Better emissions data would cut the modelled CBAM charge by €3.5 billion, but not for every exporter

The result varies sharply by country and product. For Russia, country averages reduce an €893 million modelled steel charge by only €10 million. India records a €231 million net reduction, but that national total hides opposite effects: the country average raises the charge on its main flat-steel flows while reducing it on ferro-alloys and stainless products.

The cost of relying on default values

CBAM requires importers to surrender certificates for the emissions embedded in imported goods. The charge is reduced by the free allocation an equivalent EU producer would receive under the relevant benchmark. If verified emissions data is unavailable, the importer must use an EU country default value instead.

Those defaults carry a mark-up — 10% in 2026, rising to 30% by 2028 for steel, aluminium, cement and hydrogen, and a flat 1% for fertilisers. The mark-up is intended to make defaults conservative and to encourage verified reporting. It does not guarantee that verified data will reduce the modelled charge for every producer, as the Indian flat-product flows below show.

The table shows the modelled charge if every consignment is assessed using default values. Read it as a default-value scenario, not as a forecast of revenue or as a maximum possible charge for every producer.

Aluminium is small because the Regulation currently counts only direct emissions, and an aluminium smelter’s emissions are overwhelmingly in the electricity it buys. If indirect emissions are ever brought into scope, this is the line that moves most.

Where better data reduces the modelled charge

Run the same calculation against published country-level emissions intensities — the European Commission’s own figures, on the same direct-emissions boundary the Regulation uses — and the difference estimates where better data would matter most. These are industry averages for each country, not verified figures for any particular consignment.

National totals can conceal large differences between product groups. India is the clearest example: the country average raises the modelled charge on its main flat-steel flows while reducing it on several ferro-alloy and stainless-steel codes.

A country total can hide opposite effects within it. Across all of India’s codes the country average reduces the modelled charge by €269M and raises it by €38M, netting €231M — but the €38M falls on the carbon-steel flat products that make up most of the tonnage, while about half the reduction comes from two codes, ferro-alloys and stainless bar, where the default sits far above the published country value. An exporter’s answer depends on what it ships, not on its national average.

Why the earlier estimate was revised

An earlier version of this work estimated the value of verified emissions data at about €8.8 billion for the same group of steel flows. Using published country-level intensities reduces that estimate to €3.5 billion. What caused the revision matters more than its size.

The earlier comparison scaled each plant’s activity by its production route. The method could estimate the scale of a plant, but it could not reliably distinguish emissions performance between countries: each production route used the same global emissions factor, so every blast-furnace works on earth carried the same number. Most of the apparent country differences simply reflected how much steel each country made in electric-arc furnaces.

Replacing it with published country intensities changed both the total and its shape. Two country figures moved far enough to reverse their reading: India fell from €951M to €231M and Russia from €678M to €10M. An exporter in either country acting on the earlier figure would have received the wrong advice.

The activity calibration remained sound, and independently so: scaling plant activity to measured emissions gives a blast-furnace ratio of 0.99 in Australia, 0.98 in the United States and 0.96 in Canada — three regulators, three sets of rules, one answer. The problem was the use of a single global emissions factor for each production route.

From country averages to individual plants

Country averages show where better data may matter at a market level, but an importer’s actual liability depends on the facility that produced the goods. The next question is therefore whether individual plants can be identified, and whether anyone has published their emissions.

We hold 1,830 located plants across steel, aluminium and ammonia. Fifty-two of them carry an intensity that a regulator has published. The other 1,778 carry identity, ownership, production route, scale and a bounded estimate.

China contributes 748 plants to the layer with no measured value attached to any of them; India 133, Russia 62, Türkiye 32. Published plant-level data is concentrated in jurisdictions that account for only a small share of the modelled charge — 1.0% as the layer stands. Merging the three datasets we have already obtained but not yet matched to plant identity, from the United States, Canada and Taiwan, would take that to 5.3%. The remaining €13.8 billion is associated with origins where our dataset contains no regulator-published plant-level figure.

The countries associated with the largest modelled CBAM charges are generally not the countries for which plant-level regulatory data is available.

Our current dataset includes regulator-published plant-level emissions from six jurisdictions: Australia, the United States, Canada, Taiwan, Japan and the European Union. EU data is included as a benchmark and validation source, even though EU production is not itself an imported CBAM origin. For three of the largest origins we checked the primary sources directly — China’s provincial emitter registers list names and production routes but no emissions, India’s crediting scheme covers aluminium and not steel, and Türkiye’s scheme has published no values yet. Elsewhere the absence reflects what we have searched for and obtained, not a certainty that nothing exists.

