US efforts to revive primary aluminium hinge on securing long-term, industrial-scale power contracts of 500+ MW per smelter, with advocates warning that data centres and hyperscalers are outbidding producers for grid capacity. Imports of Canadian primary aluminium fell from 2,745 kt in 2024 to 1,950 kt in 2025 after Section 232 tariffs rose to 50%, even as Century Aluminum and Emirates Global Aluminium plan the first new US smelter since 1980 in Oklahoma, still stalled pending a power deal. Rio Tinto’s $1.5 billion AP60 low-carbon smelter expansion in Quebec and EU CBAM-driven demand are meanwhile pulling more Canadian hydropower-based metal towards Europe, risking longer-term diversion of supply away from the US.

Rio Tinto has begun a $1.5 billion expansion of its AP60 low‑carbon aluminium smelter at Arvida, Quebec, adding 96 new AP60 pots to lift capacity to about 220,000 tonnes per year, a 160,000‑tonne increase. Powered by hydroelectricity, the AP60 process emits about one‑seventh of the global industry average and is expected to cut Rio’s carbon emissions by roughly 290,000 tonnes per year as older Arvida potlines are phased out by next month. The project, backed by up to $113 million from Quebec, has created more than 1,500 construction jobs and will employ about 100 permanent staff, with a nearby recycling facility planned to integrate post‑consumer aluminium.

Advanced Magnet Lab has secured a US$2 million, two-year Defense Logistics Agency contract to qualify domestically produced high‑grade sintered NdFeB PM‑Wire magnets, including defence‑grade compositions such as N48SH and N35EH. The Florida firm will develop alloying, supply chain management and advanced manufacturing routes to scale permanent magnet production using its PM‑Wire process, which is designed to fit existing magnet‑making lines. AML is also progressing SmFeN, MnBi, anisotropic NdFeB and (Mischmetal‑Nd)FeB magnets, working with Phoenix Tailings, Ionic Rare Earths and Momentum to reduce critical rare earth content and improve material traceability.