The Battery Capital Decision

The board is being asked to choose between two versions of the same ambition. Both aim at a large, durable position in India’s electrification value chain. They differ in where the capital sits and, critically, in what has to be true for the capital to come back.

Our conclusion is that cell manufacturing is the single worst place in this value chain to put $1bn today — not because India should not have cell plants, but because the economics of being the eighth Indian entrant into a globally over-supplied commodity, using licensed technology we cannot obtain, behind a tariff wall the government has repeatedly declined to build, are severely negative. Every other link in the chain has better returns, lower capital intensity, faster reversibility, and — the point most often missed — benefits from the falling cell prices that would destroy a cell plant.

The five findings that produce this answer

1. The margin does not exist at current protection. At year 5 a base-case plant sells at $52.1/kWh against a cash cost of $50.2/kWh. Gross margin over the imported materials basket is about $16/kWh; conversion cost consumes essentially all of it. Solving for the conversion cost that clears the hurdle returns no solution: even at zero conversion cost the project returns −$229m, because headroom over the materials basket decays from $18.1/kWh in year 3 to $9.5/kWh by year 12 as Chinese prices fall faster than the basket does.
2. The whole thesis reduces to one wager: that Chinese LFP prices stop falling. Holding a generous 20% duty constant and varying only the price path, NPV swings from +$415m to −$805m. Flat prices are break-even. Every credible forecaster expects continued decline.
3. The technology is not available on acceptable terms. China’s export controls on cathode, anode and process technology resume 10 November 2026. In January 2026 a Chinese licensor withdrew from technology-sharing talks with India’s largest energy group, citing those controls. Wood Mackenzie attributes India’s 10–15 year gap in part to deep technology dependence on Chinese and Korean licensors.
4. The strategic-autonomy argument is backwards. A cell plant converts a liquid, substitutable dependency on finished cells into an illiquid, hostage dependency on Chinese cathode and graphite, with $1bn of fixed assets standing behind it.
5. Policy is moving away from protection, not toward it. Budget 2026–27 cut duties on cell machinery and lithium precursors. The one protective measure — 20% local content for storage — is satisfiable by pack assembly at $5m/GWh, not cells at $80m/GWh.

What would reverse this recommendation

One input dominates: captive demand. If the group already controls, or can contract, offtake for more than half of a plant’s nameplate output — a vehicle OEM, a large IPP with a committed storage pipeline, a telecom or data-centre backup fleet — the utilisation and price-realisation assumptions that sink the base case improve materially, and the upside stops being a fantasy.

Even then the answer is sequence, not scale now: build the captive demand first, prove it with three years of contracted volume, then build cells behind it. A plant built ahead of its captive load is a merchant plant wearing a captive plant’s business case. We have modelled the group as having no material captive offtake today. If that assumption is wrong, the phasing in Section 10 should be re-timed — the direction does not change.

Four things settle this question before any model is opened.

1. A Chinese licensor walked away from India’s best-capitalised entrant

In January 2026 Bloomberg reported that Xiamen Hithium had withdrawn from technology-sharing talks with Reliance, following Beijing’s curbs on overseas transfers of battery technology. Reliance stated that there had been no change to its plans and that its 2026 timeline remained intact. What is not in dispute is the withdrawal itself and its stated cause: a counterparty well placed to license LFP process technology into India stepped back because its own government restricted the transfer.

That matters more than any single company’s response to it. Reliance holds an ACC-PLI award, a ₹75,000 crore new-energy complex and captive renewable demand. If a licensor withdraws from that counterparty, a later entrant holding none of those advantages has no stronger claim on the same technology.

2. The technology is not for sale on acceptable terms

China’s Ministry of Commerce brought LFP and NCM cathode materials, artificial graphite anode material, granulation and graphitisation process technology, high-density cells and the associated production equipment under export licensing effective 8 November 2025. The measures were suspended on 7 November 2025 — but only until 10 November 2026, roughly eleven weeks from this meeting. Every credible Indian cell programme depends on a Chinese or Korean licensor. Chinese licensors now require a licence granted at MOFCOM’s discretion; Korean licensors are not offering LFP process technology to Indian third parties on commercial terms.

This is not a risk that contract language hedges. It is a foreign government’s discretionary veto sitting upstream of a $1bn asset, and it becomes live again inside this financial year.

3. The tariff wall does not exist, and policy is moving the other way

The investment case for domestic cells rests entirely on import protection. India has consistently chosen the opposite. The Union Budget 2026–27 removed customs duty on lithium-ion cell manufacturing machinery, extended those capital-goods exemptions to battery storage systems, and took basic customs duty on lithium carbonate, hydroxide and oxide to zero from 7.5%. Every one of those moves reduces the cost of importing the inputs and does nothing to raise the cost of importing the finished cell.

The revealed preference is consistent and rational: the government’s priorities are EV affordability and cheap grid storage for the 500 GW non-fossil target, and both are served by cheap imported cells. Four or five cell manufacturers lobbying for protection will lose to vehicle OEMs, distribution companies and consumers who benefit from its absence.

4. India has 2 GWh commissioned against 226 GWh of announcements

Wood Mackenzie’s August 2026 assessment puts India 10–15 years from a globally competitive, self-sufficient cell industry, with roughly 2 GWh commissioned against China’s 2,695 GWh. The ₹18,100 crore ACC-PLI scheme, approved in May 2021, targeted 50 GWh. Government data cited in February 2026 puts installed capacity at 1 GWh, by Ola, with the remaining beneficiaries yet to commission. Penalties have been imposed for missing the December 2024 milestone — ₹12.5 lakh a day for Ola, roughly ₹35 crore accrued, and ₹5 lakh a day for Reliance and Rajesh Exports. As of July 2026 no firm had claimed any incentive under the scheme despite ₹5,180 crore of investment. Awardees seeking extensions attribute the delay to difficulty sourcing equipment and machinery, primarily from China.

Two readings of that gap are possible. The optimistic one is that the field is empty and the prize unclaimed. The correct one is that four consecutive cohorts of well-funded, government-subsidised Indian entrants have found this harder than they projected, and the constraints they hit apply to us identically.

India’s battery demand is real and growing fast. It is also consistently smaller than the forecasts that justify gigafactory announcements, and its composition is shifting toward the segment least willing to pay a premium for domestic supply.

ICEA projects Indian lithium-ion demand of 115 GWh by 2030; other published forecasts for that year span roughly 70–160 GWh, a dispersion worth holding in mind. Two- and three-wheelers dominate volume and are ferociously price-sensitive. The fastest-growing block is grid storage: India tendered over 130 GWh of storage in 2025 alone, and cumulative tendered capacity has gone from 6.8 GW in 2018 to 90.7 GW in 2025. Wood Mackenzie puts India’s 2026 demand pipeline from competitive tenders at roughly 260 GWh, of which domestic manufacturing supplies under 1%. Storage buyers select on lowest levelised cost, with no meaningful preference for domestic cells beyond the 20% content floor.

Set 226 GWh of announced Indian cell capacity through 2035 against 70–160 GWh of annual demand by 2030. A tender pipeline is larger than annual consumption because it procures several years of future delivery, so the two should not be set against each other directly — but on either measure, and even discounting announcements heavily for the historical execution rate, the credible domestic pipeline plus imports covers credible demand. We would not be filling a supply gap. We would be competing for share in a market already contested by better-positioned incumbents and by an import channel that is structurally cheaper.

We modelled a 10 GWh LFP plant at $80m/GWh of greenfield capex — consistent with the ~$80m/GWh benchmark for greenfield gigafactories outside China, close to double the in-China figure. Capex is spent across three years; production ramps from 25% to 85% utilisation; cells are sold at import parity, defined as the landed Chinese LFP price grossed up for effective protection and a 3% domestic logistics advantage. Cash cost is built from an imported materials basket of about $39/kWh declining slowly, plus a conversion cost carrying a fixed-cost absorption penalty at low utilisation. Full model specification is at Appendix 18.

Why the base case loses money even though the plant runs

At year 5 the base-case plant sells at $52.1/kWh against a cash cost of $50.2/kWh. It is not losing money on operations; it is earning about $2/kWh of contribution against a capital charge requiring roughly $11/kWh of EBITDA to justify $80m/GWh. Gross margin over the imported materials basket is around $16/kWh, and conversion cost consumes essentially all of it. Solving the model for the conversion cost that would clear the hurdle at current protection returns no solution. Even at zero conversion cost the project returns −$229m at a 5.9% IRR — not because the plant cannot cover its capital charge in the middle years, but because headroom over the materials basket decays across the horizon, from $18.1/kWh in year 3 to $11.2 by year 10 and $9.5 by year 12, as Chinese prices fall 3%/yr against a basket falling 1.5%/yr. The margin is a melting asset unless protection ratchets upward over time.

That result is the arithmetic behind Wood Mackenzie’s finding that locally manufactured cells cost 25–40% more than imported ones — a premium it attributes to limited scale, higher financing costs and an underdeveloped supplier ecosystem rather than to any fundamental cost disadvantage. The same assessment puts India second only to China among major manufacturing destinations on underlying cost, with a 9% advantage over South Korea. That is the strongest fact available to the opposing case, and it does not rescue the investment: the gap that matters is scale and ecosystem, and scale and ecosystem are what a first plant does not have. We are not modelling incompetence. We are modelling a competent Indian plant losing to scale, learning-curve position, and a materials chain that sits in someone else’s country.

Exhibit 4 is the most important page in this report. Strip away the strategic language and the investment reduces to a single wager: that the price of Chinese LFP cells stops falling. That is a bet against the two most powerful forces in the industry — structural Chinese overcapacity and the LFP learning curve — placed with $1bn of shareholders’ capital, irreversibly, in a fixed asset, on someone else’s technology.

We would not be entering an empty field. Wood Mackenzie counts 226 GWh of Indian cell capacity announced for construction through 2035 against 2 GWh commissioned, and four ACC-PLI beneficiaries are already contracted to build. Against that field, a marginal entrant would hold no advantage on cost, technology, subsidy or customer access.

The more important observation is what the field has in common. Wood Mackenzie attributes India’s 10–15 year gap in part to deep technology dependence on Chinese and Korean licensors, and ACC-PLI awardees seeking extensions attribute their own delays to difficulty sourcing equipment and machinery primarily from China. Two consequences follow. First, licensed technology is not scarce technology, and non-scarce technology does not produce durable margin — it produces a race in which the licensor captures the rent. Second, access is contingent on a foreign export-control regime that has already been exercised.

The Indian precedent that rhymes

A decade of ALMM listing and import protection produced roughly 193 GW of Indian solar module assembly capacity under ALMM List-I, against only about 31 GW of domestic cell capacity under List-II. The shallow step localised; the deep, capital-intensive step did not, despite sustained policy support. ALMM List-II, which finally forces domestic cells, came into force on 1 June 2026 and immediately exposed the shortage — roughly six times more module capacity than the cell capacity needed to feed it.

The policy achieved its national objective. Whether it rewarded the manufacturers who answered its call is a separate question, and the distinction — national success, shareholder disappointment — is precisely what this board has asked us to guard against.

The case for domestic cells is most often made on dependence-reduction. On inspection it does not survive.

India has no cell-scale domestic supply of the key inputs. Wood Mackenzie puts China’s share at between 85% and 98% of global capacity across every major supply-chain component, and China now licenses the export of the process technology as well. India’s domestic materials build-out is real but early: Himadri’s 200,000 tpa LFP cathode plant is phased over five to six years, Epsilon’s graphite anode plant reaches 30,000 tpa by 2028, and Altmin’s cathode pilot produces 100 kg a day.

A domestic cell plant therefore does not remove the dependency. It converts a liquid, substitutable dependency into an illiquid, hostage one. Today, if a supplier is cut off, we buy cells from a different country next quarter. After the investment, if cathode or graphite is cut off, we own a $1bn asset that cannot run — and we still cannot make the cells. When China restricted graphite exports in October 2023, shipments to India were delayed two to three months: survivable for a buyer, ruinous for a plant with fixed costs and take-or-pay obligations.