Activity is the annual throughput modelled for 2024. “None published” means our dataset holds no regulator-published plant-level figure for that jurisdiction; it does not mean the plant is unmeasured internally. Taiwan and the United States publish facility emissions that we hold but have not yet matched to plant identity.

Australia as a validation case

Australia accounts for only a small share of the modelled CBAM charge, but it provides one of the strongest checks on the plant-level method. Its entire modelled charge is €63 million, 22% of the value of the goods it sends, and almost all of it rests on one steelworks at Port Kembla.

What Australia provides is the only place where the whole chain can be checked end to end, because the Safeguard Mechanism publishes measured facility emissions and the regulator also publishes a determined intensity per product. The modelled intensity for Port Kembla comes out at 1.918 tonnes CO₂e per tonne of steel. The regulator’s determined value is 1.907.

That agreement is a consistency check rather than an independent test — Port Kembla is one of the two Australian plants the calibration was built on, and it should not be presented as independent validation. The independent evidence is the United States, where seven plants reporting to a different regulator under different rules give the same ratio.

What early 2026 trade data can — and cannot — show

Trade data for the first seven months of 2026 shows what changed after the definitive period began. It cannot by itself show that CBAM caused those changes.

Compared with the first seven months of 2025, import volumes in the first seven months of 2026 fell across every CBAM sector in this analysis, measured by weight on goods originating outside the EU and the EEA: fertilisers by 54%, cement by 21%, aluminium by 14% and steel by 11%.

The largest single movement by far is Russia — steel down 52%, aluminium down 78%, fertilisers down 81%, a single origin accounting for 2.8 million tonnes of the fertiliser fall on its own. Sanctions and separate tariff measures on Russian fertiliser are more immediate explanations for this decline, and they make any CBAM effect difficult to isolate. Over the same months imports from several origins rose: Chinese steel by 5%, Brazilian steel by 99%, Indonesian steel by 2% and Canadian aluminium by 80%. That pattern is not consistent with a simple claim that CBAM alone is already reducing imports.

One further caution on the published trade data. The share of aluminium arriving with its origin undeclared rose from 2,197 tonnes in the first seven months of 2025 to 103,959 tonnes in the first seven months of 2026, almost all of it unwrought metal. Part of what looks like a fall in any single country’s aluminium is simply origin no longer being stated.

Download the data

The figures on this page are published as CSV. They carry the country layer: the charge by origin, and the comparison against published country intensities. The plant layer is not included — its cross-source identifiers, the matching rules behind them and the per-plant calibrated intensities stay unpublished, for the reasons set out in the methodology note.

Published under the site’s terms. The underlying fields remain subject to the licences of the sources they come from, which is what the third file is for: EU trade and legal texts are reusable under Decision 2011/833/EU, the country intensities and the plant registers under CC BY 4.0, and a few sources are cited without their values being reproduced. Trade year 2025, 2026 rules, €80 certificate.

How this was checked, and its limits

The analysis was tested in four ways using public data. Each check produces a numerical result that can be independently reproduced.

Limits

• The €14.5bn is a default-value scenario, not a forecast. What is actually collected depends on how much verified data importers obtain, which cannot be observed from outside.
• Country-level intensities are analytical proxies, not verified facility data. They estimate where better emissions data may matter; they cannot determine the charge for an individual consignment, which depends on verified emissions for that product and facility.
• The estimates do not account for carbon prices paid in the country of origin. Any recognised carbon price would reduce the number of CBAM certificates ultimately due. The implementing regulation governing that deduction was still in draft at the time of writing and sets no list of qualifying countries.
• Indonesia’s €1,950M should not be read as exposure. The default value for its principal export corresponds to ferro-nickel and is not comparable with the others in the table.
• Outside the six jurisdictions named above our dataset holds no published plant-level intensity — the plant layer gives identity, route and a bounded estimate, labelled as such.
• The United Kingdom’s figures are inferred. Its scheme starts in 2027 and its default values have not been published; these are derived from the allowance price and the statutory rate formula, and will be restated when the values appear.
• Indicative desktop analysis of public information, not advice. Nothing here is a compliance determination. Any commercial, financing or compliance decision needs independent verification.
• The downstream extension is not priced here. Parliament reached its first-reading position in September 2026 and the scope is still in negotiation; the codes proposed so far carry about 10 million tonnes a year, worth roughly twice the current steel scope in value.