Forecasting mineral demand for electric vehicle batteries starts as arithmetic. EV unit sales multiplied by kWh per vehicle multiplied by material intensity per kWh, broken out by cathode chemistry, anode chemistry, electrolyte composition. The annual reports from IEA, BNEF, Wood Mackenzie, and the rest follow this logic chain with minor variations. Each link in that chain is a moving target, and the links interact with each other in ways the multiplication does not capture.

LFP took over 60% of Chinese installed capacity from a base below 25% in 2019. That shift has been written about extensively.

Less discussed is what it did to the lithium chemicals market internally. NCM consumes lithium hydroxide. LFP consumes lithium carbonate. When NCM dominated, hydroxide traded at a sustained premium to carbonate, and project economics for new lithium processing plants were built around that premium as if it were permanent. LFP's surge flipped the relationship repeatedly. A processing plant designed around hydroxide output needs months and significant capital to pivot toward carbonate if the market demands it. FS reports from lithium developers tend to handle this with a throwaway line about flexible product configuration. The flexibility exists on paper. In a physical plant with fixed reactor geometry and purification circuits, "flexibility" means shutting down a line, retrofitting, requalifying product with downstream customers, and eating the margin loss during transition. The gap between "technically possible" and "economically painless" is wide enough that it should show up in project risk sections. It usually does not.

LFP also introduces a phosphate rock demand channel that sits outside the battery minerals frame entirely. LFP cathodes require iron phosphate, which requires industrial phosphoric acid, which requires phosphate rock. Phosphate rock's primary market is fertilizer. Battery-grade purity requirements are higher than fertilizer-grade, but the ore comes from the same deposits. Battery demand for phosphate rock is currently in the single-digit percent range of global consumption, which looks trivial until you consider that phosphate rock supply elasticity is among the lowest of any industrial mineral. Reserves are concentrated, with Morocco and Western Sahara holding over 70% of the global total according to USGS estimates. Developing new phosphate mines takes over a decade. The battery industry scaling LFP and the fertilizer industry feeding a growing global population are drawing on the same constrained resource base. At current LFP growth rates, the crossover from "statistical noise" to "policy-relevant competition" is a matter of years, not decades. How many years depends on how fast LFP penetrates outside China, which nobody can forecast with confidence because it depends on a tangle of trade policy, patent licensing, and local manufacturing incentives that changes quarter to quarter.

The standard demand model multiplies unit sales by per-vehicle kWh, and per-vehicle kWh has been climbing steadily as automakers competed on range. That trend is decelerating. Past 500 km of range, consumer willingness to pay for incremental range drops sharply. 800V charging architectures reduce the practical penalty of a smaller pack. CTP and CTB integration raise volumetric utilization at the pack level, meaning the same usable energy can be delivered with less active material. The combined effect: mineral demand growth will lag EV sales growth, and the lag widens over time. Demand models that proxy mineral growth from sales growth without a per-vehicle material intensity adjustment will run hot. The size of the error depends on CTP/CTB adoption rates, which are tracked in the pack engineering world and largely ignored in the mining world, so the information needed to make the correction exists but sits in a different community's databases.

Storage's share of annual carbonate consumption is on track to overtake passenger EVs around 2028 based on current project pipelines. Modeling storage and passenger EV demand on a single aggregated curve obscures this structural difference.

Graphite gets a fraction of the analytical attention that lithium and cobalt receive, despite being the heaviest single material in a lithium-ion cell by weight. A 60 kWh pack contains roughly 50 to 70 kg of graphite anode material. The mismatch between graphite's physical importance and its share of industry discussion has persisted for years and has a straightforward explanation: graphite prices do not spike dramatically, so graphite does not generate headlines, and the processing monopoly is so complete that there is nothing actionable to say about near-term diversification.

Natural flake graphite must be spheroidized and surface-coated before it can function as anode material. China performs essentially all of this processing globally. Not most of it. Essentially all of it. North America and Europe have no operating commercial-scale spheroidized graphite lines. Graphite mining projects exist in Mozambique and Tanzania, at various stages of development, and without spheroidization and coating capacity co-located or at least located outside of China, the ore from those mines will end up in Chinese processing facilities regardless of mine ownership.

The technical barrier to spheroidization is moderate. The concentration happened through a combination of scale economics and willingness to absorb the environmental costs of a process that generates substantial fine particle waste. Breaking this concentration requires building processing capacity from scratch in jurisdictions with higher environmental compliance costs and without the existing chemical supply chain ecosystem. IRA lists graphite as a critical mineral. That designation has not yet produced a single operating spheroidization plant in North America.

China placed graphite under export permit requirements in 2023. A permit regime creates uncertainty about approval timing and outcomes, which forces downstream inventory builds, which raises working capital requirements and battery costs. The mechanism is indirect and the cost increase is diffuse, which makes it harder to quantify than a tariff or quota, but the effect is cumulative.

Longer term, if solid-state batteries bring lithium metal anodes to production, the graphite anode market faces displacement rather than evolution. Silicon-carbon composite anodes are a partial hedge, reducing graphite content while keeping graphite in the bill of materials, and some anode companies are moving in this direction for a mix of performance and risk management reasons. How much graphite demand survives in a world of lithium metal anodes and high-silicon composites is an open question that all current graphite mine investments are implicitly betting one way on.

Fluorine shows up in battery manufacturing through LiPF6 electrolyte salt, PVDF binder, and various additive chemistries. Upstream is fluorspar, majority-produced in China, with limited global expansion pipeline. Fluorine is not included in any major battery mineral demand model. It sits outside the analytical boundary.

Manganese has a different problem. Commodity manganese ore is cheap and abundant. Battery-grade high-purity manganese sulfate is not the same product, and dedicated production capacity for it is small globally, because demand has been small. If LMFP cathodes scale, manganese sulfate demand could outrun available capacity. New high-purity manganese sulfate plants take time to build and qualify.

The dominant structural feature of cobalt supply is that DRC cobalt production is overwhelmingly a byproduct of copper mining. Cobalt output tracks copper expansion cycles, not cobalt price signals. Copper price rises, copper mines expand, cobalt supply increases regardless of cobalt market conditions. Copper contracts, cobalt supply contracts passively.

This also accounts for why cobalt price behavior bears so little resemblance to lithium price behavior. Lithium's cycles are internal. Cobalt's cycles are derivative of copper. The two metals share a supply chain in batteries but have completely disconnected supply dynamics, and the disconnect traces back to geology.

There may be specific operations where cobalt is a more significant contributor to mine economics and therefore more responsive to its own price, but as a market-level description, the copper-cobalt coupling is the dominant feature.

INSG and most commercial databases report nickel supply-demand balance as a single number combining Class 1 (99%+ purity, suitable for dissolving into battery-grade sulfate) and Class 2 (NPI, ferronickel, consumed by stainless steel). A headline figure of "100,000-tonne nickel surplus" almost certainly refers to Class 2 surplus. Indonesia's capacity expansion has been predominantly NPI and ferronickel. The Class 1 supply-demand picture may be tight at the same time the aggregated number shows surplus.

This classification problem is well understood within the nickel trading community and poorly understood outside it. Battery investors and EV supply chain planners frequently reference the aggregated number without decomposing it, which leads to misallocation of capital and mispricing of supply risk for battery-grade nickel specifically.

Australian spodumene is the volume workhorse. Mining to lithium salt output in 3 to 5 years. Cost position in the lower half of the global curve. Ore mostly ships to China for toll conversion, a point that matters for reasons discussed below.

South American brine operations in Chile and Argentina offer lower theoretical costs. Evaporation-based extraction is constrained by climate, brine composition, and increasingly by water access conflicts with local communities and agriculture. DLE is the proposed alternative to evaporation. Lab-scale DLE results and field-scale DLE results diverge substantially. Brine chemistry variables like magnesium-to-lithium ratio, boron content, and seasonal temperature variation degrade DLE performance in ways that do not appear in vendor presentations. How much they degrade it, and whether engineering iteration closes the gap, is not settled. Supply projections that assume DLE reaches commercial maturity on an aggressive timeline carry meaningful downside risk if the technology takes longer. It might take longer. It might not. The range of outcomes is wide and anyone claiming certainty about DLE timelines is overconfident.

Chinese lepidolite in Jiangxi. Low lithium grades relative to Australian spodumene, longer and dirtier processing chains, high environmental pressure. These operations sit at the top of the global cost curve and act as swing supply. They flooded the market during the 2022-2023 price spike and shut down fast when prices dropped below their breakeven. When prices recover, they restart. This batch of capacity functions as a buffer that gives Chinese lithium supply substantially more price elasticity than external models typically credit. Lithium carbonate finding a floor below RMB 100,000 per tonne rather than continuing to fall is partly explained by lepidolite supply exiting and tightening the market from the cost curve's high end.

China's share of midstream processing, by mineral: lithium chemicals roughly 60 to 70%, cobalt refining above 80%, nickel sulfate above 60%, graphite spheroidization near 100%. Ores come from every continent. Processing converges on one country.

This is a cost structure outcome. Chinese processors operate with low energy costs, particularly hydropower in Sichuan and Yunnan. Chemical engineering supply chains are mature and locally available. Scale effects spread fixed costs. Building equivalent processing capacity in the West, even with subsidies, carries operating costs 30 to 50% higher. This is an estimate that varies by location and mineral, but the direction and rough magnitude are broadly accepted in the industry.

IRA's FEOC provisions and the EU's CRMA aim to build processing capacity outside this system. The execution is generating compliance costs that flow into battery prices. Traceability requirements for mineral sourcing under FEOC create situations where ore from the same mine, processed through different pathways, has different subsidy eligibility outcomes. Chain-of-custody auditing is becoming a cost line item in cell manufacturing. This amounts to a geopolitically-driven transaction cost layered onto the supply chain without corresponding physical value creation.

China's expansion of export permit requirements to graphite and other materials changes downstream inventory behavior. Companies shift from JIT toward precautionary stocking, which shows up as a permanent increase in working capital. This cost does not appear in mineral price data but increases total battery cost of ownership.

