 80% remaining at year 10. Discounting at 80% is already the generous reading.

The reserve window takes the last cut. You can't shave with energy you're holding for an outage, and most projects fence off 20% of state of charge, leaving 80% for daily cycling.

Multiply the three and you get 0.54. Seen from the other side, a 1,000 kWh nameplate gives about 540 kWh of usable shaving.

Cabinets ship in 200, 215 and 261 kWh blocks, so a 1,000 kWh requirement lands on 5 units. Nobody sells you 4.7 of anything.

When you ask for a number, ask for usable energy. Two quotes both say 1,000 kWh, but one delivers 480 usable and the other 620. That 140 kWh gap is 112,000 yuan at 800 yuan/kWh.

Auxiliary load is the one everybody forgets. Cooling, fire systems and controls draw 2% to 5% of rated power, which is 8 kW to 20 kW on a 400 kW system. Run it 2 hours a day for 330 days and that's 5,300 to 13,200 kWh a year.

What duration costs

Batteries are sold in 1, 2, 4 and 8-hour blocks, and EIA's market survey tracks how capacity, duration and cost move across the operating fleet.^[4] Double the duration and the per-kW price rises 15% to 42%, while the per-kWh price falls 29% to 42%.

NREL puts overnight capital cost at $1,277, $1,474, $1,869 and $2,658 per kW for the 1, 2, 4 and 8-hour blocks. Per kWh that's $1,277, $737, $467 and $332, a 3.8x spread from shortest to longest^[3].

Battery costs have fallen faster over the past decade than any other clean energy technology, which is why the storage line in a project budget behaves nothing like the module line.^[5]

Moving a 1 MW system from 2 hours to 4 hours adds $395 per kW, or $395,000. You get double the energy throughput, and the equipment only costs 27% more^[3].

Chinese pricing quotes per Wh: 0.8 to 1.2 yuan, which is 800 to 1,200 yuan per kWh.

A 1 MW system on a 2-hour block moves about 660,000 kWh a year. On a 4-hour block it moves 1.32 million. That gap is what separates how commercial and utility-scale projects are justified: one is buying peak reduction and the other is selling energy.

Equipment parameters at a glance

The front page of a spec sheet gives you three numbers: power, energy, price. The twelve below decide how much of that nameplate you can actually use.

The battery pack runs 35% to 45% of initial capex. At 800 yuan/kWh, a year-10 replacement costs 280 to 360 yuan per kWh.

A 500 kWh indoor project works out to 4 m² at 0.8 m² per 100 kWh. Add a 0.9 m service aisle and fire separation and you're actually sitting on 15 to 25 m².

The demand charge side

On a US commercial bill, demand charges are billed on the single highest 15-minute window of the billing period.

"Demand charges are typically based on a customer's maximum average demand over a 15-minute interval within the billing period. For many commercial customers they represent 30% to 70% of the total monthly electricity bill." — NREL and LBNL, survey of US demand charges^[1]

That survey covers more than 10,000 utility tariffs and roughly 70% of US commercial buildings^[1]. At the American rate of $10 to $20 per kW per month, shaving the same 400 kW is worth $4,000 to $8,000 a month.

One 15-minute peak sets the whole month, and it's usually three things switching on together. A 150 kW air compressor, an 80 kW cold-store defrost cycle and a 120 kW charger overlap, and 350 kW is enough to set a new record. The charger is the only one of the three a site can schedule, which is why solar-linked EV charging starts from the daily charging demand curve rather than the array.

Some tariffs also ratchet. The floor for the next 6 to 11 months gets set at 50% to 80% of the highest demand in the past 12 months. At a 70% ratchet, one 1,200 kW spike in July keeps billing at 840 kW for the rest of the year.

The peak-valley spread side

Shandong splits the day into five bands: sharp peak, peak, flat, valley and deep valley. Deep valley drops to 0.2 yuan/kWh and the sharp peak climbs to 1.2 yuan/kWh, roughly 90% below and 100% above the flat rate. The widest spread approaches 1 yuan^[2].

Arbitrage only pays on the energy you discharge, and one cycle moves exactly the battery's own capacity. A second daily cycle therefore doubles the annual throughput without adding a single cell. What it needs is a second valley window to charge in.

That window has to run 2 to 4 hours. In provinces that give only one hour at midday, the second cycle covers about half the energy.

A hotel in Nanjing draws 17,600 kWh a day during peak hours. It runs a 9 MW / 18 MWh system on valley-charge, peak-discharge and saves about 3.67 million yuan a year^[2]. At 18,000 kWh a day, 330 days a year, that's 5.94 million kWh, around 0.62 yuan per kWh, or 11,000 yuan a day.

