As demand for copper rises alongside AI data centers, power infrastructure and electrification, mining companies face a difficult challenge: finding economically viable deposits and expanding existing resources without spending excessively on drilling. Fleet Space Technologies is targeting that problem with Tau 3.8, an AI world model for mineral systems launched in the United States and Canada. Trained on 10 million synthetic geological environments, the platform combines seismic intelligence with drilling and other geological measurements to reconstruct underground structures, rank exploration targets and recommend where additional data collection could reduce uncertainty.
Fleet Applies AI World Models to the Search for Copper
The launch comes as North America faces growing pressure to secure supplies of minerals needed for its expanding technology and energy infrastructure. Copper is essential to data centers, electricity transmission, power generation and industrial equipment. Yet discovering and developing new deposits can take years, with exploration decisions often relying on incomplete information from widely spaced drillholes.
Fleet argues that the next constraint on AI infrastructure may not be limited to chips, electricity or data center construction. It could also involve the minerals needed to build and connect that infrastructure.
The scale of the challenge is significant. S&P Global forecasts that copper demand associated with data centers will increase from 1.1 million metric tons in 2025 to 2.5 million metric tons by 2040. The research also projects that AI training data centers will account for 58% of data-center-related copper demand by 2030.
Supply concerns are adding to the pressure. Natural Resources Canada reports that Canadian mine production of copper declined 26.2% between 2015 and 2024. Meanwhile, the U.S. Geological Survey estimates that net import reliance for U.S. copper consumption reached 57% in 2025, up from 45% in 2024.
Against this backdrop, Fleet is positioning Tau as an AI-assisted exploration system that could help mining companies extract more value from the geological information they already possess.
A world model designed for geology
Unlike conventional generative AI systems that primarily process language or images, world models aim to learn the structure and behaviour of physical environments. Fleet has applied this concept to the Earth’s subsurface, where geological formations are hidden and direct observations are expensive.
Tau 3.8 was trained on 10 million simulated geological environments. According to Fleet, the system models what seismic sensors and drillholes would detect across these synthetic worlds, teaching the model to infer geological structures from incomplete measurements.
The platform combines Fleet’s three-dimensional seismic intelligence with drilling records, geology, geophysics and mineralogy data. It uses those inputs to reconstruct subsurface geology, identify potential extensions of mineral deposits and rank targets for further exploration.
An important feature is its approach to uncertainty. Rather than only predicting what may lie underground, Tau identifies which additional measurement could most reduce uncertainty. This gives exploration teams another way to decide whether the next investment should go toward drilