AI that digested 300 years of geology reports helped find Zambia's biggest copper deposit in a century

KoBold Metals digitized three centuries of handwritten geology records to train AI models that pointed to where to drill — and the result, the Mingomba project, is already a $2.3 billion mine under construction.

At an event on mining modernization in South Africa, on July 27, 2026, PwC presented the method KoBold Metals used to locate what is now described as Zambia's biggest copper deposit found in a century: the company digitized 300 years of handwritten geology reports, trained artificial intelligence models to interpret those records alongside geophysical, geochemical and historical drilling data, and used that combination to predict where high-grade copper mineralization was most likely before drilling. The result, the Mingomba project, is already a mine worth more than $2.3 billion with construction underway, including the start of main shaft sinking.

What sets this case apart from other AI-guided exploration projects is the data source used: not just modern geophysical surveys, but three hundred years of handwritten reports, historically unreadable by any automated system, turned into structured data a model can process. It's a reminder that much of AI's value in mineral exploration isn't about collecting new data, it's about recovering old data that already existed but had never been used at scale because it was locked in a manual format. Mingomba is also the proof of concept that was missing: it's no longer a lab pilot, it's the first world-class deposit whose characterization was AI-guided and that turned into a real mine, with defined investment and construction timeline — and a target of more than 300,000 tonnes of copper a year once it enters operation, in the early 2030s.

The case is being cited publicly as a model to replicate: mining modernization experts are already discussing using the same approach to fill South Africa's mineral cadastre, currently considered underused. For exploration geologists and project managers, the message is that underused historical archives — old maps, handwritten reports, decades-old drilling records — can be worth as much as a new geophysical survey, as long as there's a method to digitize and cross-reference that data with AI.