Only One of the Three Korean Q2 Improvements Is a Number You Can Use

All three Korean cell makers posted sequential improvement on July 30, and only one of the three numbers survives contact with a model. LG Energy Solution disclosed its US production credit at the same layer where it books profit, which produces a point estimate of underlying operating profit. Samsung SDI disclosed its credit one layer above the segment a BESS buyer actually contracts with, which produces an upper bound and nothing underneath it. SK On disclosed a large headline reversal built on at least two unquantified non-recurring items, which leaves the sign of recurring operating profit open rather than merely its magnitude. Rank the three on Q2 profitability and you have quietly asserted that a point estimate, a one-sided bound and an undetermined sign are the same measurement.

USD equivalents use KRW 1,500.48 per dollar throughout, the average of the 63 published business-day observations in the Federal Reserve H.10 series between April 1 and June 30, 2026, applied to the reporting quarter only. Where a Korean issuer writes AMPC, read 45X. Same statutory credit, issuer naming convention.

LG Energy Solution, the only computable bridge

ᵃ The issuer disclosed this in the relevant quarter's release; we have not carried it into this cycle's verified source set. Q1 2026 is fully sourced. That is a retrieval limit on our side rather than a disclosure gap, which is why it gets its own marker. The ex-incentive row is a maintained series, established in our Q2 preliminary read, and basis-consistent across all four columns.

ᵇ Not disclosed by the issuer in any of the trailing four quarters.

45X isolation. The Q2 release puts North American production incentives at KRW 241.0B against reported operating profit of KRW 113.3B. Strip the incentive and Q2 was an operating loss of KRW 127.7B, or 1.69% of disclosed revenue. One number, no range. LGES is the only one of the three that hands it over.

What matters is the shape of the series, not the level. The ex-incentive loss narrowed KRW 57.2B between Q4 2025 and Q1 2026, then KRW 269.9B between Q1 and Q2: the first derivative was already positive, and the rate of improvement accelerated from KRW 57.2B QoQ to KRW 269.9B QoQ. Run the credit as coverage of the ex-credit hole and you get 73% in Q4 2025, 48% in Q1 2026, 189% in Q2. All three points come off disclosed incentive figures. The credit covered less than half the underlying hole two quarters ago and nearly twice it now, and the move came from the hole closing, not the credit growing.

Our preliminary read framed the risk as a credit bridge shortening faster than the gap it had to span. On the final filing the sequential arithmetic runs the other direction, so that framing was wrong on the quarter and we are retiring it. The credit line is still down roughly half against the year-ago quarter on LGES's own disclosed figures, which remains the forward constraint. But Q2 is an operating improvement, not a subsidy artifact.

Utilization trend. No company-wide figure, fourth consecutive quarter. The release attributes the improvement to higher European pouch utilization, a richer cylindrical mix, and better fixed-cost absorption as North American ESS output rose. SNE Research's January–May data has LGES at 35.0 GWh of non-China EV installations, up 1.0% YoY into a market up 21.8%, share down from 20.2% to 16.7%. That series excludes ESS, stops before June, and counts cells sitting in registered vehicles rather than cells shipped, so it cannot decompose Q2 revenue. It can bound the inference, and the bound earns its keep: KRW 269.9B of sequential underlying improvement did not come from EV volume. It came from ESS ramp, mix and fixed-cost dilution, three inputs with three different durability profiles that should not be modeled as one improvement.

Guidance versus actuals. Full-year revenue growth of 20% or more against 2025 was reaffirmed, per The Elec's account of the Korean-language call; our reading runs through that report and the English deck, not a primary transcript. H2 ESS production is guided at not less than twice H1, with ESS profitability excluding IRA incentives targeted for Q4 once five production sites stabilize. That is the one dated, falsifiable claim any of the three made on July 30, and it is denominated in unsubsidized economics.

Capex is tracking the January plan of a 40%-plus cut versus 2025, at approximately KRW 1T in Q2, roughly 40% below Q1, with no revised annual figure. On orders, January guidance set a 2026 target above 90 GWh of new ESS orders; the July release reports more than KRW 3T of H1 ESS orders. Energy in the target, won in the progress report, and no disclosed ESS contract price to bridge them from public materials. The 90 GWh figure was not repeated, not revised, not withdrawn. H1 progress against it is unverified, which is a different status from on track.

