Water Electrolysers: The 1.23 V Floor, the 0.5 g/kW Iridium Tax, and the ₹306/kg Behind India's 5 MMT Green Hydrogen Bet

Engineering guide to PEM and alkaline water electrolysers: thermodynamics, Faraday sizing, overpotential budget, stack BOM and manufacturing, iridium supply limits, and a ₹/kg LCOH model for India.

·

Water Electrolysers: The 1.23 V Floor, the 0.5 g/kW Iridium Tax, and the ₹306/kg Behind India's 5 MMT Green Hydrogen Bet

India has awarded 3,000 MW per year of electrolyser manufacturing capacity to fifteen companies and 7,56,100 tonnes per year of green hydrogen production to sixteen more, against a 2030 target of 5 million tonnes. The incentive ladder runs ₹4,440/kW in year one down to ₹1,480/kW in year five for the machines, and ₹50/kg, ₹40/kg, ₹30/kg for the molecules. As of the August 2026 parliamentary answer on the programme, no SIGHT-supported electrolyser plant had been fully commissioned — against scheduled commissioning dates of 9–10 August 2026 for the entire 1,500 MW Tranche I.

That is not a construction failure. It is what happens when a subsidy meets a machine that is thermodynamically boxed in and metallurgically awkward. Of the 1,500 MW awarded in Tranche I, exactly 137 MW is PEM — Ohmium's Bengaluru line — and the remaining five awardees (Reliance, L&T, John Cockerill Greenko, Adani, Matrix) all bid alkaline. The technology split was decided by iridium availability and cell cost, not by efficiency preference.

This guide works the engineering behind those numbers, because the electrolyser is the most fabrication-intensive energy machine currently being built at scale. A 1 MW stack is 250 stamped plates, 250 coated membranes, 500 pieces of sintered titanium powder, 502 gaskets, two machined end plates and twelve tie rods — held to a helium leak rate and a flatness spec, bolted to a 2,000 A rectifier.

1.23 volts is not a target, it is a wall

Water splitting is a two-electron reaction per molecule of hydrogen:

The Gibbs energy of formation of liquid water is 237.2 kJ/mol at 25 °C, so the reversible cell voltage is fixed:

Below that voltage, nothing happens. Not slowly — nothing. The current is an exponential function of overpotential, so at 1.229 V the net reaction rate is zero, and at 1.20 V it is negative (a fuel cell). Every commercial electrolyser in the world operates between 1.7 V and 2.1 V per cell, and every improvement programme is an attempt to move that number down by tenths of a volt.

The enthalpy of the reaction is 285.83 kJ/mol, which gives the thermoneutral voltage:

At exactly V_{tn} the cell is adiabatic — the electrical input equals the enthalpy of the products, and no heat needs to be supplied or removed. Below V_{tn} the cell absorbs heat from its surroundings; above it, the cell generates waste heat that the cooling loop must reject. A 1 MW stack at 2.0 V dissipates roughly 350 kW as heat, which is why electrolyser balance-of-plant is mostly a water and heat management problem.

Two efficiency conventions collide here, and confusing them is the most common arithmetic error in hydrogen writing. On a higher heating value basis (39.4 kWh/kg), efficiency is 1.481/V_{cell}. On a lower heating value basis (33.3 kWh/kg) it is 1.253/V_{cell}. The LHV figure is what the industry quotes, so a cell at 2.0 V is 62.7% efficient, not 74%.

Because the two conventions cancel, specific energy consumption collapses to a single linear expression that is worth memorising:

At 1.9 V, 50.5 kWh/kg. At 1.8 V, 47.9. At 1.6 V, 42.6. Every 100 mV of cell voltage is 2.66 kWh/kg, and at ₹3/kWh that is ₹8/kg of hydrogen — which is why the SIGHT bid criterion scores specific energy consumption, not capex, as the primary technical metric.

The reversible voltage drifts with temperature because the entropy term does:

Running at 80 °C rather than 25 °C drops V_{rev} from 1.229 V to about 1.18 V and V_{tn} from 1.481 V to 1.44 V. That 50 mV is free efficiency, and it is the entire reason PEM stacks run hot. It is also why the membrane's glass transition temperature — not the catalyst — sets the upper bound on operating temperature.

