I’ve watched eight-figure sizing decisions turn on a definition of the word “usable.”

More money is lost in this industry to sizing mistakes than to fires, by orders of magnitude. Most of those mistakes reduce to one confusion: power is not energy, and nameplate is not what you deliver. Let’s pin down what each word actually means.

If you want to see where each percent actually goes on the way in and out, I mapped the full loss stack cell-to-meter in this Cell Losses & Efficiency visual. It’s the same waterfall we’re about to build, drawn out end to end.

MW, MWh, duration, C-rate

  • Power (MW): how fast the system charges or discharges. Set by the PCS and grid connection.
  • Energy (MWh): how much the system stores. Set by the batteries.
  • Duration (hours): energy ÷ power. A 100 MW / 400 MWh system is a “4-hour” battery.
  • C-rate: power relative to energy. That same system runs at 0.25C. A 1-hour battery runs at 1C — harder on the cells, warmer, faster-aging. The direction matters more than any single number: warranties on modern LFP typically cap continuous C-rate somewhere around 0.5C for 2-hour-and-longer systems, and every step up in C-rate raises internal heating and pulls warranted cycle life down. Most grid-scale stationary systems live in the 2–4 hour (0.25–0.5C) band because below roughly 0.5C the cells barely warm up and the warranty curve stays flat, which is where the cost per usable MWh actually bottoms out.

When someone says “a 500 MW battery,” your first question is always: for how long? A 500 MW / 500 MWh asset and a 500 MW / 2,000 MWh asset are different machines with different economics.

Nameplate vs usable: the waterfall

The number on the container sticker is not the number at the meter. Energy erodes through a waterfall:

  1. DC nameplate — the sum of cell capacity at beginning of life (BOL).
  2. SoC operating window — systems typically don’t swing 0–100%; reserve margins protect the cells and guarantee response headroom. Many LFP systems are operated roughly 5–95% SoC, i.e. ~90% usable depth of discharge, and some duties (heavy cycling, tighter warranties) run narrower still. That’s a real deduction: right off the top, ~10% of nameplate is fenced off before degradation even enters the picture.
  3. Degradation — capacity fades with age and use. End of life (EOL) is a contractual definition, commonly ~70% of BOL (some contracts use 80%).
  4. Conversion losses — PCS, transformers, cabling.
  5. Auxiliary loads — cooling and controls consume energy. Note that much of this is a continuous load driven by ambient conditions rather than by how hard you cycle, so its share of delivered throughput actually rises as utilization falls. A lightly-cycled asset in a hot climate can lose a surprising fraction of its output to thermal management alone.

What your offtake contract cares about is usable AC energy at the point of interconnection, at a given point in project life, at a reference temperature. Every word in that sentence has been the subject of a dispute somewhere.

Round-trip efficiency: build the budget, don’t quote a number

RTE is a stack of small losses. Representative modern LFP figures:

  • Battery DC round trip: ~95–97%
  • PCS: ~98–98.5% each direction
  • MV transformer: ~99.5% each direction
  • Auxiliary consumption: ~1–3% of throughput as a rule of thumb — but remember aux is largely a continuous load, so its share of throughput rises as utilization falls

Multiply the mid-case through: 0.96 × 0.985² × 0.995² × 0.98 ≈ 0.90 — about 90% AC-to-AC. Slide each assumption to its weaker end (95% battery, 3% aux, hotter climate, more idle standby) and you land closer to 86–88%. That’s why quoting a single round-trip efficiency without stating the measurement boundary, temperature, and duty cycle is meaningless. Always ask for all three.