Shandong is widening spreads. Sichuan hands retail time-of-use scheduling and pricing to the market. Jiangsu extended eligibility to nearly all commercial and industrial users while letting those below 100 kVA opt out^[2].

Zhejiang raised its peak-valley price ratio in 2025 but excluded transmission and distribution charges, government funds and surcharges from the floating portion. The arbitrage you actually capture shrank by about 30%. A headline ratio is not the same as money in hand^[2].

How much is already out there

US grid-scale battery storage passed 26 GW of cumulative capacity by the end of 2024, with 10.4 GW added that year, up 66%. That's 2% of the country's 1,230 GW of generating capacity^[6].

On the 2023 year-end count, California held 7.3 GW and Texas 3.2 GW, with every other state combined adding about 3.5 GW^[7] for roughly 14 GW nationally, before the 10.4 GW that came online during 2024. The DOE lists five uses: frequency and voltage regulation, energy arbitrage, backup power, peak shaving, and deferring transmission and distribution investment^[8], and pairs that list with a cost target of $0.05 per kWh-cycle for long-duration storage by 2030.^[9]

For a commercial site, only three of those cash out. Shaving, arbitrage and deferred capacity upgrades. The other two are things a utility buys.

The commercial segment is the smaller and more varied one. industrial and commercial installations run from a few hundred kW on a factory roof through to multi-megawatt ground-mount systems.

When there's solar on site

Solar generates at midday, peak prices land in the evening, and the two curves sit 6 to 8 hours apart. That gap is what an integrated PV and storage design has to close. Jiangsu made 11:00 to 13:00 in summer, and 10:00 to 14:00 in spring and autumn, valley periods to push consumption toward the solar hours^[2].

A 1 MW array on a clear summer midday puts out 700 to 800 kW, and where it lands in that range comes down to module specification and cell temperature. How much array the site needs at all runs off the same 12 months of meter data used for sizing a commercial PV system. Subtract the 400 kW the plant is drawing and you have 300 to 400 kW spare. Exported at the 0.3 yuan/kWh valley rate, 1,400 kWh brings back 420 yuan.

By evening the array is at zero. The power number doesn't move. The energy number does: the 800 kWh sized for 2 hours of overage now has to absorb another 1,400 kWh from midday, so 2,200 kWh. That is 5.5 hours on a 400 kW system, which pushes the choice up to the 8-hour block.

Midday surplus is 1,400 kWh and the evening gap is 2,000 kWh. The 600 kWh in between still gets bought at the top rate, which is 720 yuan at 1.2 yuan/kWh.

The battery still cycles once a day, but each cycle moves 1,400 kWh instead of a few hundred. Over 330 days that's 462,000 kWh, and 4.62 million over ten years. Warranty throughput caps usually sit right about there.

Backup and islanding

Backup energy is the critical load multiplied by the hours you need to hold it. 40 kW for 4 hours is 160 kWh, sitting idle almost all year and discharging once or twice.

On an 800 kW plant, the load that truly can't drop is often about 5%, or 40 kW. A DOE project added a few non-critical loads and the system couldn't ride through a 5-day outage. Once those loads moved back where they belonged, it did^[10].

Price the backup block on its own. At 800 to 1,200 yuan/kWh, 160 kWh is 128,000 to 192,000 yuan. That's an insurance premium you claim a few times a year, and it stretches the payback on everything else the system does.

Share one system between two jobs and you are carving up the same tank. Holding 200 kWh back out of a 1,000 kWh system takes a fifth of the nameplate out of the daily cycle for good.

40 kW for 3 days is 2,880 kWh. For 4 hours it's 160 kWh. An 18x difference, and two completely different systems. Pull the outage duration off the record. 1.5 times your longest actual outage is a common starting point.

For evaluation, the DOE runs an optimization model that puts outage duration, critical loads and equipment constraints into one solve, outputs an 8,760-hour dispatch profile and reads power and capacity off it^[10], then back-tests the result against 12 months of real data.

Life and capacity fade

Year-10 capacity runs 20% to 30% below year one. Calendar aging accounts for 15 to 20 points of that, cycling makes up the rest. Do the sizing against the year-10 figure.

LFP cells at 80% depth of discharge are rated 4,000 to 6,000 cycles. Warranties usually read "10 years or 6,000 cycles, whichever comes first," with 70% to 80% capacity left at the end.

