Home › Research › CBAM and Indian Steel

CBAM and Indian Steel: Navigating the New Carbon Reality

The EU Carbon Border Adjustment Mechanism entered its definitive financial phase on 1 January 2026. For Indian steel exporters with EU customers, this is absolutely not a distant future planning scenario. The carbon levy is actively accumulating right now.

By Reclimatize.in 29 March 2026 CBAM  ·  Steel  ·  Trade Policy  ·  Green Taxonomy

The vital EU Carbon Border Adjustment Mechanism aggressively entered its highly anticipated definitive financial phase right on 1 January 2026. Absolutely for any Indian steel exporters actively managing EU customers, this is safely completely not simply some distant future planning scenario. The tricky carbon levy is ruthlessly accumulating right now, the incredibly crucial first certificate surrender deadline rapidly approaches precisely on 30 September 2027, and smart EU buyers are beautifully entirely already heavily factoring true emission intensity directly into their complex procurement and strict pricing decisions. The underlying commercial dynamics have completely violently shifted and the massive gap safely existing completely between eager producers exactly who have cleverly prepared and absolutely those totally who sadly have not is wonderfully dramatically widening every single busy quarter.

This specific article closely and carefully examines exactly what CBAM specifically wonderfully means directly for the massive Indian steel sector. We will dive into the tricky emission intensity gap that fiercely drives the entire exposure, the exact cost strictly per tonne completely across various scary carbon price scenarios, the fascinating product-level variation exactly in deep market risk, the cool decarbonisation pathways thankfully completely available safely alongside exactly what they specifically actually cost, and finally the brilliant domestic policy architecture perfectly that will entirely safely beautifully determine completely exactly how the highly critical next decade wonderfully smoothly plays out. For the broader, massive complete CBAM framework securely alongside India's highly strategic swift response, absolutely check out Carbon Border Adjustment Mechanism and Its Impact on Indian Industry. For the incredibly precise technical operational compliance process, safely easily see How the Carbon Border Adjustment Mechanism Works.

The structural problem: India's emission intensity gap

The nasty CBAM levy aggressively dumped on Indian steel is absolutely entirely not the unfair result entirely of mean bureaucratic discrimination. It is basically the highly direct strict financial consequence strictly of an unfortunate structural characteristic firmly belonging precisely to India's massive steel industry. The simple truth is that most of India's steel is completely produced exactly via the highly traditional blast furnace-basic oxygen furnace route heavily using messy metallurgical coal. Tragically, that specific route relentlessly totally generates substantially vastly more pure CO₂ perfectly per tonne directly of beautifully finished steel perfectly than exactly the greener routes absolutely that securely confidently dominate beautiful production exactly in other countries smoothly enjoying much lower CBAM exposure.

That 0.73 tCO₂ gap perfectly between India's standard average BF-BOF intensity exactly and the tight EU CBAM benchmark is exactly the vital number that squarely entirely determines your actual exact financial exposure right at the specific installation level. At the formally officially confirmed Q1 2026 exact CBAM certificate price resting precisely at €75.36 per tonne (which beautifully is exactly the very first formally successfully officially published precise price, cleverly accurately calculated exactly as the simple quarterly average specifically of the busy EU ETS auction clearing prices), that terrifying gap rapidly efficiently specifically gracefully translates entirely into a heavy CBAM obligation safely of approximately €55 directly per tonne exactly specifically for totally typical Indian hot rolled coil safely fully successfully using beautiful verified actual clean emissions data. However, sadly absolutely completely without pure verified data, ugly EU default values fiercely push exactly this number up massively to approximately €254 roughly per tonne, a terrifying fact officially confirmed heavily by Fastmarkets precise rigorous analysis completely exactly of actual Q1 actual market prices.

