Mario Stifano runs Galantas Gold, a company trading on Toronto's TSX Venture Exchange and London's AIM, with a track record spanning more than 25 years in mining exploration, development and operations. The firm is working to bring the Andacollo gold project in Chile back into production and is developing the Indiana gold-copper project. We spoke with him about how the global push to build AI infrastructure is reshaping demand for precious and industrial metals.

Gold's role in technology

Most people associate gold with vaults or jewellery, overlooking its critical function inside everyday electronics. Engineers have depended on gold for decades because of its electrical conductivity, resistance to corrosion, and ability to form extremely thin wires and coatings. When a connection must perform reliably every single time, these characteristics become essential.

As computing demands intensify, this role should persist. Yet rising AI investment doesn't automatically translate into proportional gold consumption. Manufacturers continuously seek ways to reduce their use of expensive materials. The World Gold Council documented roughly 323 tonnes of technology-driven demand in 2025, essentially unchanged from 2024. Although AI applications boosted demand, thinner wires, more efficient designs and declining consumer electronics consumption offset those gains. Gold can grow more critical in specific premium components without triggering a major surge in overall tonnage.

Data centres and metals demand

The infrastructure underpinning AI data centres extends far beyond the servers themselves. Reliable electricity must flow continuously, requiring substations, cables, transformers and generating capacity. The International Energy Agency's latest research shows data-centre electricity consumption jumped 17% in 2025, with projections to roughly double by 2030, accelerating further for AI-focused facilities. This scale of expansion demands substantial physical infrastructure.

Copper permeates this entire electrical ecosystem. AI is intensifying copper demand precisely when utilities, transportation networks and industrial sectors are simultaneously investing in electrification. Mining must plan for all these competing demands together. Gold operates differently—it appears selectively in electronics where reliability justifies its cost, while copper's reach spans the physical infrastructure backbone. For gold, technology matters, yet it remains one component of a market primarily shaped by investment, central banks and jewellery demand.

Supply chain bottlenecks

The technology and mining sectors operate on fundamentally misaligned timescales. Committing billions to AI infrastructure doesn't guarantee materials will be available when construction teams need them. S&P Global's analysis revealed that mines beginning production between 2020 and 2024 required nearly 18 years on average from initial discovery. Before any construction starts, companies must drill deposits, validate economics, complete engineering, secure approvals and arrange financing.

Copper faces pressure extending well beyond AI. Data centres compete for supply alongside electricity networks, electric vehicles and other industries. The IEA's 2026 outlook projects that under current policies, existing and announced projects could leave primary copper supply roughly 25% short of requirements by 2035. Bridging this gap requires more than new discoveries—it demands financing for projects that can actually deliver, expanding and restarting viable operations, and extracting more metal from ore already being processed. Without these steps, technology companies will encounter higher costs, extended equipment lead times and postponed projects.

This reality shapes our strategy. Our Andacollo gold project in Chile is a former producing mine. We still face refurbishment, technical and financing work, but we're not beginning from zero.

AI in gold exploration

A compelling opportunity lies in extracting more value from work already completed. Established mining districts contain decades of drilling, surveys and geological records. Different companies gathered this information at different times, often pursuing different objectives. Consolidating it and posing new questions that were previously difficult to investigate creates genuine value.

Canada is pursuing this exact approach through a federal and territorial pilot in the Northwest Territories using AI to scan and analyse drill core stored in government facilities. Major operators are following suit. Agnico Eagle partners with GeologicAI, merging AI with drill-core scanning to strengthen resource models. At Galantas, our Chile-based team spends considerable time examining drill core and surface geology, sharing portions of that work via our Field Notes video series. Technology can assist, yet experienced geologists must still interpret the core.

Drilling remains mandatory. A model can indicate where to look; it cannot prove an economic deposit exists. The practical test is straightforward: does it enable the team to make a better decision about the next drill hole? Since exploration capital is finite, superior targeting significantly extends how far that capital reaches.

New uses for gold in AI hardware

Specialised applications don't diminish importance. A component can hold minimal gold yet still depend on it for reliable performance. Several emerging applications warrant attention. Co-packaged optics, which positions optical connections adjacent to processors, enables AI systems to transfer data faster while consuming less power. Some optical components and high-frequency electrodes in these designs incorporate gold.

