Battery gigafactory: why you don't know what it costs you to manufacture a kWh

Battery gigafactory: why you don't know what it costs you to manufacture a kWh

• Cronometras Team

Real cost per kWh and lead times in battery gigafactories: what to measure in the dry room, formation and aging to decide with data.

You have invested in a lithium cell plant with tens of GWh of nameplate capacity. The problem comes afterwards: the actual conversion cost per kWh does not match your model, the lead times you promise your customers depend on a learning curve that no one has measured, and Finance discovers too late that installed capacity is not what the plant delivers. A gigafactory that announces 40 GWh/year actually produces 28-32 GWh in year 1, 32-36 GWh in year 2 and only exceeds 40 GWh in year 3. That difference is your margin.

The context does not forgive ignorance: declared global gigafactory capacity exceeds 2.500 GWh/year in 2025 (according to Benchmark Mineral Intelligence) and the European figure will grow from 120 GWh in 2018 to 789 GWh projected for 2030. Whoever knows how much it costs to manufacture a kWh and how long it takes to commit it negotiates from a position of advantage; whoever does not promises lead times they cannot sustain.

What is happening in your plant

More than 60 % of the conversion cost (indicative figure) of a cell is concentrated in three areas where time cannot be accelerated by more operator effort: the dry room, electrochemical formation and aging. All three are poorly managed when they are treated as a conventional assembly line.

The dry room is an expensive resource that people consume. It is a cleanroom with a dew point between −40 °C and −60 °C (classified ISO 7 under standard ISO 14644) where the electrodes are stored, the cell is wound or stacked and the electrolyte is injected. Every person who enters and every door opening degrades humidity control. That is why the resource to watch is not an operator's minute, but the square-metre-minute of dry room: time is lost in doors, missing consumables and machine stoppages, not in slow hands.

Formation cannot be accelerated without breaking the cell. After assembly, each cell undergoes slow charge and discharge cycles to form the SEI layer (solid electrolyte interphase), a passivating film 3-10 nm thick on the anode. That layer takes as long as it takes: if the charge rate is pushed (for example 0,5C instead of 0,05C), it comes out heterogeneous, metallic lithium deposition appears and the cell degrades prematurely. The industry's usual protocols (indicative data) are:

Protocol Charge rate Total time Application
Conventional 0,05C / −0,05C ≥ 24 h (24-48 h) R&D, premium cells
Accelerated 0,2C / −0,2C ≈ 14 h Mass production (LG, CATL)
Hot formation (45 °C) 0,3C / −0,3C ≈ 8-10 h NMC only; emerging

These are physicochemical limitations, not debatable habits. Trying to “improve” a 14 h formation guarantees that the next batch comes out defective.

Aging freezes working capital. Before shipping, each cell spends typically 28 days in an aging chamber (common industry practice, not required by UN 38.3 or IEC 62660-1): 28 days of idle product with no accounting value added, but essential, because in that window most latent defective cells manifest themselves —internal micro-shorts, early dendrites, unstable SEI— (indicative example). Shortening it saves process time and ruins the customer return rate. The calculation is simple: chamber inventory is daily production multiplied by the aging days, and the required chambers come from dividing that inventory by the capacity of each chamber. With 5.000 cells/day × 28 days = 140.000 cells sitting in the chamber. If aging is reduced to 14 days, inventory falls to 70.000 cells and the required chambers are halved, but the detection of latent failures is shortened.

To this add the cycle rates, which throw off any director accustomed to another industry. On an automotive line a car is built every 60-90 seconds and between 30 % and 50 % of operator time is value-adding manual work. In a gigafactory the rates run from 2,5-4,0 seconds per cell on mature Chinese lines to 6,0-9,0 seconds per cell on a European ramp-up, and that share of value-adding manual work falls to 8-12 % (indicative figure). The bulk of the value is provided by the machine and time is lost in humidity, formation and logistics: managing this with body-shop line criteria leads to wrong decisions.

How it is solved: measure to decide

The operations director's question is not “how long does the operator take?”, but “what does it cost me to manufacture one unit and what date can I commit to?”. To answer it, four things must be measured and cross-referenced.

1. How much time goes into each type of activity. The right tool is work sampling (technique described by Tippett in 1935; glossary of terms in standard BS 3138): instantaneous observations at random moments, without the operator knowing when they are being observed and with more than 15 minutes between rounds. The application calculates the number of observations required for 95 % reliability. The result breaks down plant time into actionable categories for this sector: value-adding manual work, automated work, inspection, non-value tasks (adjustments, coil changes, scrap collection), internal movements and four classes of stoppage —humidity, formation, logistics and quality— each with its corrective department. Without that snapshot, a drop in performance may be a badly opened door, a saturated filter or a miscalibrated rack, and all of them get confused.

2. What it costs to manufacture one unit. Where there is measurable manual work —electrode coating inspection, manual stacking, tab alignment— it is timed and the standard time is calculated: what it costs to manufacture one unit and the basis on which planning is done. This is the work that Cronometras (cronometras.com) solves: time study with a stopwatch, ILO and TAL allowances, synthetic performance rating, standard time calculation and export to PDF, Excel or JSON. Predetermined time systems such as MTM (Maynard, 1948) contribute in repetitive manual assembly of module and pack, but not in the automated phases nor in the physical formation and aging processes: 70-90 % of the value is automated and the machine's control logic has no “movements” to measure.

3. What formation costs in energy, not just in hours. The electrical consumption of formation can be 5-8 % of the conversion cost (indicative figure), so formation is quantified in kWh per cell. Only in this way can protocols and chamber decisions be compared with money on the table.

4. What your own data says. A modern plant generates more than 10.000 variables and almost no one uses them: the records from your GMAO —which your team enters manually— give the mean time between failures and repair time for each critical machine; line control gives the actual cycle per cell and per shift; the dry room sensors give the actual dew point and the door openings; the formation racks allow defective cells to be predicted before the aging is finished; and the automated guided vehicles (AGV) reveal waiting caused by blockages. Those who only ask for “an Excel with the observations” lose 70 % of the available information.

The result of cross-referencing the four measurements is a short list of levers prioritized by euro recovered, with a typical 12-week schedule from the start:

Weeks What is done What you get
1-2 Process mapping and area selection Process map and areas to observe
3-4 Work sampling with the industry taxonomy Actual distribution of plant time
5-6 Time study of manual tasks Standard time per unit in those tasks
7-8 Analysis of your GMAO data Failure catalogue by machine (MTBF/MTTR)
9-10 Cross-diagnosis The 3-5 largest levers, prioritized
11-12 Quick wins and measurement New effectiveness indicator and verification

Without a consolidated baseline measurement at the start, no improvement reported afterwards is credible: the before and the after must be measured with the same methodology and in equivalent periods.

Simulated practical case: from 58 % overall effectiveness to 71 % in 6 months

Simulated practical case. Example built for teaching purposes from plausible industry parameters. It does not correspond to a real Cronometras implementation nor to any specific client. No figure belongs to any identifiable real manufacturer: the data are plausible within the range published in the industry's technical literature, but they are not figures from any real plant.

Starting point. Plant with 32 GWh/year nameplate, 210 direct operators, 3 shifts × 8 h × 7 days, NMC 622 prismatic cells for automotive, 16 GWh/line × 2 lines (simulated practical case). Initial overall equipment effectiveness (OEE): 58,3 % (availability 75 %, performance 81 %, quality 96 %; 58 % rounded). Initial conversion cost: ≈ 9,8 €/kWh (indicative estimate for a European plant in ramp-up).

Weeks 1-3: time inventory. Work sampling in 6 zones, with the number of observations the application calculates for 95 % reliability, over 5 full days (3 shifts/day), with a margin of ±5 % (p = q = 0,5). The time snapshot came out as follows:

Category % of time
Value-adding manual work 8 %
Value-adding automated work 57 %
Value-adding inspection 6 %
Non-value manual tasks 8 %
Internal movements 5 %
Humidity stoppages 4 %
Formation stoppages 6 %
Logistics stoppages 4 %
Quality stoppages 2 %

Weeks 4-5: diagnosis. Formation stoppages (6 % of the time) were the largest lever: the chambers tripped falsely due to miscalibrated thermistors, with a drift ≥ 5 % after 6 months of use. Value-adding manual work (8 %) was already at its ceiling with automatic stacking. In non-value tasks (8 %) there was room: scrap collection and coil changes, 3 % of the time. And logistics stoppages (4 %) were structural —delays in electrolyte supply—: these are corrected with a 7-day safety stock, not by measuring.