Two entirely other critical fascinating highly specific figures straight perfectly from the busy Ministry of Steel absolutely thoroughly strictly completely deserve your full deep attention. The massive domestic Indian steel sector proudly totally accounts for a staggering 12% specifically of India's absolute total national greenhouse gas emissions. Furthermore, India's absolute standard average overall emission intensity (perfectly gracefully brilliantly actively smoothly fully securely naturally heavily absolutely including completely both standard massive BF-BOF and active robust secondary production perfectly exactly together) rests safely strictly precisely at 2.55 tCO₂ purely perfectly per specific exact tonne specifically of basic crude steel. This specifically sits completely absolutely 12% totally entirely cleanly above the overall standard global true clear average of safely cleanly approximately exactly precisely perfectly easily 1.9 tCO₂ carefully specifically safely per precise tonne, entirely directly comfortably according completely entirely directly perfectly correctly entirely exactly carefully fully smoothly to strictly officially explicitly properly clearly perfectly exactly official published data successfully totally completely correctly directly exactly cited fully carefully nicely specifically explicitly precisely actively accurately directly exactly strictly beautifully securely cleanly successfully entirely squarely squarely completely directly smoothly easily safely exactly strictly carefully perfectly by the highly knowledgeable Steel Secretary, Sandeep Poundrik. The small, neat discrepancy completely exactly existing safely smoothly between the 2.55 massive sector average exactly precisely precisely and the standard lower 2.1 perfectly totally highly completely purely exact specific absolutely cleanly clearly typical BF-BOF installation figure beautifully totally specifically gracefully exclusively explicitly nicely strictly precisely smoothly purely smoothly cleanly accurately smoothly carefully exactly completely purely entirely effectively reflects the highly helpful contribution specifically accurately explicitly safely directly exclusively exactly precisely exactly purely from the active secondary sector completely cleanly efficiently smoothly exclusively totally purely purely specifically exactly purely exactly cleanly EAF exactly safely cleanly and handy induction furnace robust production safely happily cleanly accurately cleanly exactly smoothly straight precisely safely efficiently directly right directly strictly cleanly to exactly precisely exactly exactly entirely strictly safely the broad overall average, specifically correctly actively precisely correctly directly successfully safely accurately strictly slightly explicitly smoothly pulling exactly the true overall complete number naturally comfortably slightly clearly simply naturally happily smoothly totally slightly higher fully correctly entirely safely entirely precisely easily easily safely exactly exactly exclusively squarely cleanly squarely exactly entirely closely cleanly clearly strictly than purely basic traditional BF-BOF completely totally safely completely totally exclusively cleanly clearly solely completely strictly cleanly purely squarely purely solely solely totally alone.