What a kilogram of hydrogen costs in electrons

Faraday's law sets the production rate. Two electrons per molecule means 192,970 coulombs per mole of H₂, and one mole of ideal gas is 22.414 normal litres:

One kiloamp produces 0.4182 Nm³/h, or 37.6 g/h, of hydrogen. That single constant sizes every stack, rectifier and gas dryer in the plant.

Take a 1 MW PEM stack: 250 cells, 1,000 cm² of active area each, 2.0 A/cm², 2.0 V per cell.

Check against specific energy: 1,000 kWh ÷ 18.8 kg = 53.2 kWh/kg, and 26.6 \times 2.0 = 53.2. The two routes agree because they are the same physics.

The 18.8 kg/h figure is also the reason electrolysers are quoted in MW and not in tonnes per day. One tonne per day of hydrogen requires 1000/24/18.8 \approx 2.2 MW of stack at 2.0 V, and 2.5 MW once rectifier and BoP losses are counted. India's 5 MMT target is therefore 5,000,000 ÷ 365 ÷ 18.8 ≈ 30,400 MW of continuous stack power, or — at a realistic 60% capacity factor for solar-dominated supply — about 50 GW of installed electrolyser.

Against 3,000 MW/year of awarded domestic manufacturing capacity, that is more than sixteen years of output. Even if every rupee of Tranche I and Tranche II capacity were commissioned on schedule and never exported, 2030 lands at roughly 13–15 GW. The manufacturing award is not the constraint; the ₹/kg is.

Where the missing 0.6 volts goes

A cell at 2.0 V is spending 0.77 V above the reversible potential. The budget decomposes as:

The activation overpotentials follow Tafel behaviour, \eta = b \ln(i/i_0), with

At 80 °C, RT/F = 30.4 mV. With a transfer coefficient of 0.5–1.0 that puts b between 70 and 140 mV/decade. Measured Tafel slopes for IrO₂ anodes in PEM cells land at 50–70 mV/decade, which is close to the thermodynamic best case — the oxygen evolution reaction is slow because it is a four-electron, four-proton process with an adsorbed intermediate bottleneck, not because the catalyst is badly made. The cathode hydrogen evolution reaction is fast by comparison and contributes 30–50 mV.

Ohmic loss dominates. The membrane's area-specific resistance is thickness over conductivity:

A 50 µm Nafion membrane at 0.15 S/cm gives 0.033 Ω·cm², or 67 mV at 2 A/cm². Add the catalyst layers, the porous transport layer and the plate's interfacial contact resistance and the real ASR is 0.12–0.18 Ω·cm², which at 2 A/cm² is 240–360 mV. Two-thirds of the inefficiency is a resistance problem, not a chemistry problem: thinner membranes, better catalyst layers, lower contact resistance, tighter plate tolerances.

That is the origin of the DOE performance ladder for PEM stacks — 2 A/cm² at 1.9 V today, 3 A/cm² at 1.8 V by 2026, 3 A/cm² at 1.6 V at the limit, and 51 → 48 → 43 kWh/kg along with it. The entire ten-point gain is 300 mV, and it will be won in manufacturing tolerance and materials, not in a new catalyst.

Two secondary losses matter at the plant level. Gas crossover puts 0.1–0.5% hydrogen into the oxygen stream; the flammability limit is 4% in oxygen, so stacks alarm at 2% and trip at 2.5%, which caps how thin the membrane can get. And at high current density, oxygen bubbles cover part of the anode, adding a mass-transport term that shows up as a rising slope above 2 A/cm² — the practical reason alkaline cells stop at 0.5 A/cm² while PEM runs at 2.

The stack, part by part, and which parts a job shop can quote

NREL's bottom-up manufacturing model is the most useful public cost breakdown, and it says something unfashionable: at volume, stack cost is dominated by materials and by yield, not by labour. For a 1 MW stack the catalyst-coated membrane is 36–47% of stack cost, the porous transport layer 17–25%, bipolar plates 12–21%, and end plates plus assembly 3–13%. Stack cost falls from about USD 237/kW at 10 MW/year of production to about USD 101/kW at 1,000 MW/year.

