According to the World Bank’s Digging Beneath the Surface report published in 2019, Southern Africa has the world’s richest mineral deposits, and South Africa is estimated to have non-energy minerals worth more than $2.4-trillion. Those are only known deposits, and there are possibly trillions of dollars of undiscovered mineral deposits. According to the Bureau for Economic Research, only 9%-14% of the country has been sufficiently geomapped to identify mineral deposits. South Africa sits on the world’s biggest mining opportunity, but ordinary prospecting and exploration take years and, due to various factors, are waning. Stats SA reports that investment in mining exploration in South Africa dropped from R6.3bn in 2006 to just R740m in 2025, with Economic Research Southern Africa describing it as the “effective collapse of the sector’s project pipeline”. That is where AI comes in. AI is the ultimate shortcut to unlock South Africa’s hidden mineral wealth because it can instantly process decades of unorganised multilayered data. Around the world AI is being used with great success to find mineral deposits for mining, and we must learn from such examples. Earth AI is a mining and exploration startup founded in 2017 in San Francisco that uses the technology to find critical metals such as copper, nickel and cobalt, which are essential ingredients for electric vehicles and renewable energy. According to the company, the global average success rate for mineral exploration drilling is 0.5%. That is a 1-in-200 chance of finding a mineable deposit of minerals.Earth AI says its new systems have a 75% success rate by training AI to think like a geologist. CEO Roman Teslyuk explains: “We train our AI to learn from failures and successes of decades of hundreds of geologists who explored in the past to make much better predictions for where to look for metals in the future.” The company trains its neural networks on more than 50 years of data from failed and successful exploration campaigns, legacy core logs, historical handwritten field journals and geochemical studies. It also collects and uses live satellite imagery and advanced hyperspectral remote sensing data. Its algorithms do not search for surface signs but for subtle geometric changes in rock formations underground that suggest mineral-rich waters once contacted the rock. This process creates a digital “treasure map” that finds promising target areas for the next stage of testing.From years to months Once promising targets are identified, the company has built custom modular and inexpensive automated drilling and sampling technology. Those drills, as wide as a tennis ball, can reach a depth of up to 600m to retrieve rock samples. The process, which could take traditional mining logistics years to secure permits, build roads and move heavy-duty drill trucks, takes Earth AI just a few months from the prediction to the confirmation of the core sample. Then the company can either exploit the deposit itself or sell that information to the large mining companies. In 2024 Earth AI, in partnership with mining company Legacy Minerals, announced the discovery of one of the largest palladium mineral deposits in Australia. According to Global Mining Review, academic and corporate geologists had long held the view that the east coast of Australia lacked the specific rock properties required to host magmatic nickel, copper and platinum group metals like palladium and platinum. Earth AI’s neural networks identified underlying structural patterns in historical data that human observers had missed. Its business model means Earth AI paid for the geological software processing and physical exploratory drilling from its own pocket. In exchange, given that AI made the discovery, Earth AI earns a 3% royalty or the option to secure an 80% stake in the partnership with Legacy Minerals. Another company, KoBold Metals, backed by Jeff Bezos, is building a “digital Google maps for the Earth’s crust” and used its technology to discover a large copper deposit in Chililabombwe, Zambia, more than 1,700m below the surface. This Mingomba project is projected to produce over 300,000 tonnes of copper a year, worth more than $4bn a year. This historically overlooked site now has the potential to boost Zambia’s copper output by 40%. The company found the deposit by analysing 100 years of prospecting data with its AI tools (called TerraShed and Machine Prospector) and is now partnered with the Zambian government to mine the site. According to KoBold’s CEO, as reported by Forbes, the model is especially good at reading old data that nobody bothered to digitise and at noticing patterns no single geologist could hold in their head at once.Legacy records According to KoBold, the industry’s flaw is not a lack of data but rather human limitations in analysing and synthesising all the different types of data gathered over decades. KoBold’s AI analysed legacy records (including historical drill cores from Zambia’s past colonial era), mining permits, satellite imaging and state geophysical surveys. By structuring this unorganised data, its software created a 3D underground probability map of minerals under Zambia’s surface.Guided by the AI, KoBold claims to have drilled more than 120,000m at Mingomba in less than three years — more exploratory drilling than had been completed across Zambia’s entire post-independence history. Construction on the mine has started, and the company aims to extract the first copper for sale by the early 2030s. Given South Africa’s huge known and estimated unknown mineral wealth, using AI to find new deposits and revitalise the country’s mining industry could make it the world’s top mining and minerals power. • Gumede, founder of AI business assistant NowNow, is author of ‘AI Works For You’.