Genuine autonomy requires the materials chain, not the assembly step: cathode active material, graphite anode, separator, electrolyte and refined lithium. That is a further $1–2bn and another decade. The board should reject any version of this proposal that claims strategic autonomy from a cell plant alone. If autonomy is the objective, the honest sequence is materials first — which is also, conveniently, where the better returns and the policy support currently sit.

Indian industrial policy for batteries is generous toward inputs and capital goods, and conspicuously silent on protecting finished cells. The consistent pattern across three budgets is duty relief on machinery, precursors and critical minerals. Budget 2026–27 extended cell-machinery exemptions to storage systems and zeroed the duty on lithium carbonate, hydroxide and oxide. Cells themselves carry a 5% concessional basic customs duty, and the government is reviewing a proposal to extend that concession for a further two years against a background of rising imports. Whether it is extended or allowed to lapse is a dated, observable event bearing directly on gate G1.

The ACC-PLI pool of ₹18,100 crore is fully allocated and under strain, with penalties imposed and awardees seeking extensions. A new entrant should assume no ACC-PLI support. Removing that subsidy from the model is worth roughly nine dollars per kWh over five years — the difference between a marginal proposition and a clearly bad one.

The most bullish policy development available to us is a domestic content requirement for storage, modelled on solar’s ALMM. It has begun: 20% local content on VGF-backed storage projects from December 2025. But the threshold is set where local packs, enclosures and balance-of-plant satisfy it. A ratchet to 50–60% domestic value addition with five-year visibility would require cells and would change this analysis. Nothing announced to date does.

A cell plant is not a thirty-year annuity. It is closer to a semiconductor fab: continuous process reinvestment, with the technology leader taking the profit and followers taking the residual. An Indian plant built on a licensed process faces continuing re-tooling that its licensor may or may not be willing or permitted to supply.

Sodium-ion is the specific near-term threat

CATL’s Naxtra platform now includes a large-format sodium-ion cell built for stationary storage: 300-plus Ah, 160 Wh/kg, 97% round-trip efficiency, and cycle life above 15,000 at 80% capacity retention. Its passenger-EV cell reaches 175 Wh/kg. On 27 April 2026 CATL signed a three-year, 60 GWh sodium-ion supply agreement with HyperStrong for storage — the largest sodium-ion storage order to date — and a Changan passenger EV using Naxtra cells is scheduled to launch in mid-2026. CATL states cost parity with LFP by end-2026 and continued decline thereafter.

Sodium-ion attacks precisely the two segments an Indian LFP plant would serve: grid storage and entry-level two- and three-wheelers. It uses no lithium, nickel, cobalt or graphite anode, so it also routes around the materials chokepoints. It is simultaneously a threat to a cell plant and a reason the low-capital strategy is more robust: a storage asset owner and a pack integrator are indifferent to which chemistry wins and benefit from either getting cheaper.

The rest of the technology risk register

• Chemistry mix. BloombergNEF attributes part of the 2025 price decline to a market-wide shift in preference toward lower-cost LFP cells. This analysis assumes LFP throughout.
• LMFP and high-manganese variants require cathode and electrolyte changes mid-life, at the licensor’s discretion.
• Dry-electrode processing and large cylindrical formats could strand a wet-process line built in 2028.
• Lithium price direction cuts both ways. 2026 forecasts cluster in the $17,000–24,000/t band with credible deficit calls from Morgan Stanley and UBS. Rising lithium compresses cell-maker margin unless fully passed through, improves recycling economics, and strengthens the sodium-ion case. Only a sustained squeeze — Exhibit 4’s top row — helps a cell plant, and it helps by making the competition more expensive rather than us more efficient.

The board deserves the opposing argument at full strength, not a straw version.

The argument. India has protected strategic manufacturing before — ALMM listing for solar cells, and now a domestic content floor for storage. It will do the same for cells once domestic capacity is credible enough to protect, and the 260 GWh tender pipeline shows the demand is there. Land, grid connections, water allocations and state incentive packages are being locked up now and will be materially more expensive later. Cell plants take thirty months to build and five years to run well; an entrant that waits until policy is confirmed arrives four years after the incumbents and never catches the learning curve. Under this reading, the negative NPV in Exhibit 2 is simply the cost of the option, and the bull case is the real case.

