Layer 1 — Trajectory
Down, and for the first time this July: InfoLink's 280Ah/314Ah LFP ESS-cell average left the RMB0.370/Wh plateau to the downside, 2.7% WoW from RMB0.370/Wh on 22 July to RMB0.360/Wh on 29 July, while the same house marked battery-grade lithium carbonate up 2.1% WoW to RMB144,000/MT and SMM's 31 July tape stayed put.
Availability moved. The clearing level did not. A large buyer negotiating Q4 tonnage can currently pull roughly RMB0.010/Wh out of a quote, and that is the whole of what the print measures.
Layer 2 — Application Grid
EV cells. Higher on the week, higher across the month. Both prismatic LFP formats assessed at $50.90/kWh on 10 July (SMM); on 31 July they sit at $52.10 and $52.40/kWh.
| SMM weekly assessment, 31 July | $/kWh | WoW |
|---|---|---|
| Prismatic LFP, 100Ah | 52.10 | +0.60 |
| Prismatic LFP, 174Ah | 52.40 | +0.80 |
| Prismatic NMC 5-series | 82.00 | +1.00 |
| Prismatic NMC 6-series | 81.50 | +0.90 |
| Prismatic NMC 8-series | 94.00 | +1.00 |
Assessment is weekly, published Fridays at 12:00 Beijing; see the correction below. The EV tape and the ESS tape pointed opposite ways this week, which is reason enough not to read the BESS headline across.
BESS cells. Two sources, two directions.
| Source, date | Format | RMB/Wh | WoW | Reported range |
|---|---|---|---|---|
| InfoLink, 29 July | 280Ah | 0.360 | −2.7% | 0.340–0.390 |
| InfoLink, 29 July | 314Ah | 0.360 | −2.7% | 0.340–0.380 |
| SMM, 31 July | 280Ah | 0.367 | +0.002 | — |
| SMM, 31 July | 314Ah | 0.367 | flat | — |
Observed band RMB0.360–0.367/Wh. At CNY 7.20/USD that is roughly $50.0–51.0/kWh, and the divisor is mine rather than anyone's assessment, so run your own rate against RMB360–367/kWh. The source itself shows what skipping that step costs: SMM's own USD line prints both formats at $48.00/kWh, which back-solves against its RMB figure to a divisor near 7.65. Its two tables therefore sit on different FX or different tax bases. Do not put both inside one model.
No blended number from me. InfoLink samples after-tax transactions across roughly twenty leading manufacturers. SMM's 280Ah object is a delivered cell, VAT included, on one-month payment terms, sampled from Tier 1 through Tier 3. Different counterparty populations, different tenor, no matched delivery month. Averaging them would destroy the only useful thing on the table, which is the gap.
Defense-relevant formats. No update this week. Nothing application-specific surfaced in public $/Wh or $/kWh for small-format UAV, UUV or 6T cells. Checked: DLA QPL-32565, which lists no qualified products and no qualified sources; and the SAM.gov small-UAS accessories opportunity updated 22 July, which exposes neither cell nor pack price. Procurement is presumably running, out of public view. This segment has not been priceable from public data at any point in our tracking, and it was not this week.
Pack level. SMM China, 31 July:
- 51.9kWh LFP pack — $75.20/kWh
- 18.3kWh LFP pack — $71.90/kWh
- 42.8kWh 6-series ternary pack — $94.80/kWh
No WoW field was recoverable for the pack lines this pass. Take them as levels with no period-stamped rate attached. Cross-region calibration is still BNEF's December 2025 survey: $108/kWh global average, $70/kWh stationary storage, $99/kWh BEV.
Cadence correction. SMM's Chinese-language product notes specify the EV-cell, ESS-cell and 5MWh DC-cabin series as weekly, published Fridays at 12:00 Beijing. Earlier issues of this section described the change fields as daily. The levels were right and the change values were right; the period stamp was not. That error runs in one direction only: a weekly move read as a daily move overstates implied velocity by roughly a factor of five. If you annualized any SMM cell change out of our issues before this one, reset the period and keep the number.
