Battery Gigafactory Transformer: How to Size the 132 kV Supply for a 10 GWh Cell Plant, Why Dry Room HVAC and VFD Harmonics Set the kVA, and When Dry-Type Beats Oil

The answer first: a 10 GWh cell plant is a 59,462 kW load, and three 40 MVA 132/33 kV units carry it more safely than two 63 MVA units for the same money

A 10 GWh/year lithium-ion cell plant is not sized by its coating lines. It is sized by its dry room. Our reference plant makes 11.16 million 3.2 V / 280 Ah prismatic cells a year (33,516 cells a day, 1,396 an hour), has 86,800 kW of connected load and reaches a design peak of 59,462 kW after area demand factors and a 0.96 plant diversity. At 0.894 power factor that is 66,510 kVA, and 61,899 kVA at 0.96 once 12.6 Mvar of detuned capacitance is added, which is 1,083 A at 33 kV. Three 40 MVA 132/33 kV YNd11 units at 12% impedance carry it: 51.6% each in parallel, 77.4% on two after one unit is taken out for maintenance. But the kVA is the easy part. What actually decides whether this plant runs is the 18,920 kW dry room (31.8% of the peak, and the only load that cannot be shed), the 7.0% total demand distortion that sits one point under the IEEE 519 limit before filtering, and a 100 ms voltage sag on the 132 kV line that scraps USD 38,000 of coated electrode every time it happens.

Key takeaways

• 10 GWh/y = 11.16 million 896 Wh cells, 33,516 a day. At 38 kWh of electricity per kWh of cell capacity (industry experience for LFP prismatic lines) the plant draws 381 GWh a year over 8,000 operating hours: 47,570 kW average, 59,462 kW peak.
• Three 40 MVA 132/33 kV YNd11 at 12% impedance: 51.6% each, 77.4% with one unit out. Two 63 MVA units sit at 49.1% each but reach 98.3% with one out — technically inside the nameplate and outside any sane operating margin at 40 °C ambient.
• The dry room is 18,920 kW of the 59,462 kW peak (31.8%). Formation and aging — the load everyone worries about — is 3,120 kW (5.2%).
• VFDs, not rectifiers, set the harmonic filter: 16,900 kW of 6-pulse drives produces 7.0% TDD against an 8.0% IEEE 519 limit at Isc/IL = 19.9, so 4 × 3.15 Mvar detuned banks do power-factor correction and 5th-harmonic absorption in one purchase and take TDD to 2.2%.
• The electrical package costs USD 6.6–8.1 million (USD 111–136 per kW of peak, USD 0.66–0.81 million per GWh of annual capacity), and the 40 MVA transformers plus the 132 kV GIS are a 24–30 week critical path (industry experience, FOB China, 2025–2026).

What does a 10 GWh battery cell plant actually draw? The load schedule behind 61,899 kVA

Start from the product, not from a watts-per-square-metre benchmark, and convert the schedule the way the one-step sizing and budget formula does. A 3.2 V / 280 Ah prismatic LFP cell holds 896 Wh, so 10 GWh of annual output is 11.16 million cells, or 33,516 cells a day on 333 operating days. Cell manufacturing consumes 30–45 kWh of electricity per kWh of cell capacity depending on chemistry, dry-room area and how much formation energy is recovered — 38 kWh is a realistic mid-point for a new LFP plant with regenerative formation (industry experience). That puts annual consumption at 381 GWh. Spread over 8,000 hours it is 47,570 kW average, which is 80% of the 59,462 kW design peak: a cell plant runs flat, three shifts, seven days, and the transformer sees a load factor that most industrial plants never reach — the same flat profile that makes data centre transformers so sensitive to no-load loss.

Two rows in that table decide everything downstream. The first is the dry room: 22,000 kW connected and 18,920 kW demanded, 31.8% of the plant peak, and it is the one load that cannot be interrupted without scrapping material. The second is coating: 26,400 kW connected but 19,008 kW demanded, because the dryer fans, the NMP recovery unit and the mixing agitators are continuous while the calenders and the slitters batch. Note what is not in the top two: formation and aging, the DC rectifier area that dominates most conference talks about gigafactory power quality, is 3,120 kW — 5.2% of the peak. It matters for harmonics, not for kVA.

The 0.894 power factor is the other number worth pausing on. It comes from 12 MW of chiller and dryer-fan induction motors at 0.88, coating drives at 0.94 and formation rectifiers at 0.80, and it turns a 59,462 kW load into 66,510 kVA — a 10.7% kVA penalty paid on every transformer, cable and switchgear item on the site. Correcting to 0.96 removes 4,611 kVA from the design, which is more than one 2,500 kVA unit substation, or roughly USD 1.1 million of installed 33 kV infrastructure. The correction is not a convenience; it is one of the cheapest items in the package, and it doubles as the harmonic filter, as the next sections show.

Why three 40 MVA transformers beat two 63 MVA units — and why two 50 MVA units fail

The plant needs 61,899 kVA at 0.96 power factor, 66,510 kVA uncorrected if you decide to filter later. The redundancy rule for a plant that loses USD 38,000 of product per voltage event is N+1 at the level where redundancy is cheapest, which for a 60 MW load is the 132 kV incomer, not the 0.4 kV board. Run the options on that basis and the answer is not the one the capex spreadsheet suggests at first glance.

Two 63 MVA units look efficient because 126 MVA for a 61,899 kVA load is 49% average loading, and the price of two 63 MVA 132 kV units (USD 750,000–875,000 each, FOB China) is within 6% of three 40 MVA units (USD 520,000–610,000 each). The trap is the outage case: with one unit out, the survivor carries the whole plant at 98.3% of nameplate, which is permissible only at 20 °C ambient with a perfect cooling system. At 40 °C ambient, IEC 60076-7 loading guidance and the loss-of-life curve make sustained operation above roughly 95% a decision that spends insulation life, and a plant that runs 47,570 kW average has no load to shed except the dry room it cannot touch. Three 40 MVA units give 77.4% in the same outage, which leaves room for a hot day, a line start and the 8% overload the drives pull during a formation batch. The three units run in parallel permanently, so the impedance and ratio matching rules in the parallel operation guide apply: 12% ±7.5% and a ratio tolerance of 0.25% are what keep circulating current below 3%.

There are three secondary benefits. Short-circuit duty stays manageable: a 40 MVA unit at 12% impedance contributes 5,830 A at 33 kV, so three in parallel plus 4,000 A of motor contribution put the 33 kV bus at about 21.5 kA — comfortably inside 31.5 kA switchgear, where two 63 MVA units at 9,190 A each would have forced a 40 kA board. The method behind those amps is in the three-step short-circuit calculation. Energisation is staged: two units carry the first two electrode lines and the dry room during commissioning while the third arrives with line three, so the utility connection is not blocked by a single 30-week delivery. And the spare-parts inventory is one HV bushing set and one OLTC motor drive for three identical units instead of two.

What about the 132 kV incomer itself? A 60 MW load can be served at 33 kV, and many plants are. It should not be here: 33 kV at 1,083 A needs four parallel 630 mm² copper cables per phase, and a 2 km feeder at that current burns 382 kW in conductor loss. The same power at 132 kV is 291 A in one 630 mm² circuit and 24 kW of loss. The 358 kW difference is 2,864 MWh a year, about USD 200,000 at USD 0.07/kWh, which pays for the 132 kV GIS over the life of the plant and removes a 33 kV cable corridor from the site plan.

Why does the dry room, not the coating line, decide the transformer loading?

A lithium-ion dry room is held at 23 ± 2 °C and below 1% relative humidity, which means a supply-air dew point near −40 °C. In our reference plant that is 30,000 m² of dry room, served by 14 supply-air handling units of 400–630 kW each. The 22,000 kW connected load splits into 9,400 kW of chillers, 8,600 kW of desiccant-wheel regeneration and 4,000 kW of supply and exhaust fans. The regeneration heaters are resistive and run at full output year-round because the wheel has to be dried regardless of outdoor conditions; the chillers carry both the sensible load and the latent load that leaks through the airlocks. This is why the plant's load does not fall in winter the way a normal factory's does: the dry room goes from 18,920 kW in July to about 14,800 kW in January, a 22% swing, not the 45% that the HVAC nameplate suggests.