Cells age sitting still, too. Calendar fade runs 1.5% to 2% a year, so 15% to 20% across ten years. Every 10 °C above nominal roughly doubles that rate. A plant room held at 35 °C instead of 25 °C loses 15 percentage points more capacity over ten years. The mechanism is documented in the national lab work on cell degradation, and it is the reason temperature control in the plant room earns its cost back on retained capacity.^[11]

One cycle a day, 330 days a year, ten years: 3,300 cycles. Comfortably inside 6,000, but that's already 80% of the 4,000-cycle lower bound. Two cycles a day is 6,600, straight past the stated life.

The extra 600 cycles have to come back somewhere between year 5 and year 8. The usual fix is adding a 100 to 200 kWh cabinet, which is 80,000 to 160,000 yuan at 800 yuan/kWh. Or you write the augmentation responsibility and cost into the contract up front.

Interconnection and permitting

IEEE 1547-2018 sets the technical requirements, test items and operating requirements for distributed resources connecting to the grid. Storage counts as one, and the test set doesn't scale down with size. A few-kilowatt interconnection runs the same list^[12]. On the market side, Order No. 841 requires US wholesale market operators to write rules that let storage take part in energy, capacity and ancillary services markets.^[13]

At the end of 2025, US interconnection queues held 749 GW of storage, down 16% year on year. Median time from request to commercial operation runs past 5 years^[14], which makes the queue a longer wait than the hardware.

Another 549 GW have signed agreements but haven't been built, 161 GW of that in storage^[14]. Of applications filed between 2000 and 2020, only 13% of capacity had reached commercial operation by the end of 2025. 75% had withdrawn^[14].

Interconnection voltage steps with size. 100 to 500 kW usually connects at 380 V, 500 kW to 2 MW at 10 kV, and above 2 MW at 35 kV. Step up a level and switchgear plus transformer investment roughly doubles.

Transformers and switchgear quote in quarters. If distribution capacity is short you need an upgrade before you can connect at all, and a 1 MW upgrade often quotes in the same range as the battery itself.

What the safety codes cap

NFPA 855 requires 3 ft, about 0.91 m, between lithium battery units over 50 kWh, and between units and walls. Want a tighter layout? You need large-scale fire testing under UL 9540A to justify it^[15][16].

Indoor commercial systems above 600 kWh go into a dedicated battery room with explosion-proof ventilation and an emergency disconnect^[15].

NFPA 855 comes from a private standards body, so it only bites where a jurisdiction adopts it. In most US states the hook is the International Fire Code, whose energy systems chapter sets its own limits on where a system may sit and what has to be documented before a permit issues.^[17]

A 2 MWh indoor project avoids the dedicated room by splitting into 4 separate zones of about 500 kWh each.

Cabinet sizes cluster in two bands, 30 to 50 kWh and 100 to 215 kWh. The smaller band sits under the 50 kWh separation threshold, so 10 of them in a row free up 8 to 9 m of aisle compared with the larger band.

China's counterpart is GB/T 51048-2025, the design standard for electrochemical energy storage stations, issued 10 December 2025 and effective 1 April 2026, replacing the 2014 edition^[18].

"The national standard Design Standard for Electrochemical Energy Storage Stations is hereby approved as GB/T 51048-2025, effective 1 April 2026. The previous standard GB 51048-2014 is abolished on the same date." — Ministry of Housing and Urban-Rural Development announcement, translated^[18]

The new edition puts lithium and sodium battery buildings off-limits below grade or half-below grade, and requires automatic suppression tied to ventilation in lithium battery rooms and prefabricated enclosures^[19].

It also only applies from 500 kW rated power and 500 kWh rated energy upward. A 400 kW / 1,000 kWh commercial project falls outside its scope and leans on local or industry standards instead^[18].

By the end of 2025, China's new-type storage fleet passed 130 GW, more than 40 times its level at the end of the 13th Five-Year Plan. Over the same period there were more than 160 safety incidents in electrochemical storage^[19], roughly one for every 810 MW installed.

What to put in the RFQ

Two quotes on the same 1,000 kWh cabinet can differ by 30%, and the gap is definitions. Five items decide which one you are actually comparing.

• Usable energy, and the conditions it applies to: at what temperature, at what discharge rate, measured in year 1 or in year 10.
• AC round-trip efficiency, and the load factor it was measured at. Some datasheets quote full load only, and an inverter is 3 to 5 points worse at 30% load, right where peak shaving tends to sit between 20% and 40%.
• End-of-life capacity and who pays for augmentation. If the warranty says "70% at year 10," ask which test condition that refers to, who funds the shortfall, and what a percentage point is priced at.
• Continuous discharge power. Some units quote 250 kW peak and 100 kW continuous, a 2.5x gap. Peak numbers don't shave anything.
• The certification level. UL 9540 signs off the whole system. UL 9540A is a test method for thermal runaway propagation, and the report doesn't come bundled with the certificate^[16].