Layer 3 — The Synthesis
Nothing in the cost stack paid for this
Two issues ago I set out three things that had to clear in sequence before lithium weakness reached a cell quote: inventory buffers, then cell-maker pass-through, then integrator margin compression. None had cleared. Last week I hardened that from a lag into structural impairment. This week qualifies the position instead of confirming it, and the qualification is the more useful of the two outcomes for anyone with a Q4 decision in front of them.
The first downstream concession since the RMB0.370/Wh plateau formed arrived alongside a rising input, in the same publication, in the same week. Whatever bought RMB0.010/Wh, cost relief did not. InfoLink puts the move on selected specifications and orders, concentrated in long-term agreements and large-customer negotiations, with manufacturers competing for Q4 volume and utilization.
The mechanism sets the half-life. Cost-driven movement is general and durable because it resets every producer's floor simultaneously. A purchase of order book is discretionary and tier-specific: individual manufacturers filling individual capacity gaps, ending when the gap fills or the quarter does. Both produce the same print. Only one of them is still in the market in Q1.
A bounce inside a decline
The +2.1% WoW print is the second leg of a two-week sequence, and the sequence carries more information than the print. InfoLink's carbonate benchmark fell 7.2% WoW in the prior assessment, then rose 2.1% WoW to RMB144,000/MT on 29 July. The retracement recovered under a third of the drop, the two-week net is still negative, and the rate of increase is not building. On this data I call it a bounce inside a decline rather than a turn. One week of amplitude is not a Q4 cost floor.
No domestic-versus-CIF spread this week, at least not a clean one. Fastmarkets' MB-LI-0029 CIF China/Japan/Korea assessment was active and current on 27 July, but no July assessed value was publicly recoverable, and the 2 July publication-delay notice names the series and withholds the price. So both legs come off SMM, 31 July: CIF CJK battery-grade at $18.58/kg, or $18,580/t, against domestic battery-grade at $18,693.67/t ex-VAT. Domestic sits about $114/t over seaborne, roughly 0.6%. On an SMM-to-SMM basis the spread is effectively shut, with a thin domestic premium on top. Neither leg carries a recoverable period-stamped change in this pass, so whether that premium widened or narrowed WoW I cannot tell you. And this is a same-provider comparison, not a cross-provider spread. I am not going to dress it up as one.
One further basis caution, the same one I applied to SMM's cell tables. InfoLink's RMB144,000/MT and SMM's $18,693.67/t are not the same object: the SMM domestic assessment is explicitly ex-VAT, and InfoLink's public note states no tax basis at all. They will not cross-read cleanly, and the difference between them is not a market spread.
The impairment is symmetric
This is where most models will go wrong. Pass-through impaired downward is pass-through impaired upward, by the identical mechanism. So do not assume carbonate at RMB144,000/MT reaches cell quotes on the historical cathode-inventory lag of roughly six weeks. Whatever stood between June's cost relief and June's quotes is still standing between a July cost increase and an August one.
For anyone floating cell price against a lithium index, basis risk widened in both directions this week. Fixed-price Q4 tonnage struck at RMB0.360/Wh, with a hard validity date on it, is the cleaner instrument right now.
Breadth, not velocity, resolves this
Velocity tells you how fast a number is moving and nothing about who is paying it. It cannot separate a floor that dropped from a subset of buyers being paid to commit. Breadth does that: how far a concession travels across counterparty tiers, specifications and delivery windows. Four observables settle the question inside 60 to 90 days. Two of them you can check in public data. Two you cannot, and that second pair sits exactly where your own quote flow beats any published assessment, this one included.