Why it does not carry. Three reasons.

• The solar precedent argues the other way. India protected modules, not cells — 193 GW of module capacity against 31 GW of cells — and the profitable position turned out to be owning generation. Protection arrived for the shallow step and, for a decade, never for the deep one. Betting $1bn on India reversing that pattern for batteries, against the countervailing interest of EV affordability and cheap grid storage, is a policy bet dressed as an industrial one.
• The first-mover advantages are cheap to buy separately. Land, grid connectivity, state MOUs, environmental clearances and a nucleus engineering team cost tens of millions, not hundreds. That is exactly what the reserved option in Phase 0 purchases. We can hold the first-mover position without holding the fixed asset.
• Waiting is nearly free; being early and wrong is not. Equipment costs are falling, licensing terms will clarify once China’s November 2026 decision lands, and 2030 demand will still be growing when a 2030 FID commissions in 2033. The cost of a three-year delay is a few points of share in a commodity. The cost of being early and wrong is $600m–$970m of permanently impaired capital. That asymmetry, not a forecast, settles the timing question.

Four operating businesses and one option, sequenced so each phase is funded from a defensible standalone case and each is reversible if the thesis breaks.

What Exhibit 5 does and does not say

This is not a portfolio of extraordinary businesses. It clears the hurdle by 1.3 points, and two of the five lines are marginal on their own. We present it that way deliberately, because the honest comparison is not “good versus bad” but “adequate and reversible versus poor and irreversible.” Three caveats:

• Storage ownership is being bid to the bone. Standalone storage tariffs fell over 71% between 2022 and 2025 while pack prices fell 36%, and IEEFA assesses roughly 75% of allocated two-hour capacity as viability-at-risk. Lenders underwrite to 15–20% IRRs. Our 14.4% assumes we bid selectively and walk away from most auctions — four-hour configurations, hybrid contracted-plus-merchant revenue, and sites where we control land and evacuation. This leg fails if we chase volume.
• Cathode active material is the same commodity trap one step upstream. Chinese LFP cathode is also in overcapacity and its conversion margins are thin. At 9.0% it does not clear the hurdle standalone. We recommend it only with an anchor offtake contract, and only because the domestic content ratchet is the one policy lever with genuine forward momentum. If no anchor offtake is signed, this $110m goes to the reserve.
• Recycling is option value, not a profit centre yet. Feedstock is roughly 15,000 tonnes in 2026 growing toward 280,000–350,000 tonnes by 2035, with EPR obligations biting from FY2027–28 and recovery minimums already set at 90%. LFP recycling economics are structurally weaker than NMC because there is no nickel or cobalt to recover. We size it for the option and the compliance-driven feedstock position, not for near-term earnings.

Phasing

These are the conditions under which the recommendation reverses. They are written to be falsifiable and reported on quarterly. Exercise requires at least five of seven, with G1 and G5 mandatory. The thresholds are our proposal and require the board’s ratification; the status readings against them are as at August 2026.

Current standing: zero of seven passed, two partial, both mandatory gates failed. That is not a marginal call requiring judgement. It is a clear reading, and it is the reason the recommendation is deferral rather than a smaller cell plant.

Sections 12–16 set out the full execution plan for the conditional 20 GWh plant. It is approved for preparation, not construction: FID remains at the Q4 FY2030 gate review. The purpose is that if the gates turn, we are ready to build rather than ready to start planning.

A gigafactory plan fails or succeeds on a small number of choices made before any steel is ordered. Four of them determine everything downstream.

Why a single SKU into grid storage first

The launch market is stationary storage, not vehicles, and that choice cascades. Storage cells carry a six-to-twelve month qualification cycle against IEC 62619 and UL 1973 at cell level, versus eighteen to twenty-four months and IATF 16949 for automotive. Storage buyers tolerate wider capacity distribution than automotive packs, so early-production cells that would be scrapped for a vehicle programme remain saleable. And storage is where India’s demand compounds fastest and where the domestic content ratchet applies.

