What Mineral Exploration Actually Costs — And Why AI Is Rewriting the Math

$2.4 million. That's the median spend just to prove a greenfield copper prospect worth drilling — before you've pulled a single core sample. And 99 out of 100 of those prospects never become mines.

I've been staring at those numbers for months now, mostly because a friend of mine runs a junior exploration outfit in Chile and keeps asking why his investors are getting cold feet. The answer isn't complicated. Money got expensive, commodity cycles got weird, and the old playbook — geologists with rock hammers, followed by helicopters, followed by hope — stopped clearing the bar for institutional capital.

So here's what I want to walk through: where the money actually goes in exploration, why traditional methods keep bleeding capital, and what the AI-plus-satellite crowd is doing differently. Some of it's hype. Some of it isn't.

Where the money actually goes

Break down a typical early-stage exploration budget and you'll find drilling eats the biggest slice — usually 45 to 60 percent. Geophysics and geochemistry take another 15 to 20. Then there's camp logistics, permits, community engagement, environmental baseline work, salaries, and the endless line item everyone underestimates: mobilization.

Moving a drill rig into a remote part of Zambia or northern Peru can cost $80,000 before it turns a single meter. If your targeting was wrong — if you drilled the anomaly that turned out to be a magnetite blob and not the copper porphyry you hoped for — that money's gone. All of it.

And this is the part investors keep asking about. The traditional discovery cost per ounce of gold has climbed to roughly $58 in recent industry surveys, up from around $22 two decades ago. Copper's worse. You're spending more to find less, in harder places, with regulators watching every step.

Honestly, the industry's dirty secret is that most exploration budgets are consumed validating targets that a better upfront analysis would've killed on day one. That's the inefficiency AI is chewing on.

Where satellite intelligence changes the equation

A good satellite spectral survey doesn't replace boots on the ground. Anyone who tells you it does is selling something. But it does something arguably more valuable — it kills bad targets cheaply.

Here's the thing. If you can rank 400 anomalies by probability before you fly a single helicopter, and if that ranking is genuinely better than random, you've just cut your discovery cost dramatically. Not by finding more deposits. By walking away from more duds earlier.

Companies working in this space — including GeoMine AI, which uses spectral analysis across multi-band satellite imagery to fingerprint alteration zones — are essentially selling a probability filter. Their breeze geo mineral analysis workflow (their term, not mine) layers hyperspectral data with structural geology and machine learning models trained on known deposits. The output isn't "drill here." It's "these 12 targets, ranked, with confidence scores and here's why."

For a junior with $6 million in the treasury, that ranking is the difference between drilling three real prospects and drilling nine coin flips.

I used to think this stuff was mostly marketing polish on old remote sensing techniques. Then I saw the discovery numbers from a few operators using it in the Tethyan belt and West Africa. The hit rate on second-pass drilling — the expensive part — is measurably higher. Not miraculous. Just meaningfully better. Enough that risk-adjusted returns start looking sane again.

What investors are actually asking now

I had a call last month with a family office analyst who's evaluating three junior miners. His question wasn't about grade or tonnage or even management team. It was: "What's their targeting methodology, and can they show me the false-positive rate?"

That's new. Five years ago mining investment memos talked about geological intuition and "the nose" of a chief geologist. Now they're talking about model precision and recall. AI in mining has quietly forced a vocabulary shift in how due diligence gets done.

A few things I'd push any mining investor to actually ask before writing a check:

• What's the exploration cost per target advanced to drilling, and how does that compare to the peer group?
• If they're using satellite intelligence, what training data sits behind the model? A model trained on Andean copper won't do great in Australian nickel-sulfide terrain.
• How much of the budget is committed versus optioned? Optionality matters when commodity prices swing 30% in a quarter.
• What's the community and permitting overhead? Because in 2025 you can have perfect targets and still lose 18 months to a social license fight.

The answers tell you whether management understands their own risk stack or whether they're just running the old playbook with a new coat of paint.

Where I think this actually lands

Exploration isn't getting cheap. Anyone promising that is lying to you or to themselves. Drilling still costs what it costs, permits still take what they take, and the deposits left to find are deeper and messier than the ones our grandfathers pulled out of Nevada.

But the risk profile is genuinely shifting. When you can spend $150,000 on a spectral survey and satellite-driven targeting workflow and eliminate 70% of your candidate anomalies before mobilization, the math on a $30 million exploration program starts to work again. Not for every project. Not for every commodity. But often enough that the capital's coming back, cautiously, to juniors that can show a disciplined, data-first targeting story.

The geologist with the rock hammer isn't going away. She's just going to spend more of her career reviewing model outputs and less of it flying to anomalies that were never going to pan out.

Which, if you've ever sat through a $2 million drilling program that hit nothing but barren quartz, sounds like progress to me.

What would you want to see in a targeting report before you'd commit capital?