AI Mining Companies: 8 Examples, What Each One Runs

AI mining companies are metal and mineral miners, explorers and equipment makers that use machine learning, computer vision and automation to find, dig, process and haul ore. They are not crypto miners. The best documented examples are Rio Tinto and Fortescue (driverless haul trucks), BHP and Freeport-McMoRan (AI advice for copper concentrators), KoBold Metals and Earth AI (AI-led exploration), and Komatsu and Caterpillar (the autonomous truck suppliers). Rio Tinto alone ran 300 Komatsu autonomous trucks at 10 Australian sites by August 2024 (AlCircle, August 2024).

This page is written for licence and claim owners, consultants and investors who want to know which companies really run AI, what each one does with it, and what the sources say it delivered. Every company below is covered with facts from company releases, filings or trade press, with the date next to each figure. It is information, not investment advice, and it names no share prices.

Which companies are AI mining companies?

The label has no legal or industry definition, so it helps to split the field into three groups. Each group uses AI for a different job and makes money in a different way.

• Miners that add AI to existing mines. Rio Tinto, Fortescue, BHP, Freeport-McMoRan and Nevada Gold Mines (operated by Barrick) already own producing mines. They use AI to move rock with fewer people on site, or to squeeze more metal out of the same mill.
• AI-first explorers. KoBold Metals and Earth AI use models trained on geology, geochemistry, geophysics and drill data to pick where to drill. Their product is a discovery, not a faster truck.
• Equipment and software suppliers. Komatsu and Caterpillar sell autonomous haulage systems to the miners above. A miner can be an AI user without building any AI itself.

Copper is where AI is easiest to measure, because ore grades are falling and mills are already instrumented with thousands of sensors. Chile mined an estimated 5.3 million tonnes of copper in 2025 out of about 23 million tonnes worldwide (USGS Mineral Commodity Summaries, 2026), so a one or two percent gain at one large concentrator is a real tonnage. That is why most of the quantified examples below are copper mills and copper projects.

How we chose the eight examples

We did not rank these companies, and no sourced metric would rank them fairly: a driverless fleet, a flotation model and a discovery engine measure different things. Instead, each company is on the list because it meets three tests.

1. It names a real site or project, not only a strategy slide.
2. The facts come from the company, a regulator filing, a supplier release or established trade press that quotes them.
3. The claim is about AI, machine learning or autonomy in metal or mineral mining.

Other companies also use AI, and the list is not complete. Where a number is a company forecast or a supplier’s own claim, the text says so. Figures are dated, and the table below shows the one-line version before each company gets its own section.

AI mining companies compared: what each one runs

1. Rio Tinto: autonomous haulage across the Pilbara

Rio Tinto began trialling Komatsu’s autonomous haulage system (AHS) in 2008 and signed a memorandum in 2011 for 150 trucks. Komatsu delivered its 300th AHS truck to Rio Tinto’s Pilbara operations in August 2024. Komatsu said the Rio Tinto fleet had run 8.9 million operating hours and moved more than 4.8 billion tonnes, and that it had lifted effective utilisation (the share of time a truck does paid work) by 15 per cent (AlCircle, August 2024).

Chart

Rio Tinto’s autonomous trucks, first delivery to 300th

Komatsu AHS trucks at Rio Tinto’s Australian operations

The trucks work across 10 Australian sites and are reported to contribute to about 80 per cent of daily production capacity. These are supplier-reported numbers, so treat the 15 per cent as Komatsu’s claim about its own system. The reason miners adopt it is plain: a truck with no driver can run through shift changes and in places that are unsafe for people.

2. Fortescue: a billion tonnes without drivers

Fortescue was an early adopter of Caterpillar’s Command for hauling system at its iron ore mines in Western Australia. In November 2019 it reported 1 billion tonnes moved with the technology. The autonomous fleet had travelled more than 33.5 million kilometres without a single safety incident, and Caterpillar said autonomous haulage had delivered a 30 per cent improvement in productivity at the site (Canadian Mining Journal via MINING.COM, November 2019).

Caterpillar describes the trucks as responding to calls from the shovel, driving to the dump point and reporting for maintenance on their own. The figures are from 2019, so they show what the system did early on, not what Fortescue’s fleet does today.

