AI Data Centre Demand Could Widen PGM Deficits

African Rainbow Minerals (ARM) plans to reopen Bokoni Platinum in South Africa with R15.2 billion in investment over seven years, targeting annual PGM production of up to 400,000 ounces from around 2030. The project follows a new feasibility study and a previous R2.2 billion impairment in ARM’s 2025 financial year. The restart has raised investor concerns over capital intensity and near-term cash generation, particularly if ARM also proceeds with the Two Rivers Merensky project. RMB Morgan Stanley estimates that funding both projects could push ARM into negative free cash flow for two to three years. ARM shares reportedly fell sharply following the Bokoni announcement. However, the project could benefit from improving PGM market fundamentals. South African PGM production is declining while renewed automotive demand, supported by slower-than-expected EV adoption, could strengthen prices. ARM entered the project with a strong balance sheet, including R9.5 billion in net cash at end-June. Bokoni's restart could add significant future PGM supply, but its high capital requirements and challenging mining history create execution and cash-flow risks for ARM.

Researchers at Washington University in St. Louis and partner institutions have developed a new platinum-cobalt catalyst design that could improve the durability and efficiency of low-temperature fuel cells, potentially supporting power generation for data centres. The technology uses platinum-cobalt nanoparticles confined within hollow carbon nanochannels, helping prevent the nanoparticles from clumping or degrading during operation. Platinum is widely used as a fuel-cell catalyst because of its high catalytic performance but its cost remains a key challenge. Accordingly, the new catalyst maintained around 85% of its performance during testing designed to simulate 25,000 hours of operation. The development could have implications for platinum demand if fuel cells become more widely adopted for distributed power generation. This is particularly relevant as data-centre electricity consumption is expected to rise significantly with growing digital and AI infrastructure. However, further development and commercial validation will be required before large-scale deployment.