Electrically, that produces three requirements that a generic industrial substation design misses. First, the dry room cannot be shed, so it gets its own 0.4 kV substations fed from two different 33 kV bus sections — eight of the 30 unit substations serve dry room air handling and chillers, and each pair is split across two 40 MVA transformers so a transformer outage does not stop dehumidification. Second, the air-handling units are the largest single motor group on the site and they are all on VFDs: 14 units at 400–630 kW, 6,300 kW connected, ramping with dew-point control rather than on/off, which is good for the process and bad for harmonics (Section 5). Third, the transformers themselves must live outside the dew-point envelope. A cast-resin unit dissipates 20–25 kW at 88% load; thirty of them inside the dry room would add 570 kW of heat that the dehumidification system has to remove, and the dehumidification system is already 31.8% of the plant load. Put them in an adjacent electrical corridor at 25 °C with pressurised, filtered air, and specify anti-condensation heaters because a 20 °C transformer surface meeting 25 °C / 45% RH corridor air will sweat.

How do VFD and rectifier harmonics set the filter, and why is TDD the only number that matters?

Non-linear load on this plant is 16,900 kW: 4,800 kW of coating dryer and exhaust fans, 5,600 kW of chiller compressors, 2,180 kW of compressed air, 1,200 kW of NMP recovery and 3,120 kW of formation rectifiers, all 6-pulse front ends. At 33 kV that is 311 A of fundamental current on a bus whose maximum demand current is 1,083 A. Apply a typical 6-pulse spectrum — the spectra behind the transformer harmonics guide — and the current distortion at the point of common coupling looks like this.

The bus short-circuit current is 21.5 kA, so Isc/IL = 19.9, which lands in the 20 < Isc/IL < 50 band of IEEE 519-2022 Table 2 for 10–69 kV systems: 8.0% TDD, 7.0% fifth, 5.0% seventh. Every individual harmonic passes and the total passes by one percentage point. One point is not a design margin, and it says nothing about the extra winding loss the fifth and seventh harmonics cause inside the transformers themselves, which is what a K-factor rating covers — it is the difference between a compliant plant and a utility penalty letter once the plant adds a fifth coating line, and it leaves no room for the resonance that an uncorrected capacitor bank would create. The design answer is 4 × 3.15 Mvar capacitor banks with 4.7% detuned reactors, tuned to 189 Hz, which absorb the fifth and seventh and supply 12.6 Mvar of correction at the same time. Fifth-harmonic current falls from 68 A to 24 A and TDD from 7.0% to 2.2%.

The detuning is not optional. A plain 12.6 Mvar bank on a 33 kV bus with 17.5 kA of transformer fault contribution resonates near the 4.7th harmonic, close enough to the 5th to amplify it by a factor of three or four — the classic outcome in which adding power-factor correction makes the harmonic problem worse. The 4.7% reactor pushes the resonant frequency below the 5th and turns the bank into a filter. The package costs USD 420,000–520,000 as a single purchase; 13 Mvar of plain capacitors plus a 300 A active filter costs about USD 560,000 and takes more floor space in the 33 kV room.

Use TDD, not THD, in every specification and every acceptance test. THD is a percentage of the fundamental current at the instant of measurement, so a plant running at 40% load shows a higher THD than the same plant at full load while injecting fewer amps. IEEE 519 limits are expressed against IL, the maximum demand load current, which is why a plant can measure 15% THD at light load and still comply at 6% TDD. Specify Class A power-quality meters per IEC 61000-4-30 at the 132 kV PCC and at each 33 kV bus section, log 10-minute values, and judge compliance on the 99th percentile over 12 months — the same instrument and the same statistic the utility will use when it audits you.