Sodium-ion batteries cap lithium's upside. When carbonate reaches a price threshold, sodium-ion economics beat LFP in storage and micro-EVs, pulling demand away. The substitution threat constrains lithium pricing at the margin even if sodium-ion market share stays small.

Sodium-ion also changes current collector demand. Both cathode and anode use aluminum foil instead of the copper foil that lithium-ion anodes require. Battery-grade aluminum foil at 10 to 16 micron thickness with tight quality specs has a limited supplier base. Large-scale sodium-ion deployment pushes aluminum foil demand up and copper foil demand down. The copper foil industry has already invested in capacity expansion for projected lithium battery growth. If sodium-ion gains share faster than expected, those copper foil investments earn less than their business cases assumed.

LMFP brings manganese into the cathode. Scaling depends on solving manganese dissolution and low-temperature issues. Global high-purity manganese sulfate capacity is not sized for rapid LMFP adoption.

Solid-state batteries with lithium metal anodes would increase lithium consumption per kWh while potentially eliminating graphite from the anode entirely.

Recycled material will not offset primary mineral demand in any significant volume before 2030. Battery operational life runs 8 to 15 years, extended to 15 to 20 with second-life applications. The retirement wave has not started.

As LFP's share of installed capacity keeps growing, the composition of the future retirement stream shifts toward cells that are harder to recycle profitably. If lithium carbonate prices are high enough, recovering the lithium alone could close the commercial loop for LFP recycling. If prices are moderate to low, LFP recycling probably needs mandates rather than market incentives. Current recycling investments are overwhelmingly modeled on NCM cell economics. That model breaks when the feedstock shifts to LFP.

Cross-border movement of spent batteries is constrained by Basel Convention provisions on hazardous waste. Where batteries retire (high-EV-penetration markets) and where recycling capacity exists may be different places. If they are, legal barriers prevent the physical flow of feedstock to processing facilities. This coordination problem has no obvious owner and no one appears to be working on it systematically.

Lithium lacks a deep, liquid futures market. CME and LME lithium contracts trade thinly. Physical prices are set by reporting agency assessments from Fastmarkets, Asian Metal, and peers. The entire supply chain operates without effective hedging instruments.

Spodumene concentrate auctions on Pilbara's BMX platform are a visible spot pricing reference. Each auction batch is small relative to global supply. In periods of high or low sentiment, a few aggressive bids set an auction clearing price that diverges from underlying equilibrium, and that price then propagates through reporting agency assessments into broader market pricing. A thin, sentiment-driven auction sample ends up anchoring expectations for a much larger market. This amplifies price moves in both directions independent of any change in supply or demand fundamentals.

At each cycle bottom, the development pipeline in Western jurisdictions shrinks, incumbent supply survives, and concentration increases. This ratchets. Higher volatility raises financing risk for new Western projects, which reduces the pipeline further, which increases future concentration, which feeds the next cycle's volatility amplitude.

Water and energy are pushing the cost curve around in ways that mineral price analysis does not always capture. Hydropower-dependent lithium processors in Sichuan and Qinghai face electricity cost spikes in dry years significant enough to reshuffle their position on the cost curve. Brine operations in Chile and Argentina face intensifying conflict with agricultural and community water users. Water rights disputes have become a gating constraint on social license for brine projects, separate from and in addition to geological and technical constraints.

Shifting the 2030 global EV penetration assumption from 40% to 50% moves lithium demand projections by over 20%.

Chemistry share assumptions drive mineral mix. Trend extrapolation is the standard approach. Chemistry transitions tend to happen nonlinearly when a manufacturing process crosses a yield or cost threshold.

Battery life assumptions feed directly into recycling volume estimates. A two-year increase in average battery life reduces 2035 recycling return volumes by 30 to 40%.

Policy discontinuity risk is largely unmodeled. EV penetration in many markets depends on purchase incentives. When incentives expire or get cut for fiscal reasons, penetration can step down sharply. Germany's experience after removing EV subsidies in 2024 demonstrated this. Smooth penetration curves without discontinuity scenarios produce confidence intervals that are too narrow.

Flow versus stock: most models calculate mineral demand from new vehicle sales. As the EV fleet base grows, replacement sales (second or third EV purchase by owners scrapping an older EV) become a larger fraction of total sales. Replacement generates retired batteries that enter recycling. Models that do not separate fleet growth demand from replacement demand will overstate primary mineral needs as the market matures. The direction of the bias is unambiguous. The magnitude depends on fleet age distribution, which is knowable in principle and rarely incorporated in practice.

There are more underestimated variables than overestimated ones in the standard battery mineral demand model. On the underestimate side: storage demand intensity, phosphate rock competition with agriculture, graphite processing concentration, cobalt supply's dependence on copper cycles, the LFP recycling economics gap, compliance and inventory costs from geopolitical fragmentation, fluorine and manganese tightness outside the model boundary. On the overestimate side: per-vehicle material intensity is being eroded by pack engineering, and DLE commercial timelines carry downside risk. Most of the underestimates point toward supply being tighter than headline numbers indicate. The overestimates point toward demand being somewhat less steep than the most aggressive projections. Whether these offset or one side dominates depends on timing and geography in ways that a single global model handles poorly.