Low-carbon Steel for Data Centres Market (2026-2036)

The growth of this market is driven by the arrival of mandatory life-cycle carbon disclosure for new buildings in the European Union, the start of the definitive phase of the Carbon Border Adjustment Mechanism, and the embodied carbon commitments that large data centre operators have attached to their construction programmes. Steel is the single largest contributor to the embodied carbon of a data centre shell, which makes the production route of that steel the first variable a designer can change without altering the building.

Growth in this market is nevertheless a penetration story rather than a volume story. The tonnage of steel consumed by data centre construction grows with floor area, but the share of that tonnage which qualifies as low-carbon grows far faster, because it is being pulled by regulation that lands inside the forecast period rather than by voluntary preference. The constraint sits on the supply side. Announced hydrogen-based capacity has been delayed and in several cases withdrawn, which means the low-carbon steel a data centre can actually buy before 2030 is overwhelmingly scrap-based electric arc furnace material rather than the hydrogen steel that procurement commitments were written against. The market therefore grows quickly while remaining tightly rationed, and value accrues to producers able to document what they already make rather than to those able to describe what they intend to build.

Low-carbon steel for data centres is not a different material. It is the same structural sections, plate, coil, rebar and coated flat product used in any data centre, made by a production route with lower emissions per tonne and sold with documentation that evidences the difference. The product is defined by its paperwork and its process, not by its mechanical properties, which is why the market behaves unlike any other steel category. A buyer cannot inspect the attribute, cannot test for it on delivery, and cannot substitute between suppliers without revalidating the claim, so trust in the certification route carries commercial weight that would be absurd in ordinary steel trading.

Steel matters in a data centre because of where it sits in the carbon account. Operational emissions have been attacked for a decade through power purchase agreements, efficiency and siting, and in facilities running on contracted renewable supply the embodied carbon of the building becomes the larger remaining number. Within that embodied total, the structural frame, foundations, decking, support steel and reinforcement dominate, and steel dominates them. An operator that has already decarbonised its electricity has nothing left to work on but the shell, and the shell is mostly steel and concrete. That is the demand logic, and it is why data centre operators appear among the most active buyers of low-carbon steel despite being a small share of global steel consumption.

What has changed since 2025 is that this logic has acquired a legal deadline. Under Directive (EU) 2024/1275 on the energy performance of buildings, the life-cycle global warming potential of a new building must be calculated and disclosed in its energy performance certificate from 1 January 2028 for new buildings above 1,000 square metres of floor area, and from 1 January 2030 for all new buildings. The calculation covers the full emissions chain, including materials production and transport, on-site construction, operational energy, maintenance, demolition and end-of-life waste handling with reuse and recycling. Every new data centre in the European Union sits well above the one thousand square metre threshold, which means that from 1 January 2028 the embodied carbon of the steel frame appears on a document the building is required to hold. The common calculation method was set by Delegated Regulation (EU) 2026/52, which the European Commission adopted on 16 December 2025, which was published in the Official Journal on 4 May 2026 and entered into force on 24 May 2026 following the two-month scrutiny period by the Council and the Parliament. The regulation amends and replaces Annex III to the Directive, allows Member States a measure of national flexibility, aligns with the Construction Products Regulation and with the Ecodesign and energy labelling rules, encourages low-carbon and circular materials, and permits default values where product-specific data is missing. The practical consequence is that a design decision which was previously discretionary becomes a disclosed figure subject to a common method, and a number that must be disclosed is a number that gets managed.

On the import side the cost of high-intensity steel is being set separately. Regulation (EU) 2023/956, which established the Carbon Border Adjustment Mechanism, was adopted on 10 May 2023 and began to apply from 1 October 2023. The transitional phase imposed reporting duties only, and the definitive period starts on 1 January 2026, when importers must pay for certificates covering the embedded emissions of covered goods. Iron and steel is one of the six sectors in scope, alongside aluminium, fertilisers, cement, hydrogen and electricity, and the covered products are listed by CN code in the annexes to the regulation. Importers must register as CBAM declarants through the CBAM registry, must hold certificates covering at least 80 percent of the embedded emissions of goods imported since the start of the year at each quarter end, and must report and surrender certificates annually by 31 May. A proposed amending regulation would move the surrender date to 31 August, defer the 2026 purchase obligation so that the platform begins selling certificates from February 2027, and reduce the quarterly holding ratio to 50 percent. Free allowances under the European Union Emissions Trading System are accounted for in the calculation, and the CBAM cost rises through the CBAM factor as those allowances are withdrawn. For a European data centre project, CBAM and the life-cycle disclosure requirement act on the same material from different directions. CBAM prices the embedded emissions of imported steel at the border, while the Directive requires the embodied carbon of the finished building to be calculated and published. A supplier who can satisfy both with one data set holds an advantage that neither regime creates on its own.

The difficulty is that demand created by regulation has run ahead of supply created by investment. In June 2025, ArcelorMittal suspended planned investments of EUR 1.3 billion in direct reduced iron and electric arc furnace capacity intended to decarbonise its German steelmaking at Bremen and Eisenhuettenstadt, where the technology was to replace blast furnaces and where construction had been scheduled to begin by that month. The company stated that the business case for proceeding was not strong enough even with financial support from the German federal government, the state of Bremen and the state of Brandenburg, citing weak European steel demand, high levels of steel imports, high German electricity prices, delays in green hydrogen and the weak competitiveness of gas-based direct reduction, and said it was unlikely to meet its 2030 carbon emissions target. That decision is the clearest available indication of the gap between announced and buildable capacity, and it is not isolated. The economics of hydrogen-based direct reduction depend on cheap firm low-carbon electricity and on hydrogen at prices that have not materialised, and where those inputs are expensive the route does not clear its cost of capital even with public support. The consequence for a data centre buyer is concrete: a procurement commitment written against near-zero steel must in practice be met for most of this decade with scrap-based electric arc furnace material, which is genuinely lower in intensity than blast furnace steel but is not near-zero and is limited by the availability of scrap.

Scrap is therefore the quiet constraint on the whole market. Electric arc furnace steelmaking is as low-carbon as its charge and its electricity allow, and both are finite. High-quality scrap suitable for structural sections competes with every other EAF user, and in regions where the existing fleet is already largely electric the marginal tonne of low-carbon steel is harder to find than in regions converting from blast furnaces. This produces the counter-intuitive result that the regions with the cleanest steel industries have the least headroom to supply additional low-carbon tonnes, while the regions with the dirtiest have the most potential improvement and the least near-term availability.

Allocation accounting has grown into the space between the two. Where physical near-zero tonnes are scarce, mass-balance and book-and-claim methods allow a mill to assign a low-carbon attribute to a volume of conventional output, so that a buyer receives a certificate rather than a physically distinct tonne. These methods are widely used, are recognised in several standards, and are the fastest-growing certification route in this market precisely because physical supply is short. They are also contested, because the relationship between the certificate a data centre holds and the emissions actually avoided depends entirely on the integrity of the allocation rules. Any buyer relying on this route should expect the rules to tighten during the forecast period.

Competition reflects these conditions rather than conventional steel rivalry. Electric arc furnace producers with long-standing scrap-based operations can supply reduced-intensity steel today at modest premiums and compete on availability and documentation quality. Integrated producers converting to direct reduction are selling future capacity against present commitments and compete on credibility, which is now harder to establish after a sequence of deferred projects. A small number of new-build hydrogen-based producers compete on the strength of the attribute itself, with limited volume and premium pricing. Data centre operators buying across several regions encounter all three in the same tender, and the comparison between them is not a price comparison.