Fifteen of 19 Miners Now Book AI Gains, McKinsey Finds

McKinsey told a mining forum in Colorado Springs that 15 of the 19 major mining companies it tracks reported financial gains from artificial intelligence in the third quarter through Sept. 11, up from four in a prior period.

Artificial intelligence has spent several years in the mining industry as a line item in innovation budgets rather than a line item in results. That is beginning to change, according to work McKinsey presented at a mining forum in Colorado Springs, Colo.: 15 of the 19 major mining companies the consultancy tracks reported financial gains from AI in the third quarter through Sept. 11, up from four in a prior period.

The number itself is modest — 19 companies is a sample, not a census — but the direction is the story. Roughly four-fifths of the tracked group are now describing AI as something that shows up in the numbers, against roughly a fifth before, an illustrative reading of the same two counts. In an industry where capital discipline is the dominant investor theme and where technology spending has historically been justified by safety and compliance rather than margin, that is a meaningful change in how management teams talk about the spend.

From pilot projects to reported results

The distinction that matters in McKinsey's count is between activity and outcome. Almost every large miner has been running AI pilots for years: predictive maintenance on haul trucks, machine-vision grade estimation, drill-and-blast optimisation, concentrator control loops, autonomous haulage dispatch. Pilots are cheap to announce and hard to kill. What separates the 15 from the four is that companies are now attributing financial gains to those systems in their own reporting.

Mining is unusually well suited to this kind of measurement, which is part of why the shift is showing up here before it shows up in some other heavy industries. A mine site is already dense with sensors. Throughput, recovery rates, fleet utilisation, energy consumption per tonne and unplanned downtime are all tracked to the hour and audited by accountants. If a model improves mill recovery or cuts diesel burn per tonne moved, the effect lands in unit cash costs, and unit cash costs are a number every analyst on every quarterly call already interrogates.

That auditability cuts both ways. It is the reason miners can credibly claim a financial gain from AI. It is also the reason investors should ask how much of a reported improvement was AI and how much was a higher-grade section of the orebody, a softer ore blend, or a favourable currency move on an offshore cost base. Grade and mine sequencing can swing a quarter's costs more than any software deployment, and the two are not always easy to separate in a single reporting period.

Where the money actually shows up

AI in mining does not generally create revenue. It defends and extends it. The value tends to sit in four places, all of them cost or volume rather than price:

• Equipment availability. Predicting a failure on a haul truck, shovel or conveyor before it happens converts unplanned downtime into scheduled maintenance, which raises effective fleet capacity without new capital.
• Processing recovery. Small percentage gains in the share of metal pulled out of ore already mined and hauled flow almost straight to the bottom line, because the tonnes have already been paid for.
• Energy and consumables. Diesel, electricity, grinding media and reagents are among the largest controllable inputs at most operations, and are the sort of continuous process variables that control models handle well.
• Exploration and resource modelling. Faster interpretation of drill core, assay and geophysical data shortens the loop between spending on a drill program and deciding whether a deposit is worth pursuing.

That last category is the one with the longest payoff and the least immediate accounting visibility. The first three are what most plausibly sit behind a reported quarterly gain.

Why the timing lines up with a capital-discipline market

The industry context matters. Miners are being asked simultaneously to hold the line on costs, fund replacement of depleting reserves, absorb higher labour and energy bills, and deliver the copper, lithium and critical minerals volumes that electrification and data-centre construction demand. Greenfield supply is slow, expensive and politically difficult. Squeezing more metal out of assets that are already permitted, built and connected to infrastructure is the cheapest tonne available.

Seen that way, AI adoption is less a technology story than a brownfield productivity story with a technology label on it. The report from the Colorado Springs forum, covered by The Northern Miner, lands at a point where investors have limited patience for spending that cannot be tied to output or unit costs.

There is also a competitive dynamic embedded in the jump from four companies to 15. When a handful of operators claim a productivity edge, it reads as a differentiator. When most of a peer group claims it, it becomes table stakes, and the gains get competed away into industry cost curves rather than retained as excess margin. For equity investors, the interesting question shifts from "who is using AI" to "who is getting more out of it than the others, and can they prove it in unit costs over several consecutive quarters."