Why is a 100 ms voltage sag more expensive than an oversized transformer?

The coating line runs at 80 m/min and the distance from the coating head to the dryer exit is 1,800 m, so a drive trip leaves 1,800 m of wet electrode — about 9,000 m² — that cannot be recovered. At a cell cost of USD 38,000 for that length and a 45-minute restart, one event costs about USD 60,000 including lost output. A 132 kV line 20 km away clearing a three-phase fault in 100 ms leaves 32% residual voltage at the 33 kV bus. The coating drives trip below 85% for 50 ms. Three events a year is USD 180,000.

The transformer cannot fix this, and it is worth being explicit about why, because buyers often ask for a “stiffer” unit expecting sag immunity. A lower-impedance transformer passes a deeper dip to the 0.4 kV board; a higher-impedance transformer limits fault current but not the dip depth. Voltage sag is a source-side event transmitted through the transformer almost unchanged. What works is a ladder of measures, cheapest first: enable DC-bus ride-through and kinetic buffering on the 12 coating drives (free, covers a 0.5 s dip to 70% and about 60% of events); add a 2.5 MW / 1.5 s flywheel on the two lines that coat the high-value NMC electrode and a 1.5 MVA rotary UPS on the mixing and metrology lab (USD 610,000, 3.4-year payback at USD 180,000 a year); and keep a 33 kV fast transfer to the second 132 kV source, which at 100 ms is too slow for the drives but preserves the dry room and the rest of the plant.

Where does dry-type beat oil inside a gigafactory — and where does it not?

Thirty 2,500 kVA 33/0.4 kV cast-resin dry-type units sit inside the production buildings, and that is the right choice for the reasons set out in the oil-immersed versus dry-type comparison: no oil means no bund, no oil-water separator, no deluge system and no fire-rated separation to the process area. A cast-resin unit at F class insulation and 100 K temperature rise carries its rating at 40 °C ambient and reaches 125% with forced air, which is exactly the contingency a plant with N+1 at 132 kV needs at the 0.4 kV level. Putting oil-filled unit substations indoors instead would add a 4 h fire wall, a 110% bund and a deluge system per room; across 30 substations the civil and fire-protection work alone is about USD 900,000, plus a 10–15% insurance premium.

The main transformers stay oil-filled and outdoors. A 40 MVA 132 kV cast-resin unit is a special order at 2.3–2.8 times the price of an oil-immersed unit, needs a much larger footprint and loses the OLTC range that 132 kV networks require. Specify ONAN/ONAF, 40 °C ambient, 65 K rise — the derating arithmetic is in the cooling methods and temperature rise guide — C4-M paint at 240 µm and a stainless-steel nameplate, and give it a 2 h fire wall at 7.5 m from the building and a 110% oil containment pit.

The one place where the choice is not obvious is the NMP recovery area. N-Methyl-2-pyrrolidone is a combustible solvent with a flash point around 91 °C, and its recovery room is normally classified Zone 2 under IEC 60079-10-1. No transformer of either type goes inside a Zone 2 envelope. Put the serving substation 6 m outside with a pressurised electrical room, and treat the recovery area's motors and instrumentation as Ex-rated — a decision that affects the cable routing far more than the transformer price.

What must the 132/33 kV specification say? The line-by-line datasheet

Most gigafactory transformer disputes trace back to a datasheet that specified the rating but not the duty. These are the fields that must be filled in, and the values that follow from the load schedule above.

Before you sign the datasheet, run it against the pre-shipment inspection checklist: every field above has to be visibly proved at the factory, not assumed. The 12% impedance deserves a note because it is often set by habit at 10.5%. At 10.5% the same unit contributes 6,660 A at 33 kV, three in parallel plus motors reach 24 kA, and the 33 kV switchgear has to move from 31.5 kA to 40 kA — a step that costs more than the extra copper in the transformer. Going the other way, to 14%, reduces fault duty but raises voltage regulation and load loss; at 0.516 loading the difference in load loss between 12% and 14% is about 12 kW per unit, or USD 20,000 a year across three units. Impedance also sets the regulation you will live with, which is why the voltage regulation calculation belongs in the same decision. 12% is the point where the switchgear stays at 31.5 kA and the annual loss penalty is still small.