3. BHP: AI advice for the world’s largest copper mine

In May 2023 BHP and Microsoft announced a project to combine real-time plant data from the Escondida concentrators in Chile with AI recommendations on Microsoft’s Azure platform. Operators could then adjust the variables that drive ore processing and recovery. BHP’s chief technical officer said falling grades and fewer new discoveries mean AI and machine learning will have to deliver more production from existing mines (Global Mining Review, May 2023).

BHP’s Form 20-F, published in September 2023, said that increased copper recovery at Escondida had been achieved by using AI recommendations and new machine learning and data platforms to tune flotation settings. Escondida produced 5 per cent more in the fiscal year to June 2023, but the report does not credit the whole increase to AI (Bloomberg via MINING.COM, September 2023).

4. Freeport-McMoRan: machine learning in the mill

Freeport-McMoRan’s Bagdad mine in Arizona is a well-documented published case of AI at a mill. In 2018 Bagdad staff worked with McKinsey data scientists on a model that read three years of data from thousands of sensors at the concentrator. Staff had treated all ore entering the mill as one type. The sensors showed seven distinct ore types, and the mill’s controls did not suit all of them (International Copper Association, July 2020).

Chart

Freeport’s AI copper gain: result vs forecast

Tonnes of copper; Bagdad result and Americas forecast as reported

The team found that tuning the controls to each ore type could lift copper output by at least 10 per cent, and that changing the flotation pH would recover more copper. ICA wrote that Bagdad was boosting its production by 9,000 metric tonnes (the source gives no period). Freeport extended the programme to Morenci, planned it for Cerro Verde in Peru, and forecast that across its North and South American mines it could raise production by 5 per cent, or 90,000 tonnes. Those last two figures are company forecasts, not results.

5. KoBold Metals: AI-led exploration and a mine build

KoBold Metals describes its approach as AI plus human intelligence. It feeds satellite images, drilling reports and other geological data into models that predict where metals such as copper, lithium, cobalt and nickel may sit. In 2025 it announced a US$537 million Series C round at a US$2.96 billion valuation, which took its total raised to about US$1 billion (Pulse 2.0, January 2025).

Its flagship is Mingomba, a copper project in Zambia that it has worked on since late 2022. On 29 April 2026 KoBold and ZCCM-IH broke ground on shaft construction. KoBold puts the investment above US$2 billion, with about 300,000 tonnes of copper a year targeted and first output expected in the early 2030s (KoBold Metals, May 2026). Whether the AI shortened the path to discovery is the company’s claim. What is on record is the money raised and the construction start.

6. Earth AI: software that flags prospects, then drills them

Earth AI, founded in 2017, combines AI discovery software with its own drilling technology so that it can test targets itself. In March 2025 it announced six new prospects with tungsten, cobalt and gold on its Australian tenements. Samples from the Kooranjie project in New South Wales returned up to 0.26 per cent tungsten, and the Elkedra project in the Northern Territory returned up to 0.39 per cent cobalt, 1.39 per cent copper and 0.685 grams per tonne gold. Earth AI says it can go from detecting a prospect to a drill test in three to six months, and it planned a drilling programme from May 2025 (MINING.COM, March 2025).

Sample assays like these show that a target is worth drilling. They are not a mineral resource, and the company’s own next step was drilling.

7. Komatsu: the supplier behind the driverless fleets

Komatsu says its FrontRunner Autonomous Haulage System was the first commercial autonomous mining solution, launched in 2008. In April 2026 it commissioned its 1,000th autonomous ultra-class truck, a 930E-5AT with a 290-tonne payload, at Barrick’s Nevada Gold Mines in the United States. The company says FrontRunner customers have moved more than 11.5 billion tonnes of material across North America, South America, Australia and Europe (International Mining, April 2026).

Chart

Komatsu’s 1,000 autonomous trucks by model

Trucks commissioned with FrontRunner, April 2026 (about 52% / 24% / 24%)

Komatsu’s own split of the 1,000 trucks shows how concentrated the fleet is: about 52 per cent are 930E models and 24 per cent each are 830E and 980E models. Autonomous water trucks are the next machine type the company is adding.

8. Caterpillar: Command for hauling and MineStar

Caterpillar’s Command for hauling is an autonomous haulage system sold to miners, and its public result used in this review is Fortescue’s: the 2019 report of 1 billion tonnes and a 30 per cent productivity gain described above. In that report, Caterpillar says Command trucks respond to calls from the shovel, haul to the dump point and report for maintenance without an operator. Because the figures come from Caterpillar and its customer, the right reading is a supplier-reported result at one site, not an industry average.

AI or automation? What the label means

Not everything in the table is AI in the strict sense. Driverless trucks follow sensors, maps and fleet-control software, and the machine learning inside them is only part of the package. A flotation advice model like BHP’s or Freeport’s is closer to classic machine learning on plant data. An exploration engine like KoBold’s or Earth AI’s predicts where metal is likely to be from geology and drill data. Treat the three as separate products.

A useful way to tell them apart is by what the buyer gets: a safer, steadier haul cycle, a higher recovery percentage, or a short list of drill targets. The kind of evidence also differs. Autonomy is proved in tonnes moved and hours run. Recovery models are proved in plant tonnes of metal. Exploration AI is proved only when the drill hits and a resource is later reported to a recognised code. Academic reviews also note that the data work is the hard part: a 2025 review of machine learning for mining energy use covers forecasting, anomaly detection and the practical problems of putting models into plants (Frontiers in Energy Research, May 2025).

▶ What an AI mining app does, who makes one and why it is not crypto. · 3:52

Which mining companies benefit most from AI?

No source ranks miners by AI benefit, so the honest answer is a pattern, not a name. The cases in this page point to three kinds of operation where the return is easiest to see:

• Large copper and base-metal concentrators. A recovery gain of even one percentage point on a mill that treats tens of thousands of tonnes a day is a large amount of metal. Bagdad and Escondida are examples.
• Large open pits with long haul cycles. Iron ore in the Pilbara and gold in Nevada both use driverless fleets because trucks run for hours on fixed routes.
• Explorers with large areas and little drilling. AI ranking is most useful when a company holds a big ground package and can only afford to drill a few holes.

Small mines with short haul roads and a single ore type gain less. Many of the miners that buy AI do so through suppliers, so the supplier often carries the AI risk. For investors this means the supplier or the miner can be the AI company; that is a fact about business models, not a recommendation. This page does not suggest buying or selling any listed company.

AI target generation in exploration: what the tools do

Target generation is the exploration job where AI is most often sold to licence owners. The tool reads maps, satellite scenes, geophysics, geochemistry and old drill logs, learns where known deposits sit relative to those layers, and ranks the rest of the ground by similarity. The output is a prospectivity map: a colour surface, not a resource estimate.

Three limits matter for anyone paying for such a tool. First, the model is only as good as the number of known deposits it learns from, and in a new district there may be very few. Second, a high score means the ground looks like known mineralised ground, not that metal is there. Third, only drilling and a resource report prepared under a standard such as JORC or NI 43-101 turn a target into an asset. For the basics of how drilling confirms a target, see our guide to the main types of exploration drilling.

Satellite data is an input that needs no site access. Farmonaut maps alteration and structure from satellite imagery on a licence boundary and delivers exploration targets, ahead of any resource estimate, which makes it a way to screen ground before committing a drilling budget.

▶ Six kinds of AI geology tools and three questions to ask before choosing one. · 3:18

Metals mining automation beyond trucks

Haul trucks get the headlines, but automation covers more of the mine. Komatsu has added autonomous water trucks to its fleet. Plants use sensor data for predictive maintenance, which means servicing a machine when its vibration or temperature pattern shifts instead of on a fixed calendar. Drill and blast teams use software to check hole position and fragmentation, which we cover in our list of drill and blast monitoring software.

The common thread is data. The Bagdad case needed three years of sensor history before the model could find seven ore types. A mine with few sensors and a patchy history cannot buy that result off the shelf.

What a recovery gain is worth: a calculator

The Freeport and BHP examples both come down to the same sum: tonnes milled, head grade, and the extra share of the metal that ends up in concentrate. The calculator below lets you test your own mill. The default numbers are round examples, not a benchmark for any mine. Prices are inputs, so enter today’s price.