Cash and working capital. The Q2 deck shows KRW 25B of operating cash flow against KRW 1.330T of EBITDA. Working capital took KRW 1.185T. Inventory went from KRW 5.354T to KRW 6.445T, gross debt from KRW 24.682T to KRW 29.123T. The positive KRW 1.788T investing line is net, and its principal identified inflow is the sale by the Honda–LGES joint venture L-H Battery Company of all buildings and building-related assets, land and equipment excluded, to Honda Development and Manufacturing of America, valued in the corrected KRX filing at approximately KRW 3.74T on a final book value of USD 2.530B, closed May 25. On the call, more than KRW 3T of those proceeds sits inside roughly KRW 4.7T of quarterly inflows against about KRW 1T of capex. The remaining investing lines are not publicly reconciled. Liquidity improved; cash conversion did not. None of the KRW 1.788T is recurring.

We built one inference here and then killed it. A rising credit line alongside a KRW 1.091T inventory build invites the reading that credits accrued on unsold output. Section 45X requires an eligible component to be produced and sold; production alone earns nothing. But the final regulations allow a related-party election under which a sale to a related person qualifies, so downstream inventory can sit perfectly legitimately alongside earned credits. Settling which case applies would take the entities, the election, the transaction and the inventory location. None of it is in the Q2 materials. The two lines moving together is evidence of nothing, in either direction.

Samsung SDI, disclosed one layer above the segment you buy from

ᵃ Not carried in this cycle's verified source set. Q1 as reported in our Q1 decomposition.

ᵇ Not disclosed in the reviewed filings. The segment AMPC allocation and the tariff refund are both named as profit contributors and neither is given a number, which is precisely why the battery-segment result cannot be normalized.

45X isolation, bounded on one side. The July 30 KRX preliminary filing (연결재무제표 기준 영업(잠정)실적, public disclosure) states that consolidated operating profit includes KRW 107.7B of AMPC under Section 45X. Net of it, consolidated operating profit is KRW 96.1B, a KRW 332.2B swing off Q1's ex-AMPC loss of KRW 236.1B.

Then the arithmetic stops. The Korean issuer release attributes battery-segment operating profit of KRW 159.3B to high-value product sales, AMPC on increased US production, and a refund of previously paid reciprocal tariffs, and quantifies neither the AMPC nor the refund at segment level. What the documents will support: consolidated operating profit excluding both AMPC and the refund is not greater than KRW 96.1B, with no lower bound established. What they will not support in any form is the same calculation on the KRW 159.3B battery figure, because the KRW 107.7B cannot be pushed down into that segment from public materials.

Korean-language primaries checked: the July 30 KRX filing, the Korean release, the Korean IR presentation. The English deck carries the same segment result with no allocation. Samsung's business-report archive had not posted a 2026 half-year report as of July 31, leaving the Q1 report as the newest periodic filing. This refines the Q1 finding rather than restating it: Samsung does disclose its 45X credit, and discloses it in Korean before English, but the disclosure sits at the consolidated layer while the buying decision sits at the battery layer.

Utilization trend. The release says utilization improved. No percentage, no denominator. Nothing in the July materials restates the earlier trade-press expectation of Hungary above 70% in H2, and no replacement forecast was issued. SNE has Samsung SDI at 8.7 GWh of non-China EV installations for January–May, down 29.7% YoY, the steepest fall of the three. A battery segment that turns profitable while installed EV volume drops by nearly a third is being carried by something other than EV cells: high-power small cells, ESS, the credit, and a tariff refund, in proportions the filings decline to give.

Guidance versus actuals. Management expects further H2 improvement on substantially expanded ESS sales, new US lines with associated AMPC, and full utilization of high-power small-cell capacity, with EV-battery losses narrowing and reduced volumes on existing vehicle programs flagged as a continuing risk, per The Elec's call report. No numeric H2 utilization or volume target. In October there will be nothing to check it against.

SK On, a reversal with three unquantified inputs

ᵃ Not carried in this cycle's verified source set.

ᵇ Not disclosed in the reviewed filings. Three of the four rows needed to normalize Q2 are blank, and that is the finding.

ᶜ No comparable basis exists. Q1 revenue was restated from KRW 1,791.2B onto a new credit-presentation basis, and the reviewed materials do not carry that restatement back further, so a basis-consistent revenue series does not survive past Q1 in public form.

45X isolation, not attempted by the issuer. The SK Innovation deck reports battery-business revenue of KRW 2.946T and operating profit of KRW 821.8B, against the Q1 battery-business loss of KRW 349.2B first read here in Issue #4. Call remarks name three contributors: higher Asian sales, customer compensation, increased AMPC. Our reading of those Korean-language remarks runs through that report and the deck rather than a primary transcript. None of the three carries a number: no counterparty, no affected program, no characterization of cash versus receivable, no separate credit figure.