Walk it back: one worked example

Numbers in isolation don’t teach the lesson — the compounding does. So let’s chain the whole stack once, walking backward from what the contract cares about to the batteries you actually have to buy. Say the guarantee point is 400 MWh usable AC at the POI, at year 20, and work back to DC nameplate at BOL:

  1. POI → battery cells (discharge path). The ~90% figure is round-trip, so the one-way discharge chain — from the DC cells out through the PCS, transformer, and aux to the meter — is about √0.90 ≈ 95% (round-trip loss splits roughly evenly across charge and discharge, so each leg carries about half of it). So to deliver 400 MWh at the POI you must pull about 400 ÷ 0.95 ≈ 421 MWh back out of the cells.
  2. Usable → installed capacity (SoC window). If you only cycle ~90% of the pack (≈5–95% SoC), the installed DC capacity present in year 20 must be about 421 ÷ 0.90 ≈ 468 MWh.
  3. Year-20 → BOL (degradation). If the warranty holds ~70% retention at year 20, the BOL DC nameplate must be about 468 ÷ 0.70 ≈ 670 MWh.
The same walk, forward — 400 MWh at the meter demands ~670 MWh of nameplate
DC nameplate at BOL — what you procure 670 MWh
× 0.70 — year-20 retention (EOL definition) 468 MWh
× 0.90 — SoC operating window (≈5–95%) 421 MWh
× 0.95 — one-way discharge efficiency — usable AC at the POI, year 20 400 MWh

Bars are % of BOL nameplate. No single step is dramatic — the compounding is.

So a clean-sounding “400 MWh” contract quietly demands on the order of ~670 MWh of DC nameplate at BOL. That’s roughly a 1.6–1.7× overbuild once every layer of the waterfall compounds. Change any assumption (tighter SoC window, a 75% EOL definition, a hotter site) and that multiplier moves. The point isn’t the exact figure — it’s that no single loss is large, yet stacked together they nearly double the battery you procure. Size to the guarantee point, not to the sticker. We computed that overbuild across the full range of realistic assumptions — it runs 1.3–1.9×, and the single lever that moves it most is the one buyers rarely negotiate: the end-of-life SOH the warranty is written to.

Degradation: two clocks running at once

Batteries age two ways simultaneously:

  • Cycle aging — driven by energy throughput, depth of discharge, C-rate, and temperature. A typical grid duty is roughly one full cycle per day, about 365 equivalent full cycles a year, or ~7,000 over a 20-year term, and LFP warranties usually cap total throughput somewhere near that figure.
  • Calendar aging — driven by time, temperature, and time spent at high state of charge. A battery degrades even sitting still. As a rough guide, calendar fade roughly doubles for every ~10 °C of sustained cell-temperature rise (an Arrhenius rule of thumb), which is why siting and thermal management shift the whole curve.

Modern LFP warranties typically guarantee retained capacity (for example, to 70% over 15–20 years) conditional on an operating envelope: cycles per year or total throughput, temperature limits, SoC dwell restrictions, and C-rate caps. The degradation curve is typically: modest early fade, a long near-linear middle, and — if the battery is pushed too hard — a “knee” where fade accelerates. That knee tends to show up down near the low end of retention (often below ~70–80% of BOL, or once you push past the warranted cycle count), which is exactly why EOL is usually defined at 70–80%: it keeps the project on the linear part of the curve. Project economics should never plan to visit the knee — treat “below EOL retention or beyond warranted cycles” as the fence you don’t climb.

Meeting a 20-year contract: overbuild vs augmentation

Say you’ve contracted 100 MW / 400 MWh usable at the POI for 20 years. Two strategies:

Strategy A — Overbuild at BOL. Install enough DC capacity that even at year-20 degradation you still clear 400 MWh. If year-20 retention is ~70%, you need roughly 400 ÷ 0.70 ≈ 570 MWh usable-equivalent at BOL — before conversion and auxiliary margins, and assuming your dispatch stays inside the throughput warranty that the 70% figure is conditioned on. You’re buying batteries in year 0 that you won’t need until year 15 — at year-0 prices.