Weeks 6-24: implementation. Redundant thermistors and monthly calibration: formation stoppages fall from 6 % to 2 %. Automatic splicing of foil coils (auto-splice): non-value tasks fall from 8 % to 5 %. New dry room opening protocol: humidity stoppages fall from 4 % to 2 %.

Result after 6 months:

Metric Before After Difference
OEE 58 % 71 % +13 pp
Availability 75 % 84 % +9 pp
Performance 81 % 85 % +4 pp
Quality 96 % 99 % +3 pp
Conversion cost 9,8 €/kWh 8,6 €/kWh −12,2 %

Conversion cost falls ≈ 12 %, within the indicative range of −10 % to −15 % handled for ramp-up plants with process improvement interventions. These numbers are neither miraculous nor normal: going from 58 % to 71 % OEE in 6 months is ambitious but feasible —in new European plants values around 65-70 % are handled (indicative figure), and this plant would sit in the high band—. What is not realistic is any promise of “+30 % OEE in 3 months”: that is marketing, not measurement.

What you gain and what you decide with this data

  • You know what it costs you to manufacture a kWh, by line and by shift. Without that, the price you offer is set blindly and the margin evaporates during the ramp-up.
  • You commit to lead times you can meet. The difference between nameplate capacity and actual production stops being a surprise for Finance: a well-measured plant reaches 80 % of its nameplate capacity at 18 months, not 36. Public examples: AESC Douai (France, Renault) goes from 9 GWh in 2025 to 24 GWh in 2030 (according to Renault Group, press release 2024, and BBVA CIB 2023), and PowerCo Sagunto (Spain, Volkswagen) starts from 40 GWh/year of initial capacity, with preseries in September-October 2026 and commercial supply in July 2027 (according to Volkswagen Group, press release 2023, and La Tribuna de Automoción, October 2025).
  • You decide aging on economic criteria. Each day less in the chamber releases inventory and capacity, and cuts the detection of latent defects: it is a decision made with numbers, not with habit.
  • You prioritize investments with measured return. Automatic coil splicing, sensor redundancy, safety stock: each quick win enters the budget with its estimated euro and its expected effect.
  • You demand a dashboard that separates the phases. In assembly: actual cycle per cell and cells passing quality first time. In formation: cells per shift without alarm, actual average time versus target, rejection at the end of formation and kWh per cell. In aging and shipping: rejection in the chamber —2-5 % in a healthy plant and above 10 % indicates a process problem (indicative values)—, days of inventory in the chamber and customer defects —below 50 ppm in first-tier automotive (indicative value)—.

Warning sign: whoever promises big efficiency leaps in weeks, with no baseline and no subsequent measurement with the same yardstick, is selling you a headline; whoever shows you a before and after measured with data from your own machines is showing you your plant.

Let's talk about your plant

If you want to know what it costs you to manufacture each kWh and to commit to lead times with guarantees, start by measuring properly. Cronometras (app.cronometras.com) covers the time study of your manual tasks —with ILO and TAL allowances and synthetic performance rating— and gives you back the standard time per unit, exportable to PDF, Excel or JSON. And for the time inventory of the whole plant, WorkSamp (worksamp.com) does the work sampling with the value-added / non-value / obstruction taxonomy that this sector needs. If you also need industrial time and attendance control in the plant, Induly (induly.com) solves it with a mobile app and a kiosk by name and PIN.

More articles and resources for deciding with data in your plant: https://prodcont.com/


Don't want to do it yourself? We do it for you from €800

If you prefer to outsource it, Cronometras delivers a complete time and motion study from a single video: elemental breakdown, OIT/TAL fatigue allowances, standard time with tolerances, and a report in PDF + Excel + JSON ready to present to management.

👉 Request the study from €800 · More about Cronometras

Did you find this useful?

Share this article with your professional network