InfoLink's 29 July assessment puts domestic ESS cells in the 280 Ah and 314 Ah formats at RMB 0.360/Wh, down 2.7% week-on-week. Battery-grade lithium carbonate the same week: RMB 144,000/MT, up 2.1% WoW. Output price down, input price up. A week earlier the pair moved the other way round, lithium off 7.2% WoW while cells and systems held flat. That two-way decoupling is still the load-bearing fact about Chinese cell economics. The cost curve stopped setting the price somewhere in H1 2026, and nothing in this print says it has resumed.
The second derivative on that price is the one number I will not give you. The reason it is missing matters more. InfoLink attributes this week's 2.7% to specific specifications and orders, including preferential terms for large or long-term buyers locking in Q4 volume. Discounts written against a Q4 commitment are book-filling, not spot behaviour. A weekly series whose composition alternates between broad market movement and negotiated forward commitment is measuring order mix as much as price, and it will not carry a rate-of-change-of-the-rate read at weekly frequency. Benchmark's Chinese 314 Ah LFP assessments, launched February 2026, are the right instrument for that job; no current numeric value is public. So the master variable today is a level plus a composition, not a trajectory of a trajectory. The only clean second derivative in this system sits one step downstream, on export volume, which is what the rest of this maps.
NMC is absent from what follows by object, not by oversight. The routes mapped here move 280/314 Ah LFP prismatics, and every threshold that binds them (MOFCOM's energy-density cutoff, the Section 48E cost weights) acts on the large-format storage cell rather than on chemistry in general.
Almost nobody reading this buys at RMB 0.360/Wh. The price that governs a sourcing decision is the one that arrives: ex-works cell, plus freight, plus the destination duty stack, plus or minus whatever China's own tax code did to the export invoice that month, then multiplied by whether the cell disqualifies the project it goes into from a credit worth more than any duty on the manifest. That composite is the dependent variable, and four dated events reprice it between today and 1 January 2027.
Announcement No. 58 Controls Equipment, Not Storage Cells
MOFCOM Announcement No. 58 sets export licensing for cells and packs at an energy density of at least 300 Wh/kg. Large-format LFP storage cells are nowhere near it. Sunwoda publishes 180 Wh/kg for its 314 Ah ESS cell. HiTHIUM publishes 173.2–180 Wh/kg across its 314 Ah variants. The threshold sits roughly 65% above the product, and no plausible LFP chemistry path closes that gap inside this decade.
What No. 58 does control, as a separate category, is production equipment and process technology: winding, stacking, liquid-injection, hot-pressing, formation and grading machines, plus listed battery and synthetic-graphite technologies. Suspension of the whole measure runs to 10 November 2026.
For anybody sourcing LFP for storage, then, 10 November is a factory-transplantation date rather than a cell-availability date. Exposure sits with whoever is underwriting a non-Chinese cell line on winding, formation and grading equipment ordered from Chinese suppliers, or on a process recipe licensed from a Chinese cell-maker. That equipment underwriter and the procurement lead worried about cell supply have been reading the same headline for eight months; only the first has cause to.
We have logged the November date before and treated reinstatement as inference tied to Q4 bilateral dynamics rather than as observed behaviour. That holds, with one addition worth stating flatly. The reviewed official record publishes no applicant-level licensing data of any kind: no application counts, no grants or denials, no named recipients. Enforcement intensity under No. 58 is not merely uncertain; it is unmeasurable from public sources. Two things are actionable, the perimeter and the date, and the perimeter excludes storage cells.
Outward Pressure, Accelerating
Strengthening. June combined power-and-ESS battery exports: 36.2 GWh, up 23.7% MoM and 48.7% YoY, with H1 at 181.3 GWh, up 42.5% YoY (CABIA data via CnEVPost, 16 July; the structural 24–72 hour English-language lag on the underlying release applies). The second derivative lives in the gap between those two growth rates. June's 48.7% YoY runs above the 42.5% H1 pace, which makes export growth accelerating rather than decaying.