Catalyst-coated membrane (CCM). A 50–175 µm proton-exchange membrane — Nafion 115, 117 or a 20–50 µm reinforced grade — coated on the anode with iridium oxide at 1.6–2.0 mg Ir/cm² and on the cathode with 40 wt% Pt/C at 0.2–0.4 mg Pt/cm². Coating is slot-die or ultrasonic spray, then hot-pressed or decal-transferred. The membrane is 30–50% of CCM cost; the platinum-group metals are 13–46%. A 1 MW stack at 1,000 cm² per cell carries 250 × 1,000 cm² = 25 m² of membrane — ₹2–5 lakh of polymer before a single gram of iridium.

Porous transport layer (PTL). Sintered titanium powder, 0.2–0.5 mm thick, 30–50% porosity, 5–30 µm pore size, sintered at 1,100–1,300 °C in vacuum. On the anode it must survive a potential of 1.8–2.0 V in acid at 80 °C, which is one of the most corrosive environments in industrial chemistry; uncoated titanium grows an oxide film that pushes interfacial contact resistance from 4–5 mΩ·cm² to tens of mΩ·cm² within thousands of hours. The standard fix is 50–200 nm of gold or platinum by PVD, and the cheaper one is a titanium suboxide (Ti₄O₇) coating, which reaches 4–5 mΩ·cm² without precious metal.

Bipolar plates. 0.1–0.15 mm titanium (PEM anode side) or 316L stainless steel (alkaline, cathode side), stamped with a flow field of 0.3–0.5 mm channel depth on a 0.5–1.0 mm land pitch. Here the manufacturing detail bites: forming a 0.4 mm channel into a 0.1 mm sheet means a corner radius near 0.1 mm, and the outer-fibre strain is

A third of the sheet thickness in tensile strain at the channel shoulder. That is why metallic bipolar plates are hydroformed or run through multi-stage progressive dies rather than single-hit stamped, and why the alloy is chosen for elongation first and corrosion second. Die clearance for the trimming and piercing stations sits at 5–8% of thickness per side — 5–12 µm — and every plate pair is laser-welded (0.3–1.0 mm weld width, 0.1–0.2 mm penetration) and helium leak-tested to better than 1 × 10⁻⁵ mbar·L/s, because a leaking plate pair mixes hydrogen and oxygen inside the stack.

A 1 MW stack needs 251 plates at roughly 1,200 cm² of blank each — 30 m² of 0.1 mm sheet, or 14 kg of titanium. At ₹3,500/kg for Gr1 sheet that is ₹49,000 of metal; the stamping, welding and coating cost three to five times that.

Gaskets and frames. FKM, EPDM or silicone, 0.2–0.5 mm, moulded or die-cut, compressed 20–30% at 1–2 MPa. Gasket creep is the dominant stack failure mode after membrane pinholes: a 10% loss of compression over 40,000 hours shows up as a leak, not as a performance drop.

End plates and assembly. Machined or cast 316L or aluminium, 20–40 mm thick, stiff enough to hold flatness across a 1 m² active area under 12–20 tie rods at 1–2 MPa. Flatness of 0.05 mm over the plate face is a normal print callout, and it is a machining quote, not a chemistry quote.

For an Indian fabrication supplier, this is the addressable list: progressive dies for 0.1 mm titanium, hydroforming, fibre-laser welding of plate pairs, CNC end plates, gasket moulding, PVD of gold or platinum, sintering of titanium powder, and the leak-test bench. None of it requires a cleanroom. All of it requires tolerance discipline that most job shops reserve for aerospace.

Iridium: 7.5 tonnes against a terawatt

Global primary iridium production is 7–8 tonnes per year, 80–95% of it as a by-product of platinum mining in South Africa. Supply is inelastic in the strict sense — you cannot open an iridium mine, only a platinum mine, and only if the platinum price justifies it.

