The Kooranjie Project is located inside the Cobar Basin in the Central West of New South Wales, roughly 500 kilometers northwest of Sydney. It is a remote area of Australia with red earth and sparse scrub that mining geologists have flown over for decades without stopping. After analyzing satellite data, decades’ worth of drilling logs, and geochemical datasets, Earth AI’s program identified a polymetallic prospect worth looking into in that area. Copper, gold, tin, and cobalt. Drilling is scheduled to begin in a few weeks after the announcement of six confirmed prospects in two project areas in March 2025.
It would have taken years and cost several times as much to find those deposits the old-fashioned way. The track record of traditional greenfield mineral exploration is dismal; according to industry estimates, geologists discover a profitable deposit around once every 200 efforts. It’s the nature of looking for anything buried hundreds of meters deep on a continent the size of Australia, not a lack of knowledge. The easy stuff—the surface-level ore deposits that prospectors discovered a century ago—is mostly lost, and the earth makes this difficult.
In order to close this gap, Earth AI was established in 2017. The main aim was to train an AI system to identify patterns associated with ore deposits across various rock types, formation ages, and structural contexts using as much geological data as possible. The company’s model has now been trained on 400 million geological sites worldwide.
The system incorporates radiometric surveys, satellite remote sensing, drone magnetometry, historical drilling records from government and commercial repositories, and geochemical anomaly maps that may have been left unopened in a filing cabinet for forty years. It would take months for a human geologist to review that amount of data. The system completes the task in a matter of hours.
The disparity in success rates is substantial enough to completely alter the economics of exploration. The industry standard is about 0.5 percent. If Earth AI’s 66 percent assertion holds true for a variety of geological settings and commodity types, it will fundamentally alter how exploration spending is calculated. Customers who were dubious of a “unproven technology” weren’t prepared to spend millions acting on recommendations from an algorithm they didn’t understand, so the business took pains early on to validate the idea with its own hardware rather than just providing software predictions.
In order to validate its own predictions, Earth AI constructed its own modular drilling equipment, replete with crew housing and fluid treatment, that could function independently in isolated desert locations. The first evidence was the discovery of molybdenum in 2023, which was a true greenfield discovery. The next year, New South Wales implemented a bigger palladium system.
The timing is related to a larger development in the commodity markets. The metals that these AI systems are targeting include lithium, copper, cobalt, nickel, and rare earth elements. These metals are necessary for clean energy infrastructure, grid storage batteries, and electric vehicles in numbers that the mining industry isn’t providing quickly enough.
By 2035, the demand for copper alone is predicted to nearly double. The remaining deposits are deeper, more intricate, and situated in areas where obtaining permits before a drill bit hits the earth is more difficult. For them, investigating using the outdated techniques at the outdated success rates is simply too sluggish for the timescale needed for the deployment of renewable energy.