Strategy B — Augmentation. Install closer to the requirement plus near-term margin — say ~450 MWh at BOL, enough to clear 400 MWh through the first several years — then add capacity in planned tranches (commonly somewhere around years 5–8 and 12–15, project-specific) to top the fleet back up. That tranche timing isn’t arbitrary: you augment just before the degrading pack would otherwise fall below the contracted usable floor, so the year depends directly on the retention curve and how much near-term margin you installed on day one. Instead of buying that last ~120 MWh in year 0, you buy it in, say, year 7. That’s the whole bet. Lithium pack prices have fallen roughly 90% since 2010 (BNEF), from around $1,200/kWh to about $108/kWh by 2025, so a tranche deferred by seven years has historically cost a fraction of its day-0 price, even before you discount it. Add that time-value on top (a dollar spent in year 7 is worth less than a dollar spent today) and deferral usually wins on NPV. The catch is you’re forecasting a price you don’t control.

Augmentation is now the mainstream approach, but it isn’t free of engineering: new racks age differently than old ones, DC voltage windows must be compatible, and the EMS must manage a mixed-age fleet. Common patterns include dedicated augmentation containers (“sidecar” augmentation) and PCS/balance-of-plant sized on day one for the future capacity.

Sizing the power side: where the steel gets expensive

Everything above sizes the energy reservoir. The MW side is decided by a different set of hardware — PCS fleet, transformers, switchgear, the interconnection itself — and sizing mistakes here are paid for in steel rather than cells. Work it as its own stack:

  • Start at the POI. Your interconnection agreement grants megawatts at the point of interconnection — net of every loss behind it. That number, not the battery, is usually the binding constraint.
  • Size the PCS fleet in MVA, not MW. Grid codes make the plant deliver reactive power, so the converter fleet is rated in apparent power: at 0.95 power factor, delivering 100 MW takes roughly 105 MVA of PCS. Utility-scale blocks typically run 1–5 MVA each, so that’s a 20–100 unit fleet decision — and it drags transformer kVA, switchgear, and collection cabling with it.
  • Check the derating before the datasheet number. PCS continuous ratings are quoted at a reference temperature; hot sites and high altitude shave real percentage points off deliverable MW exactly when the market pays most. Ask for the capability curve at your site conditions, not the brochure’s.
  • Then reconcile duration. The MW you can inject and the MWh you installed set the duration the market actually sees. Oversize the power side against a fixed reservoir and you’ve bought a shorter-duration product — sometimes deliberately (ancillary-heavy strategies), often accidentally.

The energy stack and the power stack meet in one line item: the guarantee point. Usable AC MWh at the POI, deliverable at rated MW, at a defined year and temperature. If your spec sheet can’t state that sentence with numbers, the sizing isn’t done.

Run your own numbers: the free BESS sizing calculator on this site walks this exact waterfall — efficiency, auxiliaries, SoC window, degradation, and the PCS fleet in MVA — with your contract’s inputs.

A sizing checklist that survives contact with reality

  1. Define the service: what product, what duration, how many cycles per year, what seasonality?
  2. Fix the guarantee point: usable AC MWh at the POI, at which year, at what temperature.
  3. Build the loss waterfall back from that point to DC nameplate at BOL.
  4. Choose overbuild vs augmentation with a real price forecast and a real O&M plan.
  5. Check the warranty envelope against the intended dispatch — then check it again with the trading team in the room.
  6. Stress-test with hot-climate auxiliary loads and a conservative degradation curve, not the vendor’s marketing curve.

FAQ

What does “4-hour battery” mean? It can discharge at full rated power for four hours: energy (MWh) = 4 × power (MW). It can discharge at lower power for proportionally longer.

What is BESS augmentation? Adding battery capacity partway through project life to offset degradation and maintain contracted energy — usually cheaper than overbuilding on day one, at the cost of added engineering and future procurement risk.

Why is my usable energy less than nameplate? SoC reserves, degradation, conversion losses, and auxiliary consumption. Contracts should always specify usable, at the POI, at a defined year and temperature.

Degradation modeling, warranty structures, and augmentation economics each get dedicated modules in my Grid-Scale BESS: Complete Guide — including the augmentation-timing miss that left one project short on contracted energy in year 12, and how to spot it in the warranty envelope before you sign.