Mechanism, with the intermediate step written out, because the usual version skips it. Domestic production continues to exceed the sum of domestic power-battery installations and domestic ESS absorption, a gap we track as separate absorption channels rather than as one demand number. A Tier-2 producer cannot idle a line without creating a fixed-cost absorption problem, and, per CRU Group's published analysis "China's overcapacity: Will its battery industry consolidate?", without colliding with provincial employment and industrial-policy commitments. Those provincial mechanisms rest on CRU's proprietary research and are not independently verifiable from public English-language sources; read them as inferred. The residual volume therefore goes outward, and export orders clear at whatever price keeps a line running. That makes the export channel a pressure-relief valve on the domestic floor rather than a premium market. The chain from overcapacity to export price runs through the shutdown decision, not through demand.
Sizing the cohort that behaves this way is where public data thins out. Our standing baseline for Tier-3 utilization is 30–40%, structural rather than dated. CRU's published work supplies the surrounding structure: more than 2 TWh of Chinese capacity in 2024, roughly 60% above total demand, with non-Tier-1 output up 146% between H1 2024 and H1 2025. Against that, CATL disclosed 94.86% utilization for H1 2026. The dispersion between those two numbers is what pushes volume outward.
One named Tier-2 has published a dated 2026 figure. REPT Battero's 9 July profit alert forecasts H1 revenue of RMB 14.5–15.2 billion, up 52.8–60.1% YoY, and net profit of RMB 700–850 million against a RMB 63 million loss a year earlier. Preliminary, unaudited, and a P&L forecast rather than a utilization, margin or cash disclosure. It is a counterexample to blanket Tier-2 distress and nothing beyond that. No named Tier-2 exporter has published a 2026 utilization rate, segment margin or cash position. Every claim below about Tier-2 margin therefore rests on the dispersion and the absorption arithmetic, not on anyone's balance sheet.
Timeline: monthly, via CABIA. The October–December prints are the test set out below.
Why June Accelerated After the Rebate Cut
June exports accelerated after the VAT export rebate on batteries had already gone from 9% to 6%, effective 1 April under MOF/STA Announcement No. 2 of 2026. Three readings survive that fact:
- The three-point withdrawal was never binding on the shipment decision.
- Volume is being pulled forward ahead of the rebate going to zero on 1 January 2027.
- Overseas price realization is large enough to swamp three points.
Reading 3 is the weakest of the three. InfoLink's June work found limited direct price impact since April, weak buyer acceptance of higher prices, and incomplete pass-through to European customers. That is margin absorption, not premium realization.
Reading 2 has the wrong shape for June. The rebate test runs on the export declaration date, and a shipper optimizing against 31 December, with Rotterdam transit and a Q3 contracting cycle, concentrates that pull-forward into Q3 and Q4, not into June.
Which leaves reading 1 as the assessed explanation: three points went into supplier margin and never reached the shipment decision. This is editorial inference. CABIA publishes volume without contract dates, rebate incidence, customer ASPs or exporter margins, so the readings cannot be separated on public data.
The value in resolving it is the test it generates. If pull-forward is real, October through December break above the June run rate by more than the H1 trend, and Q1 2027 shows an air pocket with softer spot. If absorption is the whole story, Q4 tracks trend and the January elimination lands on Tier-2 margin rather than on volume. Those two outcomes point opposite ways for anyone deciding this quarter whether to lock Q1 2027 pricing or float it.
The Constraint Stack, Where Tariffs and Eligibility Are Not the Same Quantity
Mechanism first, because this force behaves unlike the other two. Nothing in the constraint stack touches the ex-works price. It acts on realized destination economics, through two channels that get modelled as one and should not be. A tariff is an ad valorem addition: continuous, scalable, negotiable inside a price discussion, reducible by classification work. An eligibility test is a binary gate: it steps rather than scales, it cannot be discounted around, and failing it forfeits a credit worth a multiple of any duty rate. A ten-point tariff move and a five-point threshold move are not comparable quantities, however similar the numbers look.