What to test in the next reporting cycle

The McKinsey count covers the third quarter only through Sept. 11, so it is a partial-quarter read rather than a closed book. The next full set of quarterly filings and operational reviews is where the claim gets stress-tested. Three things are worth watching.

First, durability. A single quarter of improved availability or recovery can be noise. Repeat it across two or three quarters, at multiple sites, and it starts to look structural.

Second, specificity. Companies that quantify the gain — tonnes, percentage points of recovery, dollars per tonne — are making a checkable claim. Companies that describe AI benefits in adjectives are not.

Third, the capital line. AI at scale is not free: it requires data infrastructure, connectivity at remote sites, retained engineering talent and, increasingly, the electricity to run it. If reported operating gains are being funded by a quietly rising technology capex line, the net economics are thinner than the headline suggests.

The wider market backdrop

The broader equity market has been carrying an AI narrative for some time, and it remains supportive. As of the last trade at 17:54 GMT on Sept. 30, 2026, the S&P 500 tracker (NYSEARCA: SPY) was at $767.64, up 0.45% on the day from a prior close of $764.20, while the Nasdaq 100 fund (NASDAQ: QQQ) traded at $743.82, up 0.80% from $737.93. The Dow 30 vehicle (NYSEARCA: DIA) lagged at $511.85, down 0.20% against a prior close of $512.88.

Mining equities have not been the primary vehicle for that trade — the money has gone to chip designers, hyperscalers and power suppliers. What the Colorado Springs presentation suggests is a second-order version of the same theme: not miners selling AI, but miners using it to hold down the cost of producing the metals the AI build-out consumes. That is a slower, less glamorous claim. It is also the kind that turns up in cash flow rather than in a multiple.

Key facts

• Miners reporting AI financial gains: 15 of 19 tracked by McKinsey
• Prior period comparison: Four of the same 19 companies
• Measurement window: Third quarter through Sept. 11
• S&P 500 tracker (SPY): $767.64, +0.45%, as of 17:54 GMT Sept. 30, 2026

Frequently asked questions

What exactly did McKinsey report at the mining forum?

McKinsey said that 15 of the 19 major mining companies it tracks reported financial gains from artificial intelligence during the third quarter through Sept. 11. That compares with four companies in a prior period. The findings were presented at a mining forum held in Colorado Springs, Colorado, and reported by The Northern Miner.

Which mining companies are in the group of 19?

The individual companies were not identified in the available reporting. McKinsey described a tracked set of 19 major mining companies without naming them, so investors cannot yet map the 15 reporting gains to specific tickers. Company-level detail would need to come from each miner's own quarterly operational reporting.

How does AI actually save a mining company money?

Most of the value sits in costs rather than revenue: predicting equipment failures to convert unplanned downtime into scheduled maintenance, lifting the share of metal recovered from ore already mined, trimming diesel, electricity and reagent consumption per tonne, and speeding up interpretation of drill and assay data during exploration.

Why is mining measuring AI gains earlier than some other industries?

Mine sites are already heavily instrumented. Throughput, recovery rates, fleet utilisation, energy per tonne and downtime are tracked continuously and feed directly into unit cash costs, a figure that is audited and scrutinised every quarter. That makes it easier to attribute a financial effect to a specific operational change.

Should investors treat these reported gains as proven?

With caution. Grade variation, ore blend and mine sequencing can move a single quarter's costs more than any software deployment, and the McKinsey window covers only part of the third quarter. Durability across several quarters and at multiple sites, plus quantified disclosure, is what would make the claim convincing.

Does this make mining stocks an AI investment?

Not directly. Miners are not selling AI products; they are using AI to lower the cost of producing metals such as copper and lithium that the AI build-out consumes. If most of a peer group achieves similar gains, the benefit tends to shift into industry cost curves rather than stay as excess margin for individual companies.

Sources

• Mining Forum: AI begins paying off, McKinsey says — Northern Miner

Photo: Nikiko · Pixabay Content License — source