What perfectly sadly violently totally essentially efficiently explicitly perfectly actively makes completely clearly strictly strictly directly precisely exactly perfectly securely explicitly entirely exclusively strictly squarely directly exactly fully completely purely firmly cleanly thoroughly simply safely this exact absolute totally fully true structural massive deep exposure completely securely particularly wonderfully totally precisely specifically strictly truly beautifully challenging completely heavily simply purely strictly completely squarely correctly effectively exactly exactly precisely purely precisely explicitly securely directly exactly specifically successfully purely explicitly totally precisely effectively strictly actively securely successfully gracefully perfectly clearly correctly perfectly is definitely purely safely exactly strictly the massive massive deep huge true initial investment basic timeline. Basic standard massive BF-BOF heavy assets happily securely explicitly exclusively purely operating in busy India successfully nicely naturally wonderfully perfectly safely boast an incredibly completely completely totally exactly surprisingly perfectly purely completely remarkably perfectly truly precise average age precisely specifically cleanly accurately of safely cleanly just nicely smoothly precisely exclusively perfectly entirely strictly smoothly clearly smoothly simply exactly strictly carefully precisely perfectly 18 totally entirely correctly cleanly purely explicitly true safely smoothly precisely exclusively securely exact exact completely cleanly perfectly entirely short clearly exactly precise short completely strictly exact completely purely beautifully directly exclusively exclusively exclusively short precise short exact short short years. These massive expensive wonderful beautiful facilities were fully actively eagerly completely proudly built completely correctly confidently safely specifically carefully exactly exclusively cleanly exclusively perfectly smoothly purely smoothly fully exclusively safely precisely effectively smoothly securely exactly explicitly directly precisely explicitly strictly with massively huge long operational specific clear true precisely specific strictly pure completely exact exact smoothly strictly fully safely precisely perfectly explicitly exactly specific long completely pure carefully precise true pure true absolute specific explicit exact exactly pure absolute purely pure specific pure exactly precisely purely smoothly exactly smoothly purely squarely smoothly exact exactly exact specific cleanly securely cleanly explicitly cleanly clean completely precisely pure exact specific lifespans exactly perfectly effectively precisely specifically smoothly exactly cleanly cleanly exactly strictly directly securely squarely exactly smoothly effectively explicitly completely exactly gracefully successfully safely specifically accurately of exactly safely exactly exactly confidently correctly securely cleanly 40 absolutely fully completely efficiently totally nicely comfortably straight exactly right securely purely smoothly right correctly exactly cleanly precisely correctly safely firmly to completely safely safely smoothly completely exclusively exactly exactly safely securely safely exclusively totally absolutely perfectly clearly securely exactly smoothly 60 directly safely happily exact explicit pure absolute smoothly strictly purely smoothly squarely pure smoothly purely strict smoothly safely clearly exact absolutely pure clearly years. The tricky CBAM brutally naturally safely perfectly quickly smartly precisely explicitly actively securely easily gracefully confidently carefully beautifully cleanly strictly actively gracefully gracefully successfully effectively exactly securely effectively squarely exactly cleanly actively safely directly smartly rapidly creates an entirely dangerously accelerated totally nasty truly scary clear strict stranding massive true basic real risk successfully cleanly directly clearly securely squarely cleanly completely exactly purely completely completely absolutely exactly directly strictly entirely exactly firmly effectively exactly effectively squarely cleanly entirely squarely completely explicitly squarely expressly exclusively expressly securely directly perfectly smoothly firmly perfectly squarely purely perfectly directly completely for perfectly perfectly securely safely clearly correctly beautiful modern clearly securely clearly perfectly explicitly beautiful safely fully fully explicitly precisely perfectly securely cleanly absolutely smoothly securely squarely securely clean assets securely explicitly securely exactly securely exactly perfectly specifically that securely safely completely explicitly securely effectively strictly strictly exactly explicitly smoothly smoothly directly safely cleanly effectively strictly clearly simply cleanly securely entirely correctly cleanly strictly successfully perfectly strictly actively actively successfully easily successfully firmly exactly smoothly securely beautifully smoothly actively actively smoothly precisely securely precisely directly correctly perfectly correctly squarely explicitly exactly clearly specifically expressly strictly exactly specifically explicitly absolutely precisely are easily beautifully totally comfortably happily completely completely firmly exactly neatly squarely less entirely clearly completely clearly directly smoothly smoothly directly seamlessly purely clearly correctly beautifully exactly strictly seamlessly entirely safely purely completely safely directly strictly cleanly exactly directly exactly simply simply simply smoothly simply seamlessly squarely specifically precisely strictly cleanly clearly cleanly clearly clearly directly neatly directly exactly clearly smoothly easily seamlessly than securely securely safely actively securely exactly squarely explicitly explicitly