Timeline, consolidated:
- 10 November 2026 — No. 58 suspension expiry, equipment and technology only
- Q3 2026 — Commission preliminary third-country supply shares
- 1 January 2027 — 48E storage threshold steps from 55% to 60%
United States, cells and DC product: tariff constraint weakening, eligibility constraint hardening. Two layers came off this year. The China-specific and reciprocal IEEPA duties stopped being collected for entries from 20 February. The 10% Section 122 surcharge in force from 24 February lapsed on 24 July, with no extension located in the Federal Register, White House or CBP record as of 31 July.
| Layer, Chinese-origin cells under HTS 8507.60.00 | Rate | Status as of 31 July 2026 |
|---|---|---|
| MFN duty | 3.4% | In force |
| Section 301 | 25% | In force |
| IEEPA, China-specific and reciprocal | — | Not collected for entries from 20 Feb |
| Section 122 surcharge | 10% | Lapsed 24 July, no extension located |
| Combined ad valorem | 28.4% | Was 38.4% for five months |
That 28.4% sits before merchandise-processing fees and before classification of any complete system. It is the number now, and the direction of travel through 2026 has been down. Most landed-cost models have not caught up. Our earlier estimate of $75–100+/kWh for US-landed Chinese cells explicitly lacked a complete duty bridge. 28.4% closes that component, and the arithmetic lands at the low end of the range rather than the middle.
Duty is not the binding constraint. Under IRS Notice 2026-15 and P.L. 119-21, a Section 48E storage project must clear a material-assistance cost ratio of 55% for construction beginning in 2026 and 60% in 2027. The ratio measures the share of direct costs not attributable to prohibited foreign entities, so higher is safer. Nowhere in the text is there a categorical exclusion of Chinese-manufactured cells. Now apply the interim safe-harbor weights in Notice 2025-08:
| Object | Cell share of direct cost | Calculated ratio, prohibited-entity cells only | vs 55% (2026) | vs 60% (2027) |
|---|---|---|---|---|
| Grid-scale BESS | 52.0% | 48.0% | Fail | Fail |
| Distributed BESS | 26.9% | 73.1% | Pass | Pass |
Extend prohibited-entity scope on the grid-scale case to cells, packaging and pack/module production, and the ratio drops to 34.4%.
Three things a procurement team can act on this week:
- Grid-scale direct from China fails on eligibility, not on duty. Forfeiting 48E costs considerably more than 28.4%.
- Distributed storage is a structurally different eligibility case under the published weights, and the market is not pricing it that way.
- The arithmetic opens a third route nobody names. A US project that does not claim 48E is not subject to the ratio at all, so its Chinese-origin cell costs 28.4% flat, ten points cheaper than in June. That is the only US route where this year's tariff relief converts into anything.
Whether that third cohort carries volume is unmeasured. The categories where it plausibly sits are augmentation and replacement volume on existing sites, buyers without tax capacity, and industrial installations outside the credit structure. Treat the sizing as inference.
All of the above applies an IRS safe-harbor table rather than reporting IRS determinations, and all of it sits alongside separate ownership, control and licensing tests. Notice 2026-15 treats rights over facility output, restrictive IP terms, extended royalties and withheld know-how as potential sources of foreign influence, independent of any cost ratio.
A correction to the model. Any Q4 compound-tightening model still carrying the 31 December 2026 expiry of the graphite accommodation is aimed at the wrong regime. That accommodation belongs to the former Section 30D clean-vehicle rules, and 30D has been unavailable for vehicles acquired after 30 September 2025. No graphite-specific carve-out from the 48E ratio appears in Notice 2026-15; the BESS safe harbor treats the cell as the listed component and gives graphite no separate percentage. Inside a 48E storage calculation, anode origin does not bind ahead of cell origin. Drop the date.
European Union: Stable. A search of the Commission's trade-defence register and the Official Journal through 31 July returned no anti-dumping or anti-subsidy investigation, no provisional duty, no registration order and no definitive measure on lithium-ion cells or complete BESS. The July battery case, AD756 / C/2026/3479, covers non-rechargeable alkaline-manganese primary cells under CN 8506 10 11 and expressly excludes rechargeable lithium-ion. The existing Chinese measure, Regulation 2024/2754, covers completed battery-electric passenger vehicles.
Net Zero Industry Act resilience criteria get described widely as a European local-content rule. Articles 25 and 26 impose non-price criteria on public procurement, renewable auctions and public interventions, applicable from 30 December 2025 per the Commission's 22 July guidance. Privately procured cells face no border condition whatsoever. The number to watch is in the Commission's official supply communication: China holds exactly 50% of combined EU supply of packs, modules and cells, which sits just under the "over 50%" single-country trigger and is therefore unmarked, while anode active material sits at 81% and is marked. Preliminary updated shares are due in Q3 2026. One percentage point of measurement separates the combined battery category from resilience treatment, which makes that update an event rather than a routine publication.
