The Technology Gap in Critical Minerals Exploration

The supply problem is an exploration problem

The energy transition requires large quantities of copper, lithium, cobalt, nickel, manganese, graphite, and rare earth elements. International Energy Agency projections indicate that demand for these minerals could increase by a factor of four to six by 2040 under net-zero scenarios. (IEA, The Role of Critical Minerals in Clean Energy Transitions, 2021)

The deposits required to meet that demand exist in the ground. The geological evidence suggests that the Earth's crust contains far more undiscovered mineral resources than current known reserves. The problem is not geological scarcity — it is the time and cost required to find, delineate, and permit new deposits at the rate the transition demands.

Bringing a new mine from initial discovery to first production takes ten to fifteen years in favorable jurisdictions, and longer in areas with complex permitting environments. (World Bank, Minerals for Climate Action, 2021) To have new production online in the 2030s, the discoveries need to happen now. That requires not just more exploration capital, but faster and more accurate exploration methods.

Why exploration has not scaled

Global exploration spending has recovered from the downturn of the mid-2010s, but capital alone does not solve the problem. The bottleneck is not money — it is the capacity to interpret geoscientific data accurately enough to identify the small fraction of drill targets that will be economically significant.

The volume of geoscientific data available to explorers has increased enormously. Modern airborne geophysical surveys produce terabytes of data per project. Regional geochemical databases cover entire continents. Satellite multispectral and hyperspectral data provides global surface coverage at sub-meter resolution. The data exists to find many of the deposits that remain undiscovered. The constraint is interpretive capacity: the ability to integrate all of this data across an area of interest and identify the locations where it converges on mineralisation.

Manual geological interpretation does not scale to the volume of data that is now available. A geologist working on a district-scale project can meaningfully integrate a handful of datasets across a focused area of interest. Integrating twenty datasets across a regional-scale area is not practically achievable with traditional methods — and that is exactly the scale at which critical minerals exploration needs to operate to find the next generation of deposits.

The specific challenges of battery metals geology

Critical minerals present particular challenges that have historically made them harder to find than conventional gold or base metal deposits.

Lithium deposits in hard rock settings — spodumene pegmatites — are structurally controlled and often narrow, making them easy to miss with coarsely sampled surveys. Cobalt is almost always a byproduct mineral associated with nickel, copper, or gold systems, rarely the primary commodity being targeted. Rare earth elements concentrate in carbonatite and alkaline intrusive complexes that have distinctive geophysical signatures but require careful discrimination from barren intrusive rocks with similar signatures. Nickel sulphide deposits form at the margins of mafic-ultramafic intrusions and are detectable geophysically but require precise structural interpretation to locate drill-testable targets.

Each of these deposit types has a characteristic multi-dataset signature that can be learned by a machine learning model. The model does not need to understand the magmatic or hydrothermal processes that formed the deposit — it needs to learn which combination of spatial features co-occurs with known deposits of that type, and apply that learned pattern to unsampled terrain.

What AI brings to critical minerals specifically

For critical minerals exploration, the advantages of AI prospectivity mapping are more pronounced than for conventional commodities for three reasons.

First, many critical minerals jurisdictions have abundant public geoscientific data but limited exploration history. Government geological surveys in Canada, Australia, the United States, and parts of Africa have acquired extensive airborne geophysical and regional geochemical coverage over decades. This data exists in public repositories and can be used as model input even where client-specific data is sparse. The model has access to regional context that a geologist manually working a single project would not systematically draw on.

Second, critical minerals exploration often targets deposit types that are less well understood than conventional gold or copper porphyry systems. Deposit models for some battery metal deposit types are still actively being refined as new discoveries are made and studied. An AI model that learns directly from the data, rather than from a fixed deposit model template, is better positioned to capture the full range of geological signatures associated with an incompletely characterized deposit type.

Third, the economic case for AI-assisted targeting is clearest in jurisdictions where drilling costs are high. In remote northern Canada, Alaska, or Central Africa, a drill hole costs $150 to $400 per metre depending on depth, mobilisation costs, and terrain. A well-targeted drill program that reduces the number of holes needed to intersect mineralisation has a direct and quantifiable financial benefit that scales with the cost of the program.

The discovery rate problem

Mineral discovery rates have declined globally over the past three decades despite increasing exploration expenditure. (S&P Global Market Intelligence, 2023) More money is being spent to find fewer and lower-grade deposits. This is partly a geological phenomenon — the most accessible, near-surface deposits have been preferentially found first — and partly a methodological problem: exploration methods optimized for finding near-surface deposits have not kept pace with the need to find deeper, more subtle, and more complex systems.

AI prospectivity mapping addresses the methodological side of this problem. By integrating data across more layers and identifying subtler anomaly combinations than manual methods can detect, it has the potential to find deposits that would be missed by conventional targeting — deposits that are deeper, at lower grades, or in geological settings that do not match the simple deposit model templates in standard use.

This is not a claim that AI will find every undiscovered deposit. The model is limited by the data available and by the geological variability of the deposit types being targeted. But in the search space of unexplored terrain with existing public data coverage — which is vast — there are almost certainly deposits that current methods would not prioritise but that a data-driven approach would surface.

Where the technology stands today

AI prospectivity mapping is not a research concept. It is a production tool used by exploration companies, government geological surveys, and junior mining companies to rank drill targets and prioritise exploration budgets. The academic foundation — positive-unlabelled learning, spatial cross-validation, bootstrap uncertainty estimation — is well-established. The practical question is not whether the technology works but whether a specific project has the data and the geological context to use it effectively.

The technology is most effective when combined with qualified geoscientist review. The model surfaces statistical anomalies; the geologist evaluates whether those anomalies correspond to mineralisation or to geological noise. Exploration programs that integrate both — AI at the cross-referencing and prioritisation stage, geoscientists at the evaluation and target selection stage — consistently produce better drill programs than either approach alone.

The gap between the critical minerals the transition requires and the exploration methods available to find them is real and consequential. Closing it requires deploying the best available targeting tools across more projects, faster. AI prospectivity mapping is one of the most ready tools available — and the one with the clearest path from existing data to better drill decisions.