Can we specify it? Testing data-centre material claims for Australian projects
A material innovation becomes useful when the evidence covers the assembly you will build. For coatings, that means the steel member and complete protection system. For timber, it means the floor and its structural, fire, acoustic and moisture performance. For AI-assisted concrete, it means tested performance using the proposed ingredients and production conditions. The examples below are leads for investigation; the decision guide shows what must be resolved before specification.
Part 4 of 4: What the data-centre build-out is teaching construction.
A coating saving depends on the section and application conditions
An intumescent coating provides thermal protection by expanding when exposed to heat. Its thickness is one part of a tested or assessed system. A smaller nominal thickness is not enough to establish equivalent fire performance or a shorter project programme.
Start with the project’s fire-protection schedule, then ask the supplier and fire engineer to map the evidence to each proposed member:
- Required fire performance, exposure conditions and the applicable assessment basis.
- Member shape and size, orientation, exposed sides and the design assumptions used to select the coating thickness.
- The specified primer, intumescent coat, reinforcement where required and topcoat.
- Preparation, application and curing conditions, including access for inspection.
- The required dry-film thickness and the method for recording and accepting the installed work.
- Maintenance, repair and future alteration requirements for the completed protection system.
Makao’s recommendation is to compare the complete installed system and inspection sequence. Lower coating volume may help, but drying, access, repairs and acceptance can still control the completion date.
Do not treat an overseas fire period as an Australian project approval. Establish the jurisdiction, adopted NCC edition, building classification and project compliance pathway with the responsible practitioners. Then confirm which test or assessment evidence is suitable for that pathway.
KCC Firemask 3005 illustrates why the detail matters. Its English technical data sheet describes Korean three-hour approval and lists 9.45 mm for beams and columns, including 0.05 mm of primer. It also specifies mesh reinforcement, qualified primers and application conditions, and directs users to KCC for shapes other than H or I sections. The thickness is therefore not a stand-alone instruction for every steel member.
Use the current manufacturer’s documentation and the project-specific assessment together. Have the supplier, fire engineer and building surveyor or certifier confirm the proposed assembly and evidence before procurement. A product brochure cannot resolve a mismatch between the tested configuration and the design.
Microsoft's timber data centres use CLT floors on a steel frame
Microsoft’s Northern Virginia example uses hybrid construction. Thornton Tomasetti describes concrete foundations and a steel frame, with cross-laminated timber (CLT) replacing the conventional metal-deck floor slab. Gensler describes off-site prefabrication of the CLT. This is a change to a floor system, not an all-timber data centre.
Microsoft reports estimated embodied-carbon reductions of 35% against conventional steel construction and 65% against typical precast concrete. Its article does not provide a reproducible calculation or complete baseline quantities. Those percentages should not become a default allowance for another building.
Compare project-specific floor systems that deliver the same spans, loads, vibration limits, fire and acoustic performance, service life and usable space. Include any additional topping, protection, connections, supports and transport in the carbon boundary. Check moisture management during construction and local supply capacity. The design can be promising without inheriting the published percentage.
AI-assisted mix design was tested on one data-centre slab
Amrize’s July 2025 account describes an AI-assisted mix developed with Meta and the University of Illinois Urbana-Champaign for a Rosemount data-centre slab. Its later project account describes laboratory testing of more than 100 mixes to train a model before full-scale use. These are project participants’ accounts, not independent validation.
Keep the three kinds of evidence separate:
- Acceptance: Amrize says the mix met Meta’s slab requirements and was approved for further use at that facility.
- Reported performance: Amrize claims “43% faster in early strength development”. Its project account does not identify the comparison mix, test age or timing method. This is not evidence of 43% higher strength.
- Estimated carbon: the supplier reports a 35% reduction. The public account is not a reproducible project LCA.
The public SustainableConcrete repository provides models and data for strength prediction and mix optimisation. That makes the method inspectable; it does not validate every supplier claim or qualify a local mix. AI proposes candidates. Materials testing establishes whether the proposed ingredients, curing regime and production process meet the project requirements. Assess durability, workability and specified-age strength as well as early strength. A carbon reduction comes from the resulting material quantities and production impacts, not from the AI label itself.
The question that can overturn an attractive proposal
Ask: what project condition would make this evidence inapplicable? A coating assessment for one member configuration does not automatically cover a different section. A concrete strength result does not establish durability in a different exposure. An estimated carbon saving can shrink when added protection or supporting materials enter the boundary. Record the mismatch and the test or assessment needed to resolve it. An unanswered question is a hold point, not a rejection of the technology.
A specification decision record the team can reuse
Record | What the team must establish |
|---|---|
Proposed change and baseline | Identify the exact member, assembly or mix and the incumbent design. Define the function and performance to remain equivalent. |
Applicable evidence | List the report, revision, tested configuration, assessment scope and limitations. Map them to the proposed use. |
Mismatch and failure conditions | Identify what would invalidate the evidence: different geometry, materials, exposure, curing, loading or installation conditions. |
Carbon and programme comparison | Use consistent quantities, boundaries and performance. Include added supporting materials, tests, rework and affected activities. |
Trial and acceptance | Define the representative trial, measurements, acceptance criteria and responsible reviewer before ordering or installation. |
Recorded decision | Name the authorised decision-maker, accepted evidence, remaining conditions and hold point. Update specifications and affected models. |
Makao’s decision rule: retain the baseline until the proposed change has passed the relevant technical and commercial reviews. A bounded trial can resolve uncertainty without pretending it has already been resolved. The useful question is not whether a material is innovative; it is whether the proposed assembly delivers the required performance, with evidence the team can inspect.
Before ordering a proposed alternative, use the substitution approval screen to record the technical evidence and the actual approval authority.
Where the benefit is a product-carbon comparison, first test whether the EPDs are comparable for the same function and boundary.
Sources
Primary project accounts and technical sources. Supplier benefits remain attributed claims. The decision record is Makao’s synthesis, not a product approval.
- KCC, Firemask 3005 English technical data sheet, including Korean certification references and application conditions.
- Meta research team, SustainableConcrete: public code and data for mix modelling.
- Microsoft, 26 May 2025
- Thornton Tomasetti project page (dated 28 March 2025)
- Gensler project page (undated)
- Amrize, 16 July 2025
- Amrize project page (undated)
Questions or corrections: contact Makao.
