The energy transition has an insatiable appetite for critical minerals. And it’s an appetite that’s only growing; between now and 2050, demand for the minerals and metals most key to decarbonization is projected to skyrocket.
According to a report released this week by Ernst & Young, demand for graphite and cobalt is set to increase by well over 200%; demand for lithium, by 910%; and for rare earths, by 943%.
The good news for clean tech is that there’s likely enough to go around; the earth’s crust has no shortage of the critical minerals to power the energy transition. However, both exploration and processing are highly geographically concentrated, primarily in China. Plus, finding critical minerals (including discovering new ones) is very difficult, and very expensive.
Like other industries looking to move faster, the critical minerals mining sector has been increasingly artificial intelligence-curious. AI could, in theory, aid the industry in finding both new deposits of the most sought-after minerals, and entirely new materials. That’s a potential that has kept money flowing to early-stage AI solutions throughout 2023, despite a tight investment market.
In March, AI-based mineral asset generator VerAI raised $12 million for a Series A. In June, GeologicAI raised $20 million for its “core scanning robot,” also a Series A. Later that month, Berkeley-based KoBold Metals raised $195 million, in a round whose investors included T. Rowe Price, Andreessen Horowitz, and Breakthrough Energy Ventures.
Last week, Google joined the lineup of players betting AI can speed up operations far up in the minerals supply chain, introducing the DeepMind Graph Networks for Materials Exploration, a deep learning tool for predicting the stability of new materials. Of the 2.2 million predictions made by GNoME, Google said 380,000 are particularly promising for experimental synthesis.
“Among these candidates are materials that have the potential to develop future transformative technologies ranging from superconductors, powering supercomputers, and next-generation batteries to boost the efficiency of electric vehicles,” the company said.
AI’s potential role in mineral exploration is broad, and the current offerings each take a slightly different approach. GNoME, for example, is a graph neural network model trained with data on the structure and chemical stability of crystals. It identifies new minerals that have similar structures to known materials, and could therefore replace highly in-demand minerals like, say, lithium relatively easily.
Meanwhile, investor darling KoBold uses machine learning and geological data to model the sub-surface and predict where mineral deposits are likely to be found. Founded in 2018, the company has scaled quickly, in part because its business model doesn’t stop at software. Instead, it makes strategic investments on land claims, and then sells licenses to mine operators. To date, KoBold says it has more than 60 mining projects globally.
Other startups use machine learning to search through geologic data to identify promising mineral deposits, or even develop robots capable of scanning and analyzing rock samples themselves.
For clean energy companies in the United States in particular, there’s a tight timeline on ramping up mineral extraction and processing outside of China, thanks to Biden administration guidelines for Inflation Reduction Act tax credits.
Under guidance released in early December, for example, automakers have until 2025 to eliminate critical minerals that were extracted, processed, or recycled by “foreign entities of concern” if they want to qualify for lucrative benefits.
Prompted in part by the Biden administration’s efforts to decrease reliance on China, the country is likely to have a critical mineral shortage by 2030 or 2035, said Tom Moerenhout, a research scholar and adjunct professor at Columbia University.
And while processing capacity is something that can be ramped up rather quickly, he added that activity further upstream, like exploration, tends to move very slowly. According to the International Energy Agency, that phase takes an average of 12.5 years, depending on the mineral.
The vast majority of known, untapped domestic deposits in the U.S. — 97% of nickel, 89% of copper, and 79% of lithium — are located near or within Native American reservations, where companies face opposition to the development of mines that could destroy culturally and environmentally important land. In Arizona, for example, mining giant Rio Tinto has been locked in a decade of debate over a copper deposit located under an Apache religious site.
Because of the permitting and legal challenges associated with those deposits, Moerenhout said there’s been talk at the federal level of conducting “another big exploration round, starting with areas that are actually much more easy to permit, with less of the environmental and social implications than current projects have.”
That means the U.S. needs to find new mineral deposits, and do it fast. And while there haven’t been major technological innovations in mineral exploration in decades, Moerenhout said AI has been a major topic for the last several years, particularly among so-called “junior miners” that are smaller than stalwarts like Rio Tinto and are focused on one specific mineral.
For these junior miners, he added, the potential of AI-driven mineral discoveries is massive. Done the traditional way, exploration is a billion-dollar undertaking, one that often lacks immediate returns.
AI could help shorten the timeline of exploration and decrease the risk, Moerenhout said. That means bringing down the cost. And in the case of GNoME, the tech could allow miners to home in on higher-quality ore, which would allow for easier production and processing.
“All of this stuff is still being piloted, being developed as we speak,” he added. “But if you can develop that type of tech, you can almost leapfrog some of the challenges associated with exploration. The potential is quite huge.”