The World Gold Council identifies this as a potentially modest but meaningful new demand source if widely adopted across AI data centres, though it remains in early stages. From a mining standpoint, three questions matter: Is the technology achieving commercial adoption? How much material does it require? Can something cheaper perform the same function? Pressure to use less gold will persist due to its cost. Yet where it delivers clear reliability advantages, it justifies its place—pointing toward a specialised role with growth potential rather than a scenario where AI transforms the entire gold market.

Grids, copper and realistic timetables

Technology companies likely grasp their dependence on physical resources. The genuine question concerns whether all participants operate on realistic timelines. Servers can be ordered quickly. Grid connections may require new lines, approvals and equipment with extended delivery windows. The IEA estimates that without addressing these constraints, roughly 20% of planned data-centre projects worldwide face delay risk—primarily a grid problem rather than copper scarcity, yet it underscores how physical the AI expansion truly is.

The supply chain runs deeper than commonly understood. The IEA's 2026 outlook warned that near-term copper supply has deteriorated considerably, partly due to tight sulphuric acid availability. Many copper operations require this reagent to process ore, and Chile, where we operate, ranks among major producers using this process. Most technology professionals have never considered sulphuric acid, yet it influences how much copper reaches markets. When supply lags demand, companies postpone projects, pay premiums or relocate. Long-term purchasing agreements and early supplier conversations help suppliers expand with confidence. In mining, a strong asset represents only part of the equation; power, infrastructure and financing are equally vital. Digital infrastructure faces identical requirements.

Machine learning and mining economics

The principal opportunity involves extracting greater value from money already spent. Drilling carries high costs. Equipment downtime is expensive. Losing recoverable gold during processing is expensive. Technology can address these specific challenges. Better analysis of historical drilling and geological data helps teams decide where to drill next. It doesn't guarantee discovery, but superior targeting enables more efficient use of exploration budgets.

Recovery and maintenance

In production, recovery—the percentage of metal in ore that the plant extracts—matters significantly. In a simplified scenario, if processed material contains 100,000 ounces of gold, improving recovery from 90% to 92% yields an additional 2,000 ounces from identical feed. The cost of achieving that improvement must be weighed, but no additional tonnes require mining to access those ounces. BHP and Microsoft have detailed using real-time plant data and AI recommendations at Escondida to help operators adjust processing conditions targeting improved copper recovery.

Predictive maintenance addresses another cost driver. Detecting emerging equipment problems enables repairs during scheduled shutdowns rather than unexpected failures disrupting production. For smaller companies, access to specialist software and service providers can make advanced analytical tools affordable. Yet tools still require reliable data and experienced personnel. Large operators retain advantages in both dimensions. I recommend starting with a measurable problem, testing the technology and comparing benefits against full costs. Smaller companies needn't match major miners' technology spending—they must deploy their own budgets effectively.

Recycling and new supply

Recovering more metal from products already manufactured makes sense. The critical question concerns how much can be collected and recovered economically, and when it becomes available. The UN's Global E-waste Monitor found that only 22.3% of e-waste generated in 2022 was formally collected and recycled. Measured by weight rather than recovery rates for specific metals, this reveals substantial improvement potential. AI-assisted sorting and enhanced processing can help, yet they require collection networks, appropriate facilities and business models covering costs. Timing also matters—recycling depends on products reaching end-of-life, and copper installed in grids today may operate for decades.

I view recycling and mining as complementary. Increased recycling strengthens supply security and reduces how many new mines are necessary. It doesn't eliminate the need to invest in mining.

What matters most by 2030

For the decade through 2030, delivering supply ranks first. I also believe the market underweights this priority. A discovery made today will likely not produce metal before 2030. This makes projects already permitted, financed and under construction far more relevant this decade, as does what existing operations can contribute through expansions, improved recovery and thoughtful restarts. AI will support these decisions. Much of its near-term value should come from understanding known assets more thoroughly and helping operations run more consistently. Yet the asset must still function. It requires appropriate geology and metallurgy, a credible development plan, financing and operational responsibility. Better software doesn't alter that fundamental requirement.

I'd also distinguish gold from copper. Data centres form part of copper's broader infrastructure narrative. For gold, technology plays a role, but monetary conditions, investment demand and central-bank purchasing will prove more consequential. The question I continually return to is straightforward: which projects can actually deliver, at what cost, and when?

Source: The Next Web