How many GWh from 10,000 tonnes of cathode?

The round number has a dated source. On 13 September 2021, Dangsheng Technology — a listed Chinese ternary-cathode producer — answered an investor-platform question: about 1,500–1,800 tonnes of ternary cathode per gigawatt-hour of ternary lithium cells, not split by 523, 622 or 811. 1,600 sits inside that band. Through September–November 2021, as lithium-carbonate prices spiked, Dangsheng, Dynanonic (a listed LFP-cathode producer) and other listed materials companies were asked the same question in public. Industry media kept the round number and dropped the chemistry.

A coincidence then made the figure look more universal than it was. On published parameters, NCM622 lands at about 1,600 t/GWh: a research compilation gives 1,616; another house estimate of 1.54 kg/kWh is about 1,540 t/GWh, which Lina could not verify online and treats only as a magnitude check. 2020–2022 was a 5-series and 6-series shipment market. The coefficient was checked, repeatedly, on the cells it described.

A circulating claim that 1,600 is a rounding between NCM523’s 1,638 t/GWh and generation-3 LFP’s 2,016 t/GWh does not survive the arithmetic. 1,600 is below 1,638, so it is not in that interval. Lina sets the claim aside. The chemistry-layer 1,638 and 2,016 figures are useful; they are not the origin of the folklore.

The hard evidence that the coefficient has travelled is in the mix. GGII, a Chinese battery-industry statistics house, put 2025 China cathode shipments at 5.025 million tonnes, up 50%. Lithium iron phosphate was about 3.89 million tonnes, 77.4% of the total and up 58%. Ternary was 830,000 tonnes, 16.5%. Inside ternary, 6-series still led; 8-series and 9-series lost share; 5-series fell quickly. The market that validated 1,600 t/GWh is not the market that now ships the powder.

Every tonnes-per-GWh figure in the note is cell-level cathode active material: the powder that stores the alkali ion, excluding conductive additive, binder and current collector. That is also the shipment basis in listed-company reports and in the industry statistics. Mixing a materials-level voltage into a cell-level mass will understate tonnes per GWh in a systematic way.

Voltage is full-cell working voltage against the anode the cell actually uses. For the lithium powders that is graphite: 3.7 V or 3.6 V for NCM, 3.2 V for LFP. For the sodium powders it is hard carbon: 3.0 V for NFPP, 3.0 V for NFM, 3.5 V for NFS. The table does not use the half-cell average potential versus lithium or sodium metal, which is higher. Half-cell reversible capacity is used for the engineering estimate; the gap between that number and the usable cathode capacity in a formed production cell is noted, and then left alone.

Tonnes of CAM per GWh = 1,000,000 ÷ (practical mAh/g × average volts). The product of those two terms is watt-hours per kilogram of active material. One gigawatt-hour is 10^9 watt-hours; divide by the specific energy and convert kilograms to tonnes. Inverted, gigawatt-hours from 10,000 tonnes = 0.01 × practical capacity × average voltage, before yield.

This is a first-order identity. Strictly, practical capacity and average voltage should come from the same full cell, on the same discharge window, rate and temperature. Public data do not provide that pair. The note therefore takes reversible capacity from material half-cells (the usual paper and supplier number) and voltage from the full-cell working voltage of the matching anode. The combination is an engineering estimate. Read it as a range. It is not a claim to single-digit precision.

NCM811, worked once. 200 mAh/g × 3.70 V = 740 Wh/kg. 10^9 Wh ÷ 740 Wh/kg = 1,351,000 kg, or 1,351 t/GWh, which is 7.4 GWh from 10,000 tonnes before any manufacturing yield. That 1,351 is the chemistry layer. It is not yet a factory number.

Practical specific capacity is reversible capacity, below the stoichiometric ceiling. Theoretical capacity is C = nF / (3.6 × M), with F = 96,485 C/mol. The note’s ceilings are 170 mAh/g for LFP, 278 for NCM523, 129 for NFPP, 240 for NFM and 121 for NFS. Practical capacity is that ceiling times a realised efficiency: about 73% for NCM811, about 91% for generation-3 LFP. First-cycle efficiency, state-of-charge window and rate all move the production-cell number relative to the half-cell. The chemistry layer takes the half-cell figure and does not deduct formation loss a second time.

Average voltage is the capacity-weighted integral of the discharge curve, ∫V(Q) dQ / Q, not the plateau voltage and not the median. On a flat LFP curve the three almost coincide. On a sloping sodium curve they do not, and the integral is the quantity that belongs in the identity.

The chemistry layer is an idealised cell. Between powder and a saleable cell sits coating, assembly, formation and grading, and the cells that are downgraded or scrapped. The note applies a yield of 96%, inside a 95–97% band, as a scenario. Tonnes per GWh are divided by that yield; gigawatt-hours from 10,000 tonnes are multiplied by it. It is not an industry standard, and the author asks the reader to replace it.

Public yield figures cannot be averaged into one utilisation rate. Their denominators differ. Tesla has discussed 4680 dry-electrode yield above 90%. Mature chemistries are often quoted above 98% on a comprehensive basis. After formation, the share of cells below A-grade is sometimes put below 10%. A 20 GWh plant improving yield by three percentage points is a different object again. None of these is a measurement of CAM mass that survives into a GWh of graded cells, so none of them sets the 96% scenario.

Formation and grading are the last screen before a cell is warehoused. Lina notes that they account for about 15% of lithium-battery equipment investment and about 40% of line energy, which is why end-of-line rejects are easy to undercount. Those rejects sit in the yield layer. They are not a second deduction from practical capacity.

Each percentage point of yield moves tonnes per GWh by about 1%. At 95, 96, 97 and 98%, the chemistry-layer figure is scaled by 1.0526, 1.0417, 1.0309 and 1.0204 — that is, divided by the yield.

On the same 10,000 tonnes, 9-series NCM at 7.3 GWh is about 2.6 times NFPP at 2.8 GWh. Phosphate cathodes are systematically heavier per gigawatt-hour than ternary: generation-3 LFP at 2,100 t/GWh, NFPP at 3,543. That is voltage and capacity together, not a process gap between plants.

Generation-3 and generation-4 LFP almost coincide (4.8 and 4.9 GWh). The LFP “generation” story in the Chinese cathode market is largely compaction density, not a 30% jump in CAM-mass efficiency. High-nickel NCM moves the other way, which is why 1,600 t/GWh is already high for 811 and 9-series.

The 9-series voltage is taken conservatively at 3.60 V. Average voltage on Ni90 falls slightly as nickel rises; the literature range is 3.6–3.8 V, and ±0.1 V is about ±3% on the result. Practical capacity of 210 mAh/g is an estimate from an NCM90-Zr2 paper (215.5 mAh/g reversible at a 4.5 V cut-off), stepped down to a production window.

NFS is sodium iron sulfate, Na2Fe2(SO4)3, the alluaudite structure first reported by Yamada in 2014. Full-cell voltage against hard carbon is about 3.5 V. The feedstocks are ferrous sulfate and sodium sulfate — chemical by-products — which is the point of the route: iron-based polyanion, high voltage, low materials cost. NFM is O3 NaNi1/3Fe1/3Mn1/3O2. Full-cell voltage is an engineering 3.0 V (2.9–3.1): Sigma-Aldrich’s material-level 3.1 V less about 0.1 V for the hard-carbon anode. An MTI NFMO-plus-hard-carbon 7 Ah pouch, window 1.5–4.0 V, is cited as corroboration. NFPP’s 3.0 V is a full-cell working voltage at 0.2C in a 2.0–4.0 V, 100% depth-of-discharge window, from trade press rather than a paper.

None of that chemistry is the conversion. Phosphate powders are heavier per gigawatt-hour than the nickel-manganese-cobalt market that produced 1,600 t/GWh, and sodium powders are heavier still. Naming the powder is the conversion. Averaging them is how the folklore survives.

The kilowatt-hours on a vehicle window sticker are usually larger than the energy a driver is allowed to use, not smaller. In the sample Lina takes — Ford Mustang Mach-E, GM Bolt, Tesla — usable energy is 88.5–98.5% of nominal. Mach-E is the transparent case: 68 kWh usable on a 76.8 kWh pack, a 10–11% buffer, and Ford publishes the usable number. Bolt is 65 / 66 kWh, a 1.5% buffer. Tesla is about 4.5%. Pack-level usable energy is a depth-of-discharge window. It does not change how much cathode powder went into a gigawatt-hour of cells.

The N/P ratio — anode capacity over cathode capacity, typically 1.05–1.2 — is not a multiplier on this identity. It is set by anode thickness at a fixed anode compaction. Kim et al. (Electrochimica Acta, 2015) hold compaction density and move thickness from 124 to 135 to 146 micrometres to obtain N/P of 1.10, 1.20 and 1.30; the cathode side does not move. In a first-order model that already fixes usable cathode capacity, N/P is not an independent factor. In a real cell it still matters, indirectly, through first-cycle efficiency and the state-of-charge window.

Listed-company replies are treated as an order-of-magnitude check on the two-layer identity, not as a calibration. None of the disclosures states whether manufacturing loss sits inside the tonnes-per-GWh number. Five such figures sit in the same band as the back-calculations. Numerical differences of 0.6–4.8% are inside the estimate’s claimed precision and are not treated as a test of it.

Dangsheng’s 1,500–1,800 t/GWh band covers the yield-adjusted 523 and 622 back-calculations (1,706 and about 1,683). The 622 back-calculation is 172 mAh/g × 3.6 V, 1,616 t/GWh before yield, 1,616 ÷ 0.96 ≈ 1,683 after, and is the same identity as the 1,616 compilation figure. Dynanonic’s LFP range of 2,200–2,500 t/GWh is 4.8% above generation-3 at the floor and 19% above at the ceiling. Lina reads the ceiling as a possible allowance for specific-capacity shrinkage. Xiangtan Electrochemical’s LFP reply is compared with the uncorrected 2,016 t/GWh and then set aside: Xiangtan’s main business is graphite electrodes, not LFP, and the LFP name in that group is the associate Hunan Yuneng, a large LFP-cathode supplier.

No listed-company reply in the set uses a single 1,600. The two-layer identity and the official-order figures are the same kind of object: cell-level cathode consumption per gigawatt-hour of cells, with an unstated attitude to yield.