What does the electrical package cost, and what is the lead time?

Prices below are FOB China for 2025–2026 delivery and exclude civil works, installation and the utility connection charge. They are ranges, because a 132 kV GIS with 5 bays and IEC 61850 protection is not a commodity.

That is USD 111–136 per kW of design peak and USD 0.66–0.81 million per GWh of annual cell capacity — about 15–18% of the USD 40–55 million that a 10 GWh cell plant costs to build (industry experience). Transformer losses are a small but real line item: three 40 MVA units at 52% average loading plus 30 dry-type units at 88% add about 680 kW of loss, 5,440 MWh a year and USD 380,000 at USD 0.07/kWh, roughly 1.1% of the plant's electricity bill. Buying GB 20052 Grade 2 instead of Grade 3 dry-type units — the grading logic is compared across markets in the energy efficiency standards comparison — recovers 27 kW of that, about USD 15,000 a year, for a USD 90,000 premium across 30 units — a six-year payback that only makes sense if you also value the reduced cooling load.

The transformer and the GIS are the critical path, and they are not compressible by paying for air freight: a 40 MVA 132 kV unit needs 24–30 weeks including core stacking, tanking, oil processing and the routine tests that must be witnessed. If the building handover is in month 14, the purchase order has to be placed in month 6, before the steel is erected — and the factory you place it with should already have passed the 12-point factory audit rather than a video call. The full breakdown of what drives those weeks is in the delivery lead time guide. This is the single most common schedule failure in gigafactory projects, and it is a procurement problem, not an engineering one.

How do you size a gigafactory supply in six steps?

1. Build the load schedule from the process. kW per coating line, kW per air-handling unit, kW per formation rack. Never start from W/m² (the power system calculators will do the arithmetic once the schedule is right): a gigafactory's energy intensity varies by a factor of two between an LFP plant with a small dry room and an NMC plant with a large one.
2. Apply demand factors per area, then plant diversity. Here: 86,800 kW connected, 61,940 kW after area factors, 59,462 kW after 0.96 diversity. Connected load is 46% higher than the peak; sizing on it wastes USD 2–3 million.
3. Convert at the real power factor and correct it. 59,462 kW at 0.894 is 66,510 kVA; at 0.96 it is 61,899 kVA. Correction costs USD 420,000–520,000 and removes 4,611 kVA of transformer, cable and switchgear.
4. Choose the redundancy level, then the unit count. N+1 at 132 kV with three 40 MVA units beats two 63 MVA units on outage margin and short-circuit duty at essentially the same price.
5. Check fault level, motor starting and sag ride-through. 21.5 kA at 33 kV sets a 31.5 kA board; 630 kW air-handling units on VFDs start at 110% current; the coating drives need 0.5 s of ride-through to survive a 132 kV fault clearance, and the energisation sequence itself should follow the eight commissioning tests before the first cell is coated.
6. Verify harmonic compliance at the PCC, in TDD, over 12 months. 7.0% before filtering against an 8.0% limit is not compliance, it is luck. Filter to under half the limit and monitor with Class A meters.

What are the six specification mistakes that cost the most?

• Sizing on connected load. 86,800 kW instead of 59,462 kW means five 40 MVA transformers, a larger GIS and a 40 kA board — about USD 2.5 million of capacity the plant never draws.
• Buying two large units instead of three smaller ones. Cheaper on the quotation, but 98.3% loading with one unit out is not a redundancy scheme, it is a hope.
• Specifying THD instead of TDD. A 5% THD clause is satisfied at full load and violated at 40% load by the same plant. The utility measures TDD against IL.
• Forgetting the 40 °C ambient. A 40 MVA unit rated at 40 °C is not the same machine as one rated at 20 °C; the difference is about 8% of capacity, which is exactly the margin the N+1 calculation was relying on.
• Putting the unit substations inside the dry room. 570 kW of extra dehumidification load, condensing surfaces and a maintenance access problem, in exchange for saving a corridor.
• Ordering the transformers after the building. 24–30 weeks is not a negotiation position. The transformer order date sets the energisation date.