How to check an AI claim from a mining company

Whether you are a licence owner buying a service, a consultant reviewing a vendor or an investor reading a release, five questions sort real AI results from marketing:

1. Is a site named? “AI-driven efficiency” means little. “Concentrator at Bagdad, Arizona” can be checked.
2. Is the gain in physical units? Tonnes of copper, percentage points of recovery or tonnes hauled beat “up to 50 per cent better”.
3. Is it a result or a forecast? Freeport’s 9,000 tonnes was a result; the 90,000-tonne figure was a forecast.
4. Who measured it? A supplier’s own utilisation figure is weaker evidence than a figure in a mining company’s audited annual report.
5. What is the baseline? A gain against last year’s mill is different from a gain against a mill that was being run badly.

The same questions apply to exploration AI: ask which deposits the model learned from, whether it was tested on ground it had not seen, and what drilling followed. A prospectivity map is a screen. Only drill results and a resource report made to a recognised code confirm the metal.

Where satellite data fits

For licence owners and consultants, satellite imagery is the part of the AI toolkit that needs no site access. It screens a boundary for iron-oxide staining, clay alteration and structural lines, and it can do that before you pay for ground crews. Farmonaut takes a KML, KMZ or polygon boundary and returns a satellite mineral detection report with ranked exploration targets; it is a screen that sits ahead of drilling and ahead of any resource estimate, and it makes no claim of JORC or NI 43-101 compliance. You can request a quote through our mining quote form, or read how satellite-based mineral detection works. Our guide to the largest copper producers lists the operators most likely to run mill AI.

▶ A quick look at Farmonaut Mining Pro, which turns a licence boundary into a satellite-based picture. · 4:56

Frequently asked questions

What are AI mining companies?

They are companies that use artificial intelligence, machine learning or autonomous machines in metal and mineral mining. They fall into three groups: miners that add AI to existing mines, AI-first exploration companies, and equipment suppliers. The term does not mean cryptocurrency mining.

Which AI mining companies are the best known?

On the public record the best-documented names are Rio Tinto and Fortescue for driverless haulage, BHP and Freeport-McMoRan for AI in copper concentrators, KoBold Metals and Earth AI for AI-led exploration, and Komatsu and Caterpillar as suppliers.

Which mining companies are expected to benefit most from AI?

No published source ranks them. Operations with large copper concentrators, long open-pit haul cycles or big unexplored ground packages show the clearest returns in the cases above. This is not investment advice.

How is AI used in the mining industry?

In four main ways: target generation in exploration, driverless haulage, process control in mills (for example flotation settings), and safety and equipment monitoring. Each is proved by a different kind of evidence.

Does AI find mineral deposits?

AI ranks ground by how much it resembles known deposits. It produces drill targets, not discoveries. Only drilling, followed by a resource estimate under a recognised code, confirms a deposit. KoBold and Earth AI both pair their models with drilling for that reason.

How much copper can AI add to a mill?

At Freeport’s Bagdad mine, about 9,000 tonnes of extra copper was reported in 2020 after a machine-learning model changed mill controls and flotation pH. The company forecast 90,000 tonnes across its American mines. Your own mill will differ.

Are driverless trucks AI?

They are autonomous machines that run on sensors, maps and fleet-control software, with machine learning as one part. Komatsu had commissioned its 1,000th autonomous ultra-class truck by April 2026, and Rio Tinto ran 300 of its trucks by August 2024.

Can a small miner or claim owner use AI?

Yes, in exploration. Satellite-based screening and prospectivity mapping work on a licence boundary and do not need a plant full of sensors. Mill and fleet AI need data and capital that small operations often lack.

How is AI used in exploration target generation?

The tool combines geology, geophysics, geochemistry, satellite and old drill data, learns what known deposits look like, and ranks the rest of the ground. The result is a prospectivity surface that guides where to sample or drill.

Is AI in mining safe and regulated?

Autonomous fleets are used to remove people from hazardous areas, and Fortescue reported 33.5 million kilometres without a safety incident by November 2019. Regulators still treat a mine as a licensed operation, so check the rules in your jurisdiction.

Not investment advice. Company facts are from the sources linked above and are dated.