FCC action sits in this force category as a live constraint on connected power electronics rather than on cells. Treated separately in this issue's sidebar.
Third-country assembly: constraint unresolved, not removed. Operating transplant capacity is much narrower than the announcement flow implies, and two sites carry the point. CATL began module assembly in its new Debrecen building on 6 May 2026 using cells shipped in from other CATL plants, with the local cell building complete but trial cell production still awaiting permits. EVE's Kulim plant produces cells, but cylindrical cells for power tools and two-wheelers rather than 280/314 Ah prismatics. Wrong object. Gemlik (Siro/Farasis modules and packs since March 2023, local cells still described as a 2026 step) and Kénitra (Gotion, AfDB €100 million approved 24 July, still financing and construction) corroborate the same pattern.
No large-format ESS cell shipment from any reviewed third-country site is publicly evidenced. Transplant assembly is real; transplant cell supply in the format that sets the price series is not. And because prohibited-entity status follows ownership, control, relevant activity and licensing rather than customs origin, moving assembly offshore can change the tariff line while leaving the eligibility line where it was.
The Tax Clocks Run in Opposite Directions
Weakening the export invoice on a dated schedule, with a four-month reversal in the middle of it.
Announcement No. 2 cut the battery VAT export rebate to 6% for exports dated 1 April through 31 December 2026 and cancels it from 1 January 2027, with the customs export declaration date controlling. The attached schedule lists HS 85076000, "lithium-ion accumulators," with no cell-only subcode, so cells and packs both fall inside it. Whether an integrated container falls inside depends on how customs classifies the exported article, and the notice prescribes no allocation among cells, racks, PCS and balance of system.
Announcement No. 20 imposes a 2% consumption tax from 1 September 2026, rising to 4% on 1 September 2027, and its scope note reaches both the basic functional unit and packs assembled from those units. Tax paid on an input is deductible when the input goes into another taxable battery. A self-produced taxable battery transferred into a non-battery product is declared at the transfer point, which does not read as an exit route for integrators. Exports keep existing consumption-tax refund and exemption treatment.
The interaction matters more than either clock alone, and it is where most models have the sign wrong. From 1 September, a domestic sale carries 2% that an export does not, tilting the margin comparison toward export by roughly two points. From 1 January, the export side surrenders six points of rebate. Export preference improves for four months, then deteriorates by six. Anyone modelling a single monotonic tightening through Q4 is reading a reversal as a continuation.
One bounded negative finding on scope. Searches of the State Taxation Administration policy library, MOF and customs materials dated 16 to 31 July found no implementing interpretation defining taxable battery value inside an integrated ESS. That is a gap in the public record, not a statement about how tax offices are treating filings.
Four Tax Bases, One Quote-Normalization Instruction
The code operates on the invoice, not on the cell. Which means the quotes landing on your desk this quarter are not comparable until normalized. At least four bases are in circulation:
- A tax-inclusive domestic cell
- An export cell or pack under HS 85076000
- A DC container classified as a battery
- An AC or separately invoiced system whose non-battery equipment carries different codes and different treatment
Nothing in the notices makes these comparable, and no public guidance allocates value inside the integrated case.
Every quote should state its HS classification, its export declaration month, whether the price is rebate-inclusive, and whether consumption tax is embedded.
Two vendors quoting an identical $/kWh on different bases can be more than eight points apart on realized economics. After 1 January that spread widens. This is a fifth layer sitting on top of the four price objects we separated in Issue #3, and it sits inside the fourth.
Whether an Export Premium Exists
Not measurably. InfoLink put 314 Ah ESS cells at roughly $50–53/kWh "outside China" in early July, without specifying origin, destination, Incoterm, duty inclusion, warranty basis or buyer class. RMB 0.360/Wh converts to approximately $50–51/kWh at 7.1–7.2 RMB/USD, offered here as a conversion basis rather than as an assessment. The two figures overlap at the bottom of the ex-China range, which suggests a thin to nonexistent cell-level premium. They also measure different objects, so the comparison cannot bear weight.
