How Mining Companies Are Using AI in Exploration and Operations

From KoBold's Zambian copper find to 1,000 autonomous haul trucks, a look at what AI does in mining today and what it does not change.

Key takeaways

Komatsu commissioned its 1,000th ultra-class autonomous haul truck in April 2026, with its autonomous fleet moving over 11.5 billion tonnes.
KoBold Metals says AI sped up exploration at Mingomba in Zambia, where work began in April 2026 with first copper targeted for the early 2030s.
A 2025 peer-reviewed review found greenfield mineral discovery rates unchanged despite heavy investment in AI and machine learning.

Mining companies have used automation and data for years, and AI is now part of exploration targeting, haulage and processing. The public examples show real gains in how equipment runs and how data is sorted. They also show what AI has not changed: the time and money it takes to turn a discovery into a mine.

Exploration: better targeting, same drill bit

The best-known example is KoBold Metals, a privately held exploration company. In January 2025 it raised US$537 million in a Series C round at a post-money valuation of US$2.96 billion, led by Durable Capital Partners and T. Rowe Price funds, according to Mining Engineering.

KoBold’s headline project is Mingomba in Zambia. The company acquired it in December 2022 and says it used its proprietary AI to speed up exploration there, finding a highly concentrated copper resource about 1,700 metres below surface. In April 2026, MINING.COM reported KoBold had started work on the mine, with capital costs of more than US$2.3 billion, plans for over 300,000 tonnes of copper a year and first output in the early 2030s.

What AI does in exploration is sort data. Models combine geophysics, geochemistry, satellite imagery, old drill logs and geological maps, then rank where a deposit is most likely to be. That can narrow a search area and help decide where to drill next.

What it does not do is replace the drill. A 2025 review in a peer-reviewed journal by Davies and co-authors noted that “greenfield mineral deposit discovery rates remain unchanged” despite heavy investment, and that machine learning “struggles with low-quality data and inconsistent sampling” outside established mining areas. The authors argued that relying on machine learning alone “overlooks the critical role of human creativity,” and recommended pairing data scientists with geoscientists.

Haulage: the most mature use

Autonomous haul trucks are the part of mining automation with the longest track record.

Komatsu said on April 21, 2026 that it had commissioned its 1,000th ultra-class autonomous haul truck running its FrontRunner system. Its autonomous fleet had moved more than 11.5 billion tonnes of material. Komatsu first brought a commercial autonomous haulage system to market in 2008. Sites named in its release include Barrick’s Nevada Gold Mines.
Caterpillar reported in July 2020 that it had 276 autonomous trucks in operation, which had hauled more than 2 billion tonnes and driven 67.6 million kilometres without a lost-time injury. That is the most recent figure we could confirm from a published source.
Rio Tinto runs AutoHaul, an autonomous heavy-haul rail system in Western Australia’s Pilbara region that became fully operational in June 2019. Each train is about 2.4 kilometres long and carries around 28,000 tonnes of iron ore, controlled from an operations centre in Perth more than 1,500 kilometres away.

Autonomy in haulage is mostly about consistency. Trucks and trains running on software keep steadier speeds, take fewer unplanned stops and remove people from some of the most dangerous tasks on site. The companies report safety and productivity benefits, though most do not publish a single standard metric that allows direct comparison.

Processing: small percentages on big volumes

In May 2023, BHP and Microsoft announced they were using AI at its Escondida copper mine in Chile to improve copper recovery in the concentrators. The system uses real-time plant data and machine learning on Microsoft’s Azure platform to recommend adjustments to operating variables that affect ore processing and recovery.

BHP did not publish a recovery figure in that release. The logic is scale. In a plant processing very large tonnages, a fraction of a percentage point of recovery produces meaningful extra metal without mining more ore.

By the numbers: what AI does not shorten

The Mingomba timeline shows the gap between finding a deposit and producing from it, even for a company built around AI:

Step	Date	Source
KoBold acquires Mingomba	December 2022	MINING.COM
Series C, US$537 million	January 2025	Mining Engineering
Work begins, capex over US$2.3 billion	April 2026	MINING.COM
First copper targeted	Early 2030s	MINING.COM

From acquisition to first targeted output is roughly eight to ten years. For the industry as a whole, S&P Global Market Intelligence reported in October 2024 that the average lead time from discovery to production was almost 18 years for mines that started up in 2020 to 2023.

Put another way, AI can help find a deposit faster. It does not issue permits, finance a US$2 billion build, or sink a shaft 1,700 metres. Those steps set the pace.

What changes and what does not

Based on the public record:

Changes: how fast large data sets are screened, how drill targets are ranked, how trucks and trains are dispatched, and how plants are tuned in real time.
Does not change: the geology itself, the need to drill and assay, permitting timelines, construction costs, and commodity price cycles.
Still uncertain: whether AI-led exploration raises the industry’s overall discovery rate. The 2025 review found greenfield discovery rates unchanged so far.

For readers following mining companies, a claim that a company “uses AI” says little on its own. The useful questions are what data it applies AI to, what the model has predicted, and what later drilling or production results showed.

What to watch

KoBold’s published updates on Mingomba construction and financing.
Komatsu and Caterpillar autonomous fleet milestones and customer announcements.
Large miners’ annual and sustainability reports for quantified results from AI in processing and maintenance.
Peer-reviewed studies measuring discovery rates for AI-guided exploration against conventional programs.
S&P Global Market Intelligence updates on mine lead times.

Sources

Frequently asked questions

How is AI used in mineral exploration?

In exploration, AI sorts data. Models combine geophysics, geochemistry, satellite imagery, old drill logs and geological maps, then rank where a deposit is most likely to be. That can narrow a search area and help decide where to drill next. It does not replace the drill, and a 2025 review found greenfield discovery rates unchanged.

How many autonomous haul trucks are used in mining?

Komatsu said on April 21, 2026 that it had commissioned its 1,000th ultra-class autonomous haul truck running its FrontRunner system, and that its autonomous fleet had moved more than 11.5 billion tonnes. Caterpillar’s most recent confirmable figure, from July 2020, was 276 autonomous trucks in operation, which had hauled more than 2 billion tonnes.

Does AI make it faster to build a mine?

AI can help find a deposit faster, but it does not issue permits, finance construction or sink a shaft. KoBold Metals acquired Mingomba in December 2022 and targets first copper in the early 2030s, roughly eight to ten years. S&P Global reported an average of almost 18 years from discovery to production for mines started in 2020 to 2023.