Publicly checkable:
- Whether SMM's Friday tape follows. Two days after InfoLink cut, SMM had 314Ah flat and 280Ah up RMB0.002/Wh. If the concession is broadening, SMM's average should track down within two to three assessments. Checkpoints: 7, 14 and 21 August. Still flat-to-up on 21 August, and the discount never left the large-account channel.
- Which tail of InfoLink's range moves. Watch the tails, not the average. Both formats floor at RMB0.340/Wh today, with ceilings at RMB0.390 and RMB0.380. A break below RMB0.340 while the top holds means the discount is going deeper to accounts that already held the best price, which is channel deepening. A ceiling compressing toward the average while RMB0.340 holds means the concession has reached the buyers who were paying up. That is repricing.
Not publicly observable, worth logging internally:
- Quote validity periods. Utilization buying comes with short windows. A concession that holds for thirty days is a different animal from one that lapses inside a week.
- Deposit and indexation terms. RMB0.010/Wh offset by heavier prepayment, or handed back through a lithium-indexed escalator, is a financing cost booked as a price cut. No published assessment table shows you this.
Annual calibration, systems, and the US number that still isn't there
Set the week against the annual curve. BNEF's survey moved from $115/kWh in December 2024 to $108/kWh in December 2025. BNEF calls that −8% YoY; the two published levels difference to roughly −6%. Take either, then put it beside the −20% YoY BNEF reported in the 2024 survey: the annual rate of decline shed around two-thirds of itself in one cycle. Every weekly move above sits inside that deceleration. If the EV tape holds its current direction through Q4, the 2026 survey has a plausible route to a print near zero. That last sentence is my assessment, not BNEF's.
One adjacent tender at the system layer. Huaneng Ningxia's 80MW/320MWh equipment package went at RMB159.04 million, or RMB0.497/Wh. Warranty terms, acceptance criteria and full scope are absent from the public record, so the number cannot be normalized against any other tender. It does establish one thing: system pricing keeps moving on its own logic, not in step with the cell tape. InfoLink's 29 July system prints say the same. The two-hour DC average slipped 1.0% WoW to RMB0.490/Wh while the one-, two- and four-hour AC averages held flat at RMB0.840, RMB0.580 and RMB0.530/Wh, with the stickiness attributed to PCS, transformers, EMS, thermal management, fire protection, grid connection, commissioning and warranty rather than to cells.
Still no public US buyer-side equipment $/kWh with enough disclosed scope to set against China. DOE's 22 June Wisconsin Build America waiver puts approximately $1.87 million against 16kWh LFP systems across 29 microgrid sites, then withholds unit count, aggregate kWh, and the split between batteries and controls. No $/kWh comes out of that. CPS Energy's up-to-20MW BESS proposals are due 7 August. Illinois' storage bids land 26 August, with a confidential results report on 28 August. Two dated windows in which a US object could become visible. Neither has opened.
-
Fastmarkets respecs the CIF benchmark: Effective 1 September, Fastmarkets will change the quality and tonnage specifications of its battery-grade CIF CJK lithium salts assessments after feedback from more than 50 companies, which means any seaborne-versus-domestic spread series you maintain will contain a basis break at that date.
-
A hedging venue for the CIF leg: CME plans to launch a lithium carbonate CIF CJK options contract on 17 August, settled against the monthly average of the Fastmarkets assessment, with BESS buyers named as a target constituency — worth watching alongside the September respec, since the settlement reference is the series being changed.
-
The tender bid distribution: China Huadian's 12GWh framework procurement drew 59 bidders at RMB0.4923–0.6207/Wh, averaging about RMB0.54/Wh with a five-year warranty, which places the Huaneng Ningxia RMB0.497/Wh award near the aggressive tail of an observed distribution rather than outside it — different scope, so context only, not normalization.
-
A new certification barrier on the non-cell stack: The FCC has barred approval of new foreign-made power inverter models, with previously approved models apparently still able to enter the market — a discontinuity that sits in the exact layer of a US-landed system that cell pricing cannot reach.