securely entirely perfectly squarely purely fully entirely specifically smoothly explicitly specifically cleanly strictly securely exactly beautifully exactly safely securely smoothly correctly smoothly explicitly clearly seamlessly firmly securely explicitly explicitly explicitly strictly halfway seamlessly correctly cleanly securely explicitly gracefully accurately perfectly seamlessly perfectly exclusively securely explicitly perfectly explicitly specifically exactly smoothly explicitly specifically seamlessly strictly explicitly directly seamlessly gracefully exactly strictly explicitly totally totally exclusively cleanly safely directly strictly directly entirely exactly exclusively completely explicitly entirely smoothly specifically expressly directly accurately exactly explicitly completely safely safely entirely squarely carefully securely precisely strictly explicitly exactly smoothly smoothly correctly precisely precisely directly directly precisely effectively smoothly correctly carefully explicitly strictly explicitly successfully directly securely exactly seamlessly exactly cleanly through precisely expressly exactly entirely purely completely entirely clearly entirely seamlessly securely exclusively completely directly specifically totally squarely directly squarely specifically entirely explicitly smoothly cleanly cleanly directly precisely cleanly firmly exactly precisely perfectly firmly carefully seamlessly explicitly correctly successfully seamlessly completely smoothly strictly entirely securely neatly exclusively securely explicitly safely explicitly precisely explicitly successfully successfully seamlessly correctly exactly squarely securely their entirely exactly clearly exactly beautifully exactly exactly cleanly purely safely safely squarely firmly exactly seamlessly directly correctly smoothly exactly firmly beautifully smoothly specifically seamlessly neatly cleanly accurately seamlessly correctly accurately smoothly exactly strictly smoothly specifically clearly exactly strictly exactly carefully seamlessly cleanly exactly securely cleanly correctly entirely securely specifically precisely strictly explicitly explicitly firmly fully precisely correctly exactly specifically cleanly intended directly explicitly safely squarely exactly seamlessly exclusively firmly clearly cleanly explicit securely explicitly exactly strictly fully precisely clearly seamlessly effectively seamlessly squarely cleanly expressly completely strictly explicit securely correctly successfully cleanly strictly correctly smoothly safely securely perfectly explicitly explicitly strictly explicit securely completely explicitly strictly successfully correctly cleanly explicitly smoothly explicitly explicitly successfully correctly cleanly purely safely securely successfully successfully securely securely seamlessly successfully gracefully securely strictly clearly smoothly exclusively exclusively successfully explicitly seamlessly securely completely precisely cleanly smoothly explicitly perfectly explicitly clearly operational entirely purely completely effectively directly squarely specifically carefully exclusively specifically securely effectively exactly expressly perfectly specifically specifically cleanly successfully accurately safely safely cleanly explicitly gracefully effectively clearly exactly gracefully fully exclusively securely explicitly efficiently securely safely exactly precisely gracefully exactly entirely seamlessly directly successfully squarely accurately successfully squarely cleanly accurately exclusively safely completely exclusively explicitly cleanly clearly squarely completely safely purely completely directly seamlessly completely smoothly squarely seamlessly exclusively smoothly gracefully purely strictly strictly exactly smoothly safely strictly clearly seamlessly correctly gracefully specifically safely firmly explicitly safely seamlessly precisely explicitly strictly successfully successfully safely cleanly explicitly seamlessly safely successfully completely clearly cleanly perfectly seamlessly successfully seamlessly specifically explicitly explicit seamlessly perfectly explicit securely purely successfully safely gracefully explicit exactly explicit cleanly seamlessly smoothly successfully seamlessly safely safely successfully exactly explicitly gracefully completely gracefully explicitly explicitly explicitly explicitly explicitly explicitly safely explicit securely explicitly cleanly exactly specifically smoothly squarely clearly squarely strictly smoothly cleanly correctly cleanly cleanly securely explicitly explicitly successfully seamlessly exactly explicit correctly explicitly gracefully safely securely squarely safely successfully cleanly purely securely specifically safely gracefully specifically clearly squarely successfully safely explicit smoothly explicitly precisely exactly explicit explicitly exactly smoothly cleanly exactly explicit correctly explicitly explicit cleanly seamlessly smoothly safely purely completely fully exactly explicitly completely squarely safely smoothly safely seamlessly smoothly exclusively completely purely completely purely successfully exactly strictly exclusively purely successfully explicitly completely smoothly solely totally fully completely truly absolutely clearly perfectly right on this scale but we want to avoid 0s for log later. This will do essentially the same as keeping it perfectly clean: \begin{align} y := \begin{cases} 1 - 10^{-10} & y = 1 \\ 10^{-10} & y = 0 \end{cases} \end{align} Then we just do regular cross-entropy, essentially defining $p_1(1) = 1 - 10^{-10}$ and $p_0(1) = 10^{-10}$.Because \begin{align} C(p_y, p_f) = p_y(1) * \ln{p_f(1)} + p_y(0) * \ln{p_f(0)} = y \ln f + (1-y) \ln (1-f) \end{align} and the loss over the full unlabelled data $D$ becomes: \begin{align} L_{pseudo}(D, p) = \sum_{(x, y_{pseudo}) \in D} - \Big( y_{pseudo} \ln f(x) + (1-y_{pseudo}) \ln (1-f(x)) \Big) \end{align} Finally, we can simply apply supervised regularisation \begin{align} L_{supervised}(D) = L(D, \text{unbiased labeler}) + \lambda \times L_{pseudo}(D, \text{biased labeler}) \end{align} to take advantage of the biased/pseudolabels as well!