A group of investors is betting that the $1.6tn global mining industry is set to undergo the sort of digital disruption that upended the media, music and automotive industries.
T Rowe Price, Bond Capital and a dozen other investors have raised $192mn for a Bill Gates-backed start-up called Kobold Metals, which uses artificial intelligence and machine learning to find new deposits of critical metals needed for batteries and clean energy.
It comes as demand for battery metals such as lithium and nickel is expected to surge as electric vehicles become mainstream over the next two decades.
Silicon Valley-based Kobold estimates that more than $10tn of lithium, cobalt, nickel and copper needs to be mined to meet the coming demand for electric vehicles — a figure that will see a growing number of traditional mining companies turn to AI to help with the challenge.
BHP, the world’s biggest miner, and Norway’s state-backed energy company Equinor have already partnered with Kobold, and their venture capital arms are both investors.
In an interview with the FT, Kurt House, chief executive at Kobold, said mining discoveries have been getting slower and more expensive over time, reflecting what he called Eroom’s Law — the opposite of Moore’s Law.
“In the last 30 years, the number of discoveries per dollar of exploration capital has declined by six times,” he said. “So if you increased your budgets by six times, you’re going to find things at the same rate you found them in 1990.”
Low-hanging fruit locations where minerals can be seen from the surface have largely been found while exploration for harder-to-find minerals is chronically underfunded.
For example, BHP paid $15bn in dividends last year, but spent only $75mn on exploration.
Named after the German word for a goblin that controls the earth’s minerals, Kobold was already backed by Andreessen Horowitz and Bill Gates’s Breakthrough Energy Ventures fund.
Its latest funding round came after the start-up was able to demonstrate that its technology had successfully predicted the composition of bedrock in northern Quebec, finding valuable minerals in an area dismissed by conventional approaches as “non-prospective.”
Kobold collects vast troves of historical and scientific data and uses algorithms to identify where mineral deposits might be below the earth’s surface.
Its technology includes machine learning tools to sift through 20mn pages of documents in the public domain — including two centuries worth of mining rights’ agreements in countless jurisdictions — that it sorts, digitises, and streamlines into accessible information.
House said much of this “dark data” has been forgotten or unused. His team hired someone to go through state archives in Zambia where they found hand-painted maps on linen from the 1920s, covering the whole country and describing the land and any outcropping bodies.
“We have fully digitised them and now we can look at the data through spectral satellites,” House said. “We have dozens of examples like this.”
These methods are helping Kobold build a “Google Maps” of the Earth’s crust, House said. Once the company has a sense of where valuable minerals have been overlooked, it sends teams in to collect more data.
In northern Quebec, a team of six took data and rock samples at 839 locations along 142km of traverses. The exploration included 12 weeks of manning a helicopter equipped with an enormous metal detector — 115 feet across and 1,700 pounds — that would send out electromagnetic pulses a third of a mile into the earth in the search for minerals.
Kobold is, in some sense, a sophisticated real estate play. Once it has done the work, it buys the mining rights and will form partnerships with mining majors to split the revenue. In northern Quebec, it purchased the right to nearly 200,000 acres of land.
But the work does not always result in a gold mine of metals. Even after they have done months of work on a location, there is still “a reasonable chance” that no minerals can be mined profitably. In Quebec, for instance, House said “the potential (value of minerals at the site) is from zero to multiple billions.”
“Our objective is a 20 per cent success rate — that would be way better than standard practice,” said House.
Connie Chan, a partner at Andreessen who has backed two funding rounds for Kobold, said the proof points behind its methods convinced her that mining is “ripe” for digital disruption, especially given the need for new discoveries to make electric vehicles a reality.
“These aren’t rare metals,” she said, referring to lithium, nickel, cobalt and copper. “The crazy thing is that cobalt is as common as tin, but until Tesla stepped on to the stage we just didn’t need much of it, so no one had the financial incentive to go explore for it.”