What changes as the plant expands from 10 to 30 GWh?

Almost nothing at 132 kV and a great deal at 33 kV. Three 40 MVA units are already the N+1 configuration for 30 GWh at the same 38 kWh/kWh intensity — the load triples to about 178 MW, which needs six units of 40 MVA, so the practical expansion is to add three more identical units on the same GIS line-up rather than to buy a larger frame size. What must be sized for the end state on day one is the 33 kV switchgear room: 26 bays at 10 GWh become 60–64 bays at 30 GWh, and a switchgear room built for 26 cannot be extended without a shutdown. Specify the building, the cable trench and the busbar rating (1,250 A) for 30 GWh, install the bays for 10 GWh, and leave 20% spare bays energised with test sockets.

The second end-state decision is the 0.4 kV architecture, and it is the same unit-substation logic that governs a BESS step-up installation at a smaller scale. Thirty unit substations at 10 GWh become 80–90 at 30 GWh, and the temptation is to raise the unit rating to 3,150 kVA to keep the count down. Resist it: a 3,150 kVA unit at 6% impedance puts 75 kA on the 0.4 kV board, which forces 100 kA busway and higher-rated breakers across the whole plant. Keeping 2,500 kVA units and adding count keeps the 0.4 kV fault level at 60 kA and the component range standard.

Sources / 资料来源

Source: IEC 60076-1:2011 — Power transformers: General.
Source: IEC 60076-2:2011 — Temperature rise for liquid-immersed transformers.
Source: IEC 60076-3:2013 — Insulation levels, dielectric tests and external clearances in air.
Source: IEC 60076-5:2006 — Ability to withstand short circuit.
Source: IEC 60076-7:2018 — Loading guide for mineral-oil-immersed power transformers.
Source: IEC 60076-11:2018 — Dry-type transformers.
Source: IEC 62271-200:2021 — AC metal-enclosed switchgear and controlgear for rated voltages above 1 kV and up to and including 52 kV.
Source: IEC 61000-4-30:2015 — Testing and measurement techniques: power quality measurement methods, Class A.
Source: IEC 61000-3-6:2008 — Assessment of emission limits for distorting loads in MV and HV power systems.
Source: IEEE 519-2022 — Recommended practice and requirements for harmonic control in electric power systems, Table 2 current distortion limits for 120 V–69 kV.
Source: IEEE C57.12.00-2021 and C57.12.90-2021 — General requirements and test code for liquid-immersed transformers.
Source: GB 20052-2020 — Minimum allowable values of energy efficiency and energy efficiency grades for three-phase distribution transformers.
Source: GB/T 1094.11-2022 — Power transformers: dry-type transformers.
Source: IEC 60079-10-1:2020 — Classification of areas: explosive gas atmospheres.
Source: ISO 12944-2:2017 and ISO 12944-5:2019 — Corrosion protection of steel structures, corrosivity categories and paint systems.
Source: Industry experience — lithium-ion cell plant electricity consumption of 30–45 kWh per kWh of cell capacity, dry room electricity at 25–35% of plant load, and 55–70% formation energy recovery on regenerative formation, 2019–2026.
Source: Industry experience — 6-pulse VFD harmonic spectra of 20–25% fifth and 8–10% seventh at the drive terminals, and measured TDD of 6–9% on unfiltered gigafactory 33 kV buses, 2019–2026.
Source: Industry experience — transformer, GIS, switchgear and filter price ranges and 24–30 week lead times for 132 kV class equipment, FOB China, 2025–2026.
Source: Industry experience — industrial electricity tariffs of USD 0.06–0.09 per kWh used in the loss and sag calculations.
Source: QDTB Engineering Team — load schedules, harmonic studies and 132/33 kV transformer specifications for battery cell, module and pack plants in Southeast Asia, Europe and North America, 2021–2026.