CATL's Overseas Premium Nearly Doubled in Two Years. The Overseas Line Moved 0.32 Points.

CATL's H1 2026 interim reports gross margin of 29.97% on overseas revenue and 21.16% at home. The spread is 8.81 points. In the matched 2024 half it was 4.47. A premium that nearly doubles in two years invites the reading that the export channel got richer.

Decompose it and the reading does not survive. Overseas gross margin went from 29.65% to 29.97%. That is 0.32 points. Domestic went from 25.18% to 21.16%, a loss of 4.02. The two sum to the entire 4.34-point widening with nothing left over. Over two matched halves the overseas line is flat to within a third of a point. What moved was the thing it gets measured against.

That distinction is the whole planning question. A premium widening on firmer export pricing tells you the channel has headroom. A premium widening because the domestic comparator is eroding tells you the channel is holding a line, which has a shorter half-life and breaks in different ways.

Five observations, two of them disclosed

Matched interims give you two data points. The series has five. The intervening halves come out by subtraction: take annual overseas revenue and gross margin, net off the disclosed first half, and the residual is the second.

Implied rows derive from: FY 2024 overseas RMB 110.336bn at 29.45%, domestic 22.25%, group RMB 362.013bn; FY 2025 overseas RMB 129.641bn at 31.44%, domestic 24.00%, group RMB 423.702bn. The H1 2025 and FY 2025 links go to CATL's own English filings. H1 2026 is the mirrored Chinese-language interim and closes 30 June 2026. Operational sourcing below runs to late July 2026.

Two things to know before putting weight on the implied rows. Audited annuals get restated against unaudited interims, and any restatement lands entirely in the residual, which means entirely in the implied half. Second, the implied domestic column is the weaker of the two: the overseas figure falls out of two disclosed overseas lines, while the domestic figure requires assuming the geographic split exhausts group revenue. Lean on the overseas column. Treat the domestic column as indicative.

Read across all five and the overseas margin is flat only at matched interims. Sitting in between is an H2 2025 print near 33.6% and a give-back of roughly 3.6 points into H1 2026. Any model carrying a stable 30% overseas line is carrying the average of a series with four points of half-to-half range.

The premium itself moved in a step. It went to 8.89 points in H2 2025 and printed 8.81 in H1 2026: it widened, then held, without compounding into a third period. And across those two halves the geographies moved almost together, overseas down 3.63 and domestic down 3.55. Something common pushed both. The divergence that generates the headline shows up only when you compare like halves.

The same filing carries a movement running the other way. Overseas revenue grew 42.35% year on year against 54.80% for the group, so overseas share of group revenue fell from 34.22% to 31.46%. The rich geography got relatively smaller in exactly the period its margin advantage got wider. Weight the disclosed geographic margins by the disclosed geographic shares and blended gross margin runs approximately 26.5% in H1 2024, 25.0% in H1 2025, 23.9% in H1 2026. That arithmetic is mine rather than a filing line, and it is approximate. The direction carries the point: the premium widened and group margin fell anyway, because domestic is simultaneously the thinner margin and the growing weight.

What the segment note means by overseas

CATL attributes external-customer revenue to the location where goods are delivered or services provided. Not where they were manufactured, not shipping origin, not the legal entity that books the sale.

Follow the definition through and 29.97% contains at least three cost structures:

• Cells made in Chinese plants and delivered to European customers.
• Cells made at CATL's Arnstadt plant in Germany and delivered to European customers.
• Modules assembled in Debrecen, Hungary, from cells shipped in from other CATL plants, delivered to European customers.

A fourth is coming. Output from the CATL–Stellantis venture at Zaragoza will land in the same line, and the FY 2025 report states that CATL consolidates Contemporary Star Energy in full despite holding 50% of the economics, on the grounds that it controls daily operations and holds decisive board voting. Full consolidation puts 100% of that plant's revenue and 100% of its cost through the overseas gross margin line while half the resulting profit belongs to somebody else. As Zaragoza ramps, reported overseas margin degrades as a proxy for CATL's own economics, and nothing in the geographic table flags the divergence.