The iridium-specific power density is the number that governs the industry:

With a loading L of 2 mg/cm², a current density of 2 A/cm² and a cell voltage of 2.0 V, the power density is 4 W/cm² and the iridium requirement is 0.5 mg/W — 0.5 g/kW. One tonne of iridium therefore buys about 2 GW of PEM capacity, and the world's 7.5 t/yr buys 15 GW/yr if every gram went to electrolysers and none to spark plugs, glass crucibles or acetic acid catalysts.

A terawatt of PEM electrolysis needs 0.015 g/kW, which means a loading of 0.06 mg/cm² at 2 A/cm² — a 30–40× reduction from current practice. The published roadmap runs 2.0 mg/cm² today, 0.4–0.5 mg/cm² in pilot production, and 0.1 mg/cm² as the research target. Below about 0.3 mg/cm² the catalyst layer stops being electrically continuous and the "islanding" effect takes over, so the reduction is a coating and ink-formulation problem, not a purchase-order problem.

At an iridium price above USD 6,500/oz — roughly ₹18,500/g — 0.5 g/kW puts ₹9,000/kW of iridium into a stack that costs USD 450/kW (₹39,600/kW). That is 23% of stack capex in one element, before the platinum cathode. Platinum is far less binding: 200 t/yr of supply against a 0.2 mg/cm² loading.

This is the real reason 1,363 of Tranche I's 1,500 MW went to alkaline. An alkaline cell uses nickel foam, Raney nickel and a Zirfon diaphragm. There is no iridium in it, and there is no supply ceiling.

Alkaline, PEM, SOEC, AEM: the four columns of a bid table

· Liquid alkaline · PEM · SOEC · AEM

Current density · 0.2–0.7 A/cm² · 1.0–2.2 A/cm² · 0.5–1.5 A/cm² · 0.3–1.0 A/cm²

Cell voltage · 1.8–2.0 V · 1.7–1.9 V · 1.25–1.35 V · 1.7–2.0 V

Stack energy · 48–53 kWh/kg · 47–52 kWh/kg · 35–40 kWh/kg · 48–54 kWh/kg

PGM loading · none · 2 mg Ir + 0.3 mg Pt per cm² · none · low PGM

Min load · 20–40% · 3–10% · limited by heat · 10–25%

Cold start · 20–60 min · 5–15 min · hours (thermal mass) · 10–30 min

Stack lifetime · 60,000–80,000 h · 40,000–80,000 h · 20,000 h · 20,000–40,000 h

Installed capex · USD 500–1,500/kW · USD 1,000–2,000/kW · above USD 3,000/kW · USD 800–1,500/kW

Footprint per unit · 5 MW in ~15 m² · 2.5 MW in ~2 m² · 0.05 MW · small modules

The alkaline column is the one that gets built in India, for three reasons: no PGM, a domestic supply chain for nickel and steel, and a 20-year operating record in chlor-alkali plants. The cost of that choice is current density. At 0.5 A/cm² an alkaline stack needs four times the active area of a PEM stack at 2 A/cm² for the same power, so the plates, gaskets and frames — and the factory that stamps them — scale with the gigawatt number rather than the efficiency number.

PEM wins on footprint and load-following: 3–10% minimum load against 20–40% for alkaline is what allows a stack to track a solar farm's evening ramp without shutting down. SOEC wins on thermodynamics and loses on everything else — 1.25 V per cell because part of the energy arrives as 800 °C steam, but 20,000 hours of lifetime and a stack that must never see a cold start.

The ₹/kg model

Levelised cost of hydrogen is a capital charge plus electricity plus replacement:

with the capital recovery factor

Take a 100 MW PEM plant in Gujarat: ₹90,000/kW installed (USD 1,020/kW, which is where Chinese stacks plus Indian BoP land), 8% cost of capital over 20 years, 60% capacity factor from solar-dominated supply, 53 kWh/kg at 2.0 V, ₹3.00/kWh power.

Total: ₹306/kg, or USD 3.48/kg at ₹88 to the dollar. That sits inside the World Bank's 2026 estimate of USD 3.60–6.80/kg for PEM systems and comfortably above any current offtake price.