Benchmark's CIF-Europe and CIF-North-America 314 Ah LFP assessments are the correct instrument, and no current numeric value is public. China Customs reports HS 85076000 in units, kilograms and value, and the code mixes consumer, vehicle and storage batteries, so customs unit values cannot be converted to $/kWh without a mix and an energy denominator. InfoLink's European figures are system-level and will not decompose to a cell. The export premium remains inferred, not measured. One component of the landed stack is cleanly observable: Rotterdam port-to-port freight at $1.5–1.9/kWh in June for a 5 MWh two-hour DC container, up from $1.1–1.4/kWh at end-2025. Freight is moving against the buyer while cell price falls.
Two Tensions Where Inflections Will Surface
Tax withdrawal against absorption pressure. The clocks remove export margin on fixed dates. The production-absorption gap pushes volume outward irrespective of margin. Outward pressure is winning now, and September's consumption tax briefly reinforces it by penalizing the domestic alternative. 1 January is the first date on which the two forces genuinely oppose, and the resolution surfaces in whether six points get passed through, absorbed, or converted into declaration-date gaming during Q4.
The best price is foreclosed; the open route carries the thinnest margin. US grid-scale storage has the strongest realized system pricing and is credit-foreclosed to prohibited-entity cells under the published safe-harbor weights. Europe is open, with no trade remedy and no border content condition, and offers a cell premium public data cannot demonstrate exists. Distributed US storage is the anomaly: it clears on the ratio arithmetic and attracts the least attention. If this tension resolves inside two quarters, it resolves there or in the Commission's Q3 supply-share update.
Durability Through December, Repricing in January
The export price architecture holds through 31 December 2026. Nothing dated before then moves the stack materially. 10 November touches equipment and process technology rather than sub-200 Wh/kg storage cells, the EU has no live case, and the US duty stack has been easing since February rather than tightening. The repricing date is 1 January 2027, when China's export rebate goes to zero on a declaration-date test and the 48E threshold steps from 55% to 60% for projects beginning construction that year. Two independent tightenings, same morning.
The break condition, stated precisely: if battery-grade lithium carbonate holds above RMB 144,000/MT into Q4 while cell spot stays near RMB 0.360/Wh, Tier-2 exporters enter January with a rising input, a six-point revenue withdrawal, three points already absorbed since April, and no demonstrated pass-through mechanism. That cohort's last usable operating reference remains our standing 30–40% lower-tier utilization baseline set against CATL's disclosed 94.86% for H1 2026. No named Tier-2 exporter has disclosed a 2026 utilization rate or cash position, and REPT's unaudited profit alert, the only dated 2026 tier datapoint, speaks to neither. That absence is the reason the January outcome is hard to call. Watch the CABIA monthly export series for October through December. Prints materially above the June 36.2 GWh run rate confirm pull-forward and imply a Q1 2027 volume hole with softer spot. Prints on trend mean the withdrawal lands on margin, and the exits we have repeatedly noted are not happening finally get a real test.
-
Sodium-ion order book versus shipments: CATL guides to 1 GWh of cumulative 2026 sodium-ion shipments beginning with September customer deliveries, against 60 GWh contracted with HyperStrong and 5 GWh announced with Alfen for Europe — an option book until deposits, take-or-pay terms and delivered volume appear, and one that now carries a consumption-tax exemption running to end-2028.
-
GB 38031 enforcement, still text not outcome: MIIT conducted on-site conformity inspections at GAC Aion and Zhaoqing Xpeng on 24 July with vehicle and traction-battery sampling, but the published account discloses no test results, supplier exclusions or penalties; the next meaningful evidence is a homologation refusal or supplier substitution, not another inspection notice.
-
The CALB/Aion defect case: 182 complaints logged between 1 and 18 July on 177 Ah LFP cells in GAC Aion S commercial vehicles, with the warranty extended from eight years/150,000 km to eight years/300,000 km — watch whether this converts into tender exclusion and order migration rather than absorbed warranty expense, which is how consolidation shows up before any plant closes.
-
CATL's 764 GWh under construction: The H1 filing lists a construction pipeline larger than its 525 GWh of disclosed capacity, reported alongside 94.86% utilization and RMB 440.2 billion across monetary funds and trading financial assets — track it through commissioning dates and domestic/overseas split rather than adding it to nameplate.