Nowhere in the filings is overseas revenue split between China-origin export sales and locally produced sales, and there is no cross-tabulation by plant, by entity, or by production origin.

This governs everything downstream. Localization changes the composition inside the high-margin bucket without changing the label on it. No revenue moves from a rich line to a thin one, where you could watch the transfer happen in the tables. The rich line is being rebuilt from the inside.

Five candidates, three that bear weight

Five explanations are on the table. Two fail on evidential grounds before we start.

Pricing power arising from a thinner field of qualified suppliers abroad is plausible and unmeasurable here. CATL discloses that its largest customer took 13.73% of FY 2025 revenue and its top five 38.96%, with no geographic breakdown. Nothing in the filings connects concentration to the spread, so the hypothesis cannot be sized. It cannot even be signed. Shipping terms and freight incidence have no disclosed geographic cost bridge at all. Cut both.

Three remain, ordered by how much weight the disclosures will actually bear.

Export versus owned-plant cost blend. Established by the geography definition, unquantified by the filing, and carrying the largest forward significance of the three. Its current state is what the next two sections are about.

Product mix. CATL discloses margins by product line for H1 2026.

No cross-tab with geography, and the arithmetic caps how much mix can explain. 29.97% sits above every one of those segment margins, and no weighting of numbers below a ceiling gets you above it. Mix alone cannot generate the premium. Either overseas margin runs above the line average within each product line, or the unbroken-out "other business" line is carrying the gap. That line booked RMB 12.720bn of revenue against RMB 3.938bn of cost, an implied 69.04% gross margin, composition undisclosed. Mix contributes something. Its ceiling is lower than a quick read of the segment table suggests.

Contract repricing tenor. CATL's Hong Kong prospectus confirms price-adjustment mechanisms tied to lithium carbonate and other major materials. It does not disclose whether reset frequency or realized pass-through lag differ by geography. If overseas contracts reprice more slowly, falling input costs widen the overseas margin mechanically and rising costs close it. Directionally important, entirely unsized, and the most obvious candidate for the parallel half-to-half swing described above.

The charge exports carry and domestic sales do not

Domestic sales put no VAT through the profit and loss account. Output tax goes to the customer, input tax is credited, cost of sales never sees it. Exports are where a P&L charge appears, and it appears in exactly the place that matters here.

Under the Ministry of Finance's VAT accounting rules, input VAT that is not refunded on export is debited to main-business cost. The State Taxation Administration's calculation guidance sets that amount as export FOB value multiplied by the difference between the applicable VAT rate and the export rebate rate. The rebate is not revenue and not other income; the shortfall is cost of goods sold.

So the export component of the overseas line already absorbs a charge the domestic line never carries. The premium exists in spite of it, which means underlying overseas pricing power is better than 29.97% represents, by an amount CATL does not disclose.

That amount has a published escalation path.

The non-refundable column assumes the standard 13% VAT rate applies to inputs. That assumption is mine, not a CATL disclosure.

Bound the 2027 event rather than guess at it. Take 100% China-origin FOB, FOB approximating booked revenue, zero pass-through to customers: removing the remaining 6% adds six points of FOB value to cost of sales, so roughly six points of overseas gross margin is the arithmetic ceiling. The realized figure sits strictly below it on three counts: Arnstadt and Debrecen output is not a Chinese export, FOB value is not booked revenue, and CATL's commodity-linked adjustment mechanisms establish that at least some contracts reprice.

Run the same ceiling backwards into H1 2026 and it yields something more useful. The April step raised the charge by three points of FOB across three of six months, so at maximum incidence it cost roughly 1.5 points of overseas gross margin on the half. Overseas margin fell 3.63 points from the implied H2 2025 level. The rebate cut accounts for at most 1.5 of those 3.63, and the domestic line, which carries no such charge, fell 3.55 over the same interval. Most of the give-back was not the tax.

In Four Floors, Not One I treated the export tax layer as a cost the buyer absorbs on the way to a delivered price. The accounting says the seller absorbs it too, inside reported gross margin, whether or not the buyer's landed cost moves at all.