Sensitivity is brutal and one-dimensional:

Change · Effect on LCOH

Power ₹3.00 → ₹2.09/kWh (lowest 2026 solar bid) · −₹48/kg

Power ₹3.00 → ₹5.25/kWh (SECI firm RTC) · +₹119/kg

Capacity factor 60% → 90% · −₹49/kg

Capex ₹90,000 → ₹150,000/kW · +₹98/kg

Cell voltage 2.0 → 1.8 V · −₹29/kg at ₹3/kWh

Stack life 40,000 → 80,000 h · −₹13/kg

To reach USD 1/kg — ₹88/kg — with 53 kWh/kg and ₹90,000/kW capex, the electricity price has to fall to about ₹1.60/kWh, which is below every solar bid ever recorded in India. The USD 1/kg target is not an electrolyser target. It is a power-price target that the electrolyser can only make easier by moving from 53 to 43 kWh/kg.

The policy arithmetic then explains itself. SIGHT pays ₹50/kg in year one, ₹40 in year two, ₹30 in year three. Against a ₹306/kg cost, that is a 16% subsidy in year one, tapering to zero while the plant is still 80% of the way through its capital recovery. The green ammonia auctions that closed at ₹49.75–64.74/kg are the proof: one kilogram of ammonia contains 0.1776 kg of hydrogen, so at ₹306/kg the hydrogen feedstock alone costs ₹54/kg of ammonia. Developers bid ₹49.75/kg because the incentive, the captive renewable power and the ammonia premium together close a gap that the electrolyser cannot.

Storage and delivery: the 1.05 and the 10–13

Electrolysers rarely deliver hydrogen at the pressure the customer wants, and the compression energy is not small.

Isothermal compression work is

which for hydrogen from 20 bar to 350 bar at 293 K gives about 0.96 kWh/kg — DOE's programme record quotes 1.05 kWh/kg. Real machines, with intercooling and 50–60% polytropic efficiency, consume 1.7–6.4 kWh/kg. That is 5–20% of hydrogen's LHV: tolerable.

Liquefaction is not. The theoretical minimum is 3.3 kWh/kg, 3.9 with ortho-para conversion, and real plants run 10–13 kWh/kg — 30–40% of the LHV burned to reach −253 °C. For any application that does not need cryogenic density, ammonia is the cheaper carrier: synthesis adds roughly 10 kWh/kg of ammonia, and cracking at the destination costs about 4.2 kWh/kg of hydrogen recovered.

Which is why India's green hydrogen export strategy is a green ammonia strategy, and why the SIGHT ammonia allocations (6,70,000 t/yr contracted to fertilizer units) matter more to the electrolyser order book than any mobility project.

Failure modes you will actually see

The DFM checklist for electrolyser parts

If a stack bill of materials lands on a quoting desk, these are the numbers that decide whether the job is profitable:

The last item is the one that fails audits. A stamped plate that was lubricated with a chlorinated drawing oil and degreased in a shared tank will poison an iridium anode in months, and the failure will be attributed to the catalyst supplier.

What to watch

Three numbers will decide whether India's 5 MMT target is a plan or a press release. The first is the delivered renewable tariff for dedicated hydrogen supply: at ₹3/kWh the LCOH is ₹306/kg, at ₹2.09 it is ₹258/kg, and nothing else on the cost sheet moves it as far. The second is iridium loading in commercial MEAs, which is on a 2.0 → 0.5 → 0.1 mg/cm² path and determines whether PEM can scale past 15 GW/yr. The third is whether the 3,000 MW/yr of awarded electrolyser capacity actually commissions — Tranche I was due in August 2026 and, on the government's own record, was not commissioned.

The engineering is not the bottleneck. The 1.23 V floor has been known since 1923, the stamping tolerance is achievable on any modern press line, and the membrane is a purchased commodity. What is missing is a ₹/kg that closes without a subsidy, and that arithmetic is decided in a field of solar panels, not in a stack.

More FabFlow blog posts