Those five items each carry a price. A 140 kWh usable-energy gap is 14%, another 3 points of efficiency is 3% to 4%, and stacked together they move annual discharge by 18%.

Which number to accept on

Acceptance runs three times: at 1 month, 6 months and 12 months after commissioning. Only the 12-month round counts. It's based on at least 12 months of metered data, which is 8,760 points at hourly resolution. Demonstrations and spot tests don't qualify^[20].

The headline figure is measured capacity divided by nameplate. A ratio that sits below 1 for an extended period points to fade, temperature derating or a control strategy problem, and it should be settled before you sign^[20].

Write the test conditions into the contract. Ambient temperature, starting state of charge, discharge rate and cut-off voltage all get fixed. Move the temperature by 10 °C on its own and the capacity reading shifts 3% to 5%.

IEC 62933-2-1 already defines the test methods for rated energy, rated power, round-trip efficiency and response time^[21].

On the procurement side, FEMP publishes a lithium-ion storage technical specification template. Requirements are grouped under usable energy, efficiency, warranty, communications and fire protection, five categories you can lift straight into your own document^[22].

The first valid evaluation report can't exist until 12 months after commissioning. Monthly readings in between are process checks only. The report has to give three figures: capacity ratio, availability and equivalent full cycles^[20].

Common questions

Can you size off the transformer?

No. Transformer capacity is a ceiling. It tells you nothing about the shape of your load. Two plants with identical 1,000 kVA transformers: one peaks for 15 minutes and needs 60 kWh, the other peaks across 4 hours and needs 960 kWh. Same transformer, 16 times the battery.

Can one system do three jobs?

Yes, but they all draw from the same tank. Shaving and arbitrage both discharge. Backup has to sit there doing nothing until the day it is needed, and sharing one capacity budget means something gets discounted.

A common split sizes power for shaving, energy for arbitrage, and fences off 20% of state of charge for backup. On a 1,000 kWh system that fence is 200 kWh of nameplate which never enters the daily cycle, and it carries 160,000 yuan of capital at 800 yuan/kWh.

Why can't I use every kWh?

Nameplate is a cell-level figure under ideal conditions. Depth-of-discharge limit, AC round-trip efficiency, end-of-life fade and temperature derating each take a cut. A 1,000 kWh system gives up about 540 kWh for shaving.

What if capacity drops below the warranty line?

Depends on the contract. Common terms are full pack replacement, a top-up, or a payout against the missing energy. What you want settled early is the trigger: who tests, under what conditions, and how long results take.

The usual threshold is capacity below 70% of nameplate, and a full test takes 2 to 4 weeks. The downtime counts against availability.

The numbers to run before you order

1,200 kW minus 800 kW is 400 kW. Add 20% derating headroom and you're at 480 kW, so you take the 500 kW frame.

Energy works out to 400 × 2 hours × 1.8. Call it 1,440 kWh of nameplate. On a 480 kW inverter that's 3 hours, which isn't a standard block, so you step up to the 4-hour one.

1,440 kWh at 215 kWh per cabinet is 6.7 cabinets. Round to 7 and the power becomes 700 kW, 46% above the 480 kW you sized. A single 1.5 MWh container with a 500 kW inverter is where the two numbers line up again.

Capital is 1,440 kWh at 0.8 to 1.2 yuan/Wh. That's 1.15 to 1.73 million yuan. O&M runs another 1% to 2% of capex a year, which is 120,000 to 350,000 yuan over ten years.

Cycle ceiling is 2 cycles a day × 330 days × 10 years. That's 6,600, already past the 6,000-cycle figure most warranties quote.

Sources and method

Twenty-two institutional sources are cited inline, with full entries at the end. Chinese tariff and standard information comes from public releases by the National Energy Administration and the Ministry of Housing and Urban-Rural Development.

Several figures are industry ranges rather than single official values, with no single authority behind them: C-rate, depth of discharge, round-trip efficiency, overload capability, response time, footprint, weight, noise, ingress protection, availability, O&M cost, EPC pricing, cycle life, fade rates, warranty bands, free meter data over 12 to 24 months, a 2 to 4-hour midday valley in most provinces, 8 to 12 interconnection test items, and a 1 MW upgrade quoting in the same range as the equipment.

Derived figures are calculated from values given in the text above. That covers the 3.8x spread, the 0.54 conversion factor, 0.62 yuan per kWh, 1.15 to 1.73 million yuan of project cost, and 80,000 to 160,000 yuan of augmentation budget.