Low-carbon aluminum just became a Scope3 procurement line in the AI build-out

Amazon has committed to net-zero across operations by 2040, and the company's Scope 3 footprint — emissions embedded in the materials it buys — is the largest bucket it cannot decarbonize unilaterally. Aluminum is a leading candidate: smelting is electricity-intensive, and the anode reaction in conventional smelting releases both CO₂ and powerful perfluorocarbons. By specifying ELYSIS aluminum at the cable stage, AWS shifts the emissions reduction upstream and locks in auditable Scope 3 credits without touching the data-center building itself.

The mechanism matters because Prysmian targets net-zero across Scopes 1, 2, and 3 by 2035 — among the most aggressive in the cable industry — and aims for 55% of revenues from sustainability-linked solutions by 2028. The Ohio deal is the first proof point that ELYSIS-grade cable can be specified into hyperscaler RFPs at scale. If a single hyperscaler signs, others follow: AWS, Microsoft Azure, Google Cloud, and Meta all compete on Scope 3 narratives, and the loser is whichever one is still buying commodity aluminum while peers buy carbon-traced metal.

Conventional aluminum smelting carries roughly 1.5 tonnes of CO₂-equivalent emissions per tonne of metal from anode consumption alone, before any electricity is counted. The C2ES pegs the smelting step at about 17% of all aluminum-industry greenhouse-gas emissions. Rio Tinto's Quebec operation runs on hydropower, so the electricity component is already near zero — ELYSIS collapses the residual anode-emissions tail to near zero as well, leaving only embedded process and ancillary loads.

Constellium, the recycled-content specialist, posted 4.55 metric tons of Scope 1+2+3 CO₂ intensity per metric ton shipped in 2025 — better than industry-average primary aluminum, but still an order of magnitude above ELYSIS-at-hydroelectric on the smelting step alone. Recycled-content aluminum does not neutralize the same Scope 3 line because AWS's aluminum in cable is new primary metal, not closed-loop recycling from a can.

Four tiers cleared the AWS spec. Upstream, the only U.S.-accessible source of inert-anode aluminum today is Rio Tinto's Quebec operation; ELYSIS remains a controlled-technology JV with Alcoa. Midstream, Prysmian's Sedalia plant drew the rod and extruded it into building-wire and data-center cable products. Distribution runs through Wesco's EES segment, which already handles 39-plus data-center builds for a major cloud provider. The end customer is AWS Ohio, where the cable is being installed as part of the active build-out cycle.

CRU estimates the data-center sector represented ~7% of total North American cable demand in 2025 and is forecasting ~17% CAGR through 2030, with aluminum taking an increasing share of the cable product mix. Combined copper and aluminum cable and non-cable metallic demand for data centers is expected to hit 850–900 kt in 2026 and exceed 1,500 kt by 2030. Aluminum overhead cable demand has already risen at about a 9% CAGR from 2020–2026. In short, the addressable market for ELYSIS-grade cable is roughly doubling within five years, and the share that will be specified as low-carbon is set by the largest buyers.

Prysmian gave investors the clearest signal in July2026: a 10-year agreement with Molex (a Koch company) worth up to €5.5 billion, supported by a €550 million upfront payment, and a stated target of €10 billion in cumulative hyperscaler and data-center revenues through 2035, with up to €1.1 billion in incremental revenue from 2031 alone. Q2 2026 group revenue hit €6,021 million (+9.4% organic) with adjusted EBITDA of €730 million — Prysmian's strongest-ever quarter, driven by data-center demand.

On the upstream side, Rio Tinto's Aluminum business posted $9.4 billion of revenue and $3.1 billion of EBITDA in H1 2026 (+38% YoY), with average realized prices up 39% to $4,343 per tonne. The segment is carrying the highest realized-price gains in Rio Tinto's portfolio — and the only primary aluminum producer that can supply ELYSIS-grade metal is the only one positioned to lock in a premium to LME for the lowest-carbon molecules. Alcoa, the ELYSIS co-pilfer, posted Q2 2026 revenue of $4.0 billion (+24% sequential) and is acquiring South32's bauxite, alumina and aluminum assets for $4.1 billion — adding primary metal to its portfolio as the cable-grade aluminum spec tightens.

Near-term (days to quarters), three triggers could rerate the value chain. First, any follow-on hyperscaler order — Microsoft Azure or Google Cloud — converting from ELYSIS building-wire trials to data-center cable spec. Third, the start of commercial-scale ELYSIS production at Rio Tinto's Alma smelter (announced in June 2024 with first production targeted by 2027); if the 2,500-tonne-per-year commercial cells ramp on schedule, the supply constraint binding hyperscaler adoption loosens. Fourth, Wesco signing additional bundled procurement arrangements with hyperscalers — the model that monetizes the distribution layer's relationship to the spec.

Long-term (1–3 years), the structural thesis is straightforward: as data-center power demand doubles and Scope 3 emissions accounting tightens, low-carbon aluminum stops being a marketing line and becomes a bill-of-materials specification. Rio Tinto and Alcoa own the only commercial-scale primary aluminum process that eliminates direct anode emissions; Prysmian has converted that into a hyperscaler contract clause; Wesco sits in the distribution lane where the spec is enforced. The losers are conventional smelters — Century Aluminum and Russian or Chinese primary producers — without low-carbon capacity or hydropowered grids.

The main risks to the thesis: ELYSIS commercial-scale production delays beyond2027 keep supply tight and prevent broad adoption; aluminum prices fall and the low-carbon premium evaporates; hyperscalers pivot to lower-carbon copper substitutes for some applications; and trade policy — including the U.S. Section 232 tariffs on aluminum and Canadian retaliation — re-prices the geography of supply. None of these undercut the directional case, but each can push the timeline.