Earth AI spots critical minerals in Australia's forgotten grounds

For generations, finding a new mineral deposit meant sending geologists with pickaxes into the desert or sifting through decades-old survey sheets. It often took decades to locate anything worth mining, if you found anything at all.

Now, machine learning is turning those dusty archives into actual supply chains. Mining tech startup Earth AI just revealed that its algorithms discovered promising critical mineral deposits in areas of Australia that traditional operators ignored for decades.

While rival KoBold made headlines last summer with an enormous copper find in Zambia, Earth AI is proving that algorithms parsing old field data can consistently unearth overlooked spots. In Australia's Northern Territory, the company identified copper, cobalt, and gold. Meanwhile, 310 miles northwest of Sydney in New South Wales, its systems flagged silver, molybdenum, and tin.

Digging into dust-covered archives

Earth AI was born out of founder Roman Teslyuk's doctoral studies at the University of Sydney. While studying the Australian mining sector, he noticed an odd regulatory quirk: because the government owns mineral rights and leases them on six-year terms, exploration companies have been required to submit their exploration data to a national archive since the 1970s.

Almost nobody was using that mountain of public data. Teslyuk realized that if an algorithm could ingest that massive record of past failures and successes from thousands of geologists, it could predict future mineral locations far better than humans.

According to Teslyuk, the true frontier in mining today is technological rather than geographical.

From pure software to building drill rigs

Selling AI to miners turned out to be nearly impossible at first. Teslyuk originally structured Earth AI as a pure software company, generating predictive maps and pitching mining companies to go drill them. But nobody wanted to risk millions of dollars on an unproven algorithm in an industry that treats anything outside standard dogma as heresy.

So Earth AI pivoted and built its own mobile drilling equipment to prove the algorithms worked. After joining Y Combinator in spring 2019, the team spent years refining both the machine learning models and the physical drilling hardware, eventually raising a $20 million Series B this past January.

Unlike KoBold, which often tackles massive complex formations, Earth AI focuses on rapidly scanning vast swaths of land to pinpoint rich targets that legacy surveys walked right past.

Why it matters

Earth AI's journey is a masterclass in how AI startups have to adapt when tackling deeply conservative legacy industries.

We think their decision to build custom drilling rigs was the turning point. In industries like mining, heavy construction, or oil, decision-makers will rarely buy an AI subscription if verifying the prediction costs millions out of their own pocket. By shouldering the drilling risk themselves, Earth AI forced the industry to take note of real physical cores rather than just colorful heatmaps.

For the energy transition, this shift cannot come soon enough. The modern economy needs staggering amounts of copper, nickel, and cobalt, but twentieth-century exploration timelines simply cannot match the pace of demand. Reviving decades of overlooked government data with neural networks could unlock domestic resources right under our noses.

The catch, of course, is scale and grade. Finding promising mineralization is very different from proving a commercially viable, multi-billion-dollar mine. Heavy field operations burn cash quickly, and Earth AI still needs to show that its sites possess the volume and concentration required to justify full-scale extraction.