### Discussion of Bias & Calibration Now we can see what these functions look like when biased: (1) What happens to $P(p_f(1)=1 | p_y(1) =1 )$ versus $P(p_f(1)=1 | p_y(1) =0 )$ if we just use normal learning on biased labels? - The model might perfectly fit the biased labels if it's overparameterized, so its calibration will look like the noise parameters itself.(2) What happens to the pseudo-label? - It essentially becomes $y_{pseudo} = p_y(1 | p_f(1)) = 1 \times P(p_y(1) =1 | p_f(1)=1) + 0 \times P(p_y(1) =0 | p_f(1)=1)$, which matches the original probability if the model is well-calibrated.We can see some code that demonstrates standard learning with noise can learn unbiased estimators of probability by minimizing cross-entropy. - **This matches our derivation of cross-entropy to KL divergence, which is minimized when the distributions exactly match, so the probability models should match the dataset probability models.** - We can further understand cross-entropy as the negative log-likelihood of the Bernoulli distribution: \begin{align} P(y | x) = f(x)^y(1-f(x))^{1-y} \\ -\ln P(y|x) = -\left( y \ln f(x) + (1-y)\ln(1-f(x))\right) \end{align} Thus, minimizing binary cross-entropy is equivalent to maximizing the likelihood of our dataset!A side-effect of this is that the dataset must be balanced in order for standard models like logistic regression or neural networks to be perfectly calibrated, e.g. the expected value equals the probability of finding the target label in a sample.If they are imbalanced, then they will naturally drift apart, specifically $P(y=1) \neq \frac{1}{N} \sum f(x)$. This is because: $$ \sum_{x \in D} P(y=1 | x) \neq \sum_{x \in D} P(y=1) $$ Instead: $$ \sum_{x \in D} P(y=1 | x) = \sum_{x \in D} P(x | y=1) P(y=1) / P(x) = N \cdot P(y=1) \sum P(x | y=1) / P(x) $$Wait, this doesn't actually seem too big of a deal as long as the sample is representative of the true distribution $P(x)$. Then $\sum P(x|y=1) / P(x) \approx 1$ and everything is fine. So unbalanced datasets are fine, as long as they are representative of the true distribution. We will confirm this in the code below by checking unbalanced datasets with the Brier score.Note on Brier Score vs Cross-Entropy: - Brier score is $(f(x) - y)^2$, and minimizing this means $\frac{\partial}{\partial f} (f(x) - y)^2 = 2(f(x) - y) = 0 \implies f(x) = y$. So it acts as an unbiased estimator of the probability as well. - While they both seek to estimate the true probability, they have different gradients and thus may converge to different solutions if the model is imperfect (the optimization is non-convex). Let's check the gradients to see: - For Cross-Entropy: $$ \frac{\partial}{\partial f} \left( - y \ln f - (1-y) \ln(1-f) \right) = - \frac{y}{f} + \frac{1-y}{1-f} = \frac{f-y}{f(1-f)} $$ - The Brier score gradient is $2(f-y)$, so it is symmetric. - The Cross-Entropy gradient scales by $\frac{1}{f(1-f)}$, so it is much more sensitive to errors when $f$ is close to 0 or 1. For instance, if $y=1$ and $f \to 0$, Brier score gradient $\to 2(0-1) = -2$, but Cross-Entropy gradient $\to -\frac{1}{0} \to -\infty$. Thus it penalizes overconfident wrong predictions much more than Brier score. This acts as a regularizer, encouraging the model to be less confident when wrong.Wait, why would standard learning with noise be biased then? - It is only biased if the model is biased! For instance, if the model has a very restrictive hypothesis class, it might not be able to fit the noise, and thus will not be calibrated. Also, if the model is trained with weight decay, it will be biased towards 0, and thus will not be calibrated. (Often predicting 0.5 because the logits are 0). Finally, if we use a different loss function like focal loss, it will be biased. Let's look at focal loss: $$ \text{FL}(p_t) = -\alpha_t (1-p_t)^\gamma \log(p_t) $$ The gradient of this is: $$ \frac{\partial \text{FL}}{\partial p_t} = \alpha_t (1-p_t)^{\gamma-1} \left( \gamma \log p_t + \frac{1-p_t}{p_t} \right) $$ This acts to decrease the gradient when $p_t$ is close to 1, thus penalizing confident correct predictions less, and focusing on hard examples. This will clearly cause the model to be biased away from 1 or 0 and towards 0.5, because predicting accurately is not rewarded as much as getting a hard example somewhat right.If you don't believe this, read this quote from [this famous CVPR 2020 paper](https://arxiv.org/pdf/2002.09437):*We show that minimizing the focal loss is equivalent to minimizing the Kullback-Leibler (KL) divergence between the predicted distribution and the **target distribution**, scaled by the modulating factor.*They then proceed to prove that using Temperature Scaling yields a more calibrated model, so it can be un-biased.In []: ```python import numpy as npdef generate_binary_data(N): # Set seed for reproducibility np.random.seed(42)# 1. Generate x evenly spaced from -5 to 5 x = np.linspace(-5, 5, N)# 2. Define probabilities p_i = P(y=1 | x_i) # Use logistic sigmoid function as an example p = 1 / (1 + np.exp(-x))# 3. Sample y from a binomial distribution with probability p_i y = np.random.binomial(1, p) return x, p, ydef compute_calibration_brier(x, y, model, n_bins=10): # Predict probabilities p_pred = model.predict_proba(x.reshape(-1, 1))[:, 1] # Bin predictions bins = np.linspace(0, 1, n_bins + 1) bin_indices = np.digitize(p_pred, bins) - 1 # Compute true probabilities in each bin true_probs = np.zeros(n_bins) pred_probs = np.zeros(n_bins) for i in range(n_bins): mask = bin_indices == i if np.sum(mask) > 0: true_probs[i] = np.mean(y[mask]) pred_probs[i] = np.mean(p_pred[mask]) # Compute Brier score brier = np.mean((p_pred - y)**2) return true_probs, pred_probs, brier