Most importantly, it solves the offtake gate. The storage ownership and integration businesses built in Phases 0 and 1 become the anchor customer for Phase 2. At roughly 7 GWh of owned storage assets and a 10 GWh pack and integration business, the group would internally consume a meaningful share of a 20 GWh plant’s output before a single external contract is signed. That is how G2 gets met, and it is why the phasing runs in this order rather than the reverse.

This is the critical path and the reason Phase 2 cannot be compressed. Gate G5 requires a process not subject to a discretionary foreign export licence. That single clause eliminates the route every other Indian entrant has taken.

The tension this creates, quantified

Escaping the licensor veto also means escaping the Chinese equipment base, and that is expensive. Chinese equipment — coaters, stackers, formation systems — runs roughly 30–40% below European, Japanese and Korean equivalents and is now subject to the same export licensing that blocks the process technology. A predominantly non-Chinese vendor base moves capex intensity from about $70m/GWh toward $90m/GWh.

The plan therefore targets $80m/GWh as the design case — a mixed vendor base using non-Chinese equipment where export-control exposure is real (electrode coating, formation, process-critical metrology) and cost-optimised sourcing where it is not (materials handling, conveyance, building systems, utilities). The $1,400m headline in Exhibit 10 is stated at $70m/GWh for comparability with Exhibit 2; the board should plan against $1,600m and treat anything below it as upside.

Site selection is a Phase 0 activity because land, power contracts and evacuation approvals take three to four years to assemble and are cheap to hold. Criteria are weighted by their effect on conversion cost — the one lever in Exhibit 3 we actually control.

What controls the ramp

• The pilot line is the ramp. The 100–200 MWh line funded in Phase 0 exists to burn through the learning curve at 1/100th the cost of burning through it at scale. By FID it should have produced qualified cells to a frozen recipe for at least twelve months. If it has not, FID should not proceed regardless of gate scores.
• Freeze the recipe before block 1 install. The most expensive failure mode is process changes chasing performance after equipment is committed. Recipe freeze is a hard milestone at FID minus six months.
• Formation and aging set the cash cycle. A 14-day aging cycle with 796,000 cells resident is roughly $40m of inventory standing still. Compressing aging is the highest-value process improvement available and should be a standing engineering objective.
• Metrology before volume. In-line thickness, alignment and weld inspection at block 1 rather than retrofitted at block 3. Yield problems you cannot see, you cannot fix.

Gate G5 requires the materials chain secured before FID, not after. Cathode comes from our own Phase 1 plant; everything else must be dual-sourced with at least one non-Chinese option under contract.

Organisation

After technology access, people are the binding constraint. A 20 GWh plant needs roughly 2,450 staff at maturity, of whom around 400 require process, automation or metrology skills that barely exist in the Indian labour market today.

• Hire the nucleus now, not at FID. Thirty engineers — electrode, assembly, formation, quality, equipment — recruited during Phase 0 and employed on the pilot line. This team is the most valuable asset the option budget buys, and the reason the option premium is $80m rather than $10m.
• Three recruitment pools. Returning Indian engineers from Chinese, Korean and US cell makers; process organisations displaced by gigafactory failures elsewhere, which pairs with Route C; and domestic graduates trained on the pilot line over eighteen months.
• The pilot line is a training academy with a product. Budget it as both. Every operator who joins block 1 having already run the pilot line removes weeks from the ramp.
• Retention is a live risk. Once India has four or five operating gigafactories, a trained process engineer is the scarcest commodity in the sector. Assume aggressive poaching and structure long-term incentives accordingly.

Funding

Section 10 stated that Phase 2 would be funded from the reserve, Phase 0–1 cash generation and project finance rather than a second call on the group. On detailed build-up that is very nearly true, but not quite.

One structural note. ECA financing availability is a function of the vendor decision in Section 13 — Korean and Japanese equipment brings Korean and Japanese export credit with it, European equipment brings European. A predominantly non-Chinese vendor base costs more in capex but unlocks financing a Chinese vendor base would not. That partially offsets the $180m-per-$10m/GWh penalty in Exhibit 8, and should be evaluated jointly with vendor selection rather than after it.

Pre-FID workstreams

FID in 2030 is only a real decision if the four years before it have been used. The $80m cell option premium in Exhibit 5 funds the work below.