Two mechanical cautions before anyone compares this to a prior quarter. From Q2 the company changed presentation: US production credits formerly shown as other operating revenue now sit across revenue and other operating revenue where they affect customer price or transaction terms, with no stated effect on operating profit, net income or net assets. Q1 revenue is restated onto that basis, so KRW 1.870T to KRW 2.946T is basis-consistent and the originally reported KRW 1.791T is not. Separately, the same deck footnotes KRW 1.827T of operating profit for the merged SK On legal entity, a perimeter that also picks up SK Trading International and SK Entum. That is not a second reading of battery-business profit and it is not comparable with the Q1 segment loss.

Hana Securities puts Q2 AMPC at KRW 130B and roughly KRW 1.1T of combined tariff refunds and customer-compensation receipts. Strip both and the residual is negative KRW 408.2B. That is sell-side model arithmetic on estimated inputs, one of which fuses two distinct items, and it is not an issuer disclosure; it does not belong in a supplier file as a normalized result. What it does supply is one externally modeled negative case. On issuer evidence alone, whether recurring battery-business operating profit in Q2 was positive or negative is undetermined.

Utilization trend. No figure. SNE has SK On at 15.8 GWh of non-China EV installations for January–May, down 5.7% YoY, against revenue up 57.5% QoQ on the restated basis. Volume down, revenue sharply up. That pattern is consistent with the credit reclassification, with the one-off receipts, or with an Asian mix shift, and the filings do not separate them.

Guidance versus actuals. North American volumes are expected to recover in H2 on California incentives and recent demand; European negotiations are described as focused on stabilizing volumes and pricing and moving toward higher-profit programs. No numeric target. The single quantified forward item is structural cost: approximately KRW 300B of annual depreciation and KRW 200B of annual interest expense expected to come out following the BlueOvalSK reorganization, with full effect once the revised structure is implemented. Balance-sheet relief, not demand.

What this leaves for a sourcing file

None of the three published a 2026 company-wide utilization rate. The standard haircut, announced FEOC-compliant capacity discounted by trailing realized utilization, therefore cannot be run on current data at all. The 2025 baselines flagged in "Missing Denominators" (Issue #4) are the only denominators available, and those come from calendar-2025 annual reports published in March, so the underlying observations are two to five quarters old against a Q2 2026 decision. Any FEOC-compliant Korean supply figure quoted to you this quarter is announced capacity wearing a stale haircut or none at all.

Nor did any of the three give a cell ASP or an ASP direction. A quarter of synchronized profit improvement arrives with no primary price evidence whatsoever. The only ASP signal on offer is indirect, revenue against SNE installation volumes, and for SK On even that is unreadable because the presentation change pushed credits into the revenue line. EV/ESS revenue mix is undisclosed at all three; LGES's qualitative attribution to ESS ramp is the entire mix evidence for the quarter. A four-quarter ex-credit series can be built for LGES and cannot be built for the other two from currently accessible materials, and in SK On's case the presentation change means the comparable basis does not extend backward past Q1 even in principle.

LGES. Underlying operating economics improved materially, and the improvement is measurable, which is not true of the other two. The gap that remains is route-level: nothing in the Q2 materials joins accepted ESS gigawatt-hours to a facility, a product format, a customer and realized economics, so the KRW 127.7B ex-incentive loss cannot be attributed to the production route you would actually be buying from. The Q4 ESS-profitability-excluding-IRA claim is the test that resolves it. Until then, containment means qualifying to a named line and format rather than to a plant, and tying acceptance milestones to output from that line.

Samsung SDI. The battery segment turned profitable with installed EV volume down nearly 30%, and two profit-positive items inside it are unquantified at segment level, one of them, the tariff refund, non-recurring by nature. The missing facts are the segment credit allocation and the refund's value. The prismatic position is a genuine structural differentiator for BESS buyers who specify prismatic and a weak one for the many US integrators that remain format-agnostic; price it as an option, not a moat. Containment here is commercial structure: separate credit pass-through from base cell price so that a change in credit accrual or the lapse of a one-time refund does not reprice the cell.

SK On. The reversal is large and its recurring component is of unknown sign, which is a categorically different input from a measured loss. Hyundai-anchored North American volume is still the operational fact carrying the most weight, and the quantified forward items are depreciation and interest relief rather than orders. Where the sign of underlying profit is undetermined, financial-strength assessment stops bearing load and the mechanisms have to: capacity reservation against milestone payments, a second source qualified in parallel, tooling ownership and step-in rights, contracted volume-flexibility bands.

The inability to decompose a supplier's financials is itself a risk input, and it prices differently at each layer. LGES's opacity affects route attribution and comes with a dated test attached: Q4, ESS profitability excluding incentives. Samsung's affects segment normalization and has a nearer one in the 2026 half-year business report, unposted as of July 31; the Q1 report carried no segment allocation, so treat it as a check rather than an expectation. SK On's affects the sign, and nothing on any disclosure calendar is scheduled to close it.