Three modes, and what has actually been built

The barrier regime decides which mode is available in which market, and each mode carries its own margin signature. Export earns the premium and holds the tax exposure. Owned plants buy access at a cost structure nobody outside can see. Licensing converts capital into a royalty line that cannot be measured from outside the contract.

Arnstadt runs. The municipal government confirmed in March 2026 that the site operates integrated cell production, module stacking, testing and validation. Named-source reporting puts its cells in Volkswagen PPE-platform vehicles, the Porsche Macan and Audi Q6 e-tron among them. Nameplate target 14 GWh. Actual output, utilization and plant-level margin: undisclosed.

Debrecen is where the announcement and the plant come apart. As of 1 July 2026, the first cell-production unit was physically complete and had not received the permits required to begin trial cell production. Trial, not commercial. Intended cell capacity is reported locally at 40 GWh. What operates in Debrecen today is module assembly, running since autumn 2024 in a rented hall, with a further 5 GWh line opened in May 2026, together producing roughly 240,000 modules for about 60,000 European vehicles from cells shipped in from other CATL plants. Assembling modules from imported cells and manufacturing cells are different industrial activities, with different cost structures and different regulatory treatment. Phase 2 has carried no disclosed commissioning timetable since CATL said in mid-2025 that it was reviewing technology and schedule. Zaragoza began construction in November 2025 against an end-2026 production target; Aragón's final project approval in late July 2026 described first production phases as starting in early 2027. The targets differ. Both sources agree it is not producing.

Licensing is where disclosure stops. CATL describes the Ford arrangement as license, royalty and service: Ford owns and operates the Michigan LFP plant, CATL supplies technology access, training and quality practice. Neither party discloses the royalty basis, whether struck on installed capacity, cells produced, cells shipped or revenue, along with minimum payments, service fees, equipment sales, contract duration, and any CATL revenue recognized. Ford said in June 2026 that the plant had assembled its first full prismatic LFP cells end to end and moved past initial pre-production, with 2026 shipments still targeted and no commercial customer shipments reported.

The standard assumption is that licensing throws off near-zero-cost revenue and therefore flatters the overseas margin. The record will not carry it. What the record carries is a bundled license-plus-services structure with an undisclosed revenue basis and undisclosed associated cost. The 69%-margin "other business" line is where such revenue would plausibly sit. CATL attributes none of it there, or anywhere else.

The depreciation drag that has not started

The H1 2026 filing still books the Hungarian project as construction in progress: RMB 2.568bn added in the half, RMB 12.374bn cumulative, roughly USD 1.7bn at RMB 7.2 to the dollar, a scale conversion and nothing more.

Assets under construction do not depreciate. That RMB 12.374bn is not yet moving through cost of sales, which puts the Debrecen ramp drag outside the 29.97% entirely. It arrives when cell lines commission and a large fixed-asset base begins amortizing against low initial utilization and output that has not yet cleared customer qualification. When it arrives it lands inside the overseas revenue line, undifferentiated, and in the headline it will be indistinguishable from a decline in export pricing.

That is the mechanism by which the premium liquidates itself. No revenue migrates out of the high-margin geography. The high-margin geography absorbs the cost of its own localization, with no disclosure that isolates the effect.

What the number is good for

Three adjustments for anyone taking the overseas premium into a planning meeting.

1. It is not evidence that export pricing is strengthening. Across matched interims the overseas margin is flat to within a third of a point, and the implied half-year path shows a peak and a give-back that the interim comparison hides completely.

2. It is not a China-export number. It is a delivered-abroad number, already blending Chinese exports, German cell output and Hungarian module assembly, with no disclosed split.

3. Expect it to get less informative rather than more. Three forces degrade its content at once:

   • localization drag arriving out of construction in progress;
   • the rebate charge tripling on a fixed date;
   • full consolidation of a half-owned Spanish venture whose revenue and cost enter the line in full while its profit does not.

If 29.97% still prints through 2027, that will not mean nothing moved. It will mean something offset something, and the filing will not tell you what.