AI in mining: four measured results from US Steel, BHP, Vale and Fortescue

Will Drewes

Founder, Fern Strategy · LinkedIn · 3 min read

Where the measured wins are

Mining has been slower than finance or healthcare to publish AI outcomes, partly because the gains sit inside operational reporting and partly because sharing them costs more competitively. Enough has surfaced to see the pattern: the measurable wins cluster in maintenance, processing efficiency and autonomous equipment. Companies cite safety, and the numbers they publish are mostly cost avoidance.

US Steel: maintenance work orders in Minnesota

US Steel deployed a generative-AI maintenance workflow at its Minnesota iron ore operations and reported roughly 20% reduction in work-order completion time, along with productivity gains and lower maintenance costs it did not break out. Unplanned downtime on one piece of heavy equipment can run tens of thousands of dollars an hour, so 20% faster work orders is money. The part I find more interesting is where it sits: an agent-assisted process inside the technician's task, rather than a prediction in a dashboard.

BHP: processing at Escondida

At Escondida, one of the world's largest copper mines, BHP has used AI and ML to improve copper concentrate recovery. Since FY22 that work has saved more than 3 gigalitres of water and 118 gigawatt-hours of energy. Water and energy are direct input costs there, so read those as margin figures. They are also cumulative across a multi-year deployment, which says the system has stayed stable enough to keep delivering.

Vale: tires and rail fractures

At Salobo, AI-enabled maintenance extended haul-truck tire lifespan by 30%, saving $5 million in a single year. Haul-truck tires run $40,000 to $100,000 each, and the gain compounds across a large fleet.

On its rail network, Vale's AI fracture prediction cut fractures by up to 85%, with projected savings of $7 million per year. A fracture under a loaded ore train is a safety event before it is a cost. The 85% is the number for risk and safety leadership.

Fortescue: 360 autonomous electric trucks

Fortescue's planned fleet of 360 autonomous battery-electric haul trucks is the largest single commitment to autonomous haulage in the industry. The company projects it will remove approximately 400 million liters of diesel annually; diesel is 51% of its Scope 1 emissions. This is a projection. The scale of the bet is the information: Fortescue has decided that running 360 autonomous electric trucks is easier than paying for the fuel, emissions and labor of the alternative, and it changes the safety picture on haul roads, the shift structure and the maintenance profile all at once.

What this means for operators

The high-value outcomes are in maintenance and equipment lifecycle. Tires, work orders, rail fractures: unglamorous, high-frequency, rules-driven problems with messy inputs, which is the profile where agents pay off.

None of these came from a black box dropped into production. BHP's multi-year figures and Vale's fracture system both imply sustained integration work. That layer is where the result holds or falls apart.

The safety case and the cost case are usually the same deployment. Vale's rail system saves $7 million and prevents derailments in one project. If you're evaluating AI for safety compliance alone, you're undervaluing it.

Before any of that is credible you need to know which workflows are instrumented, which aren't, and where the liability sits when an automated decision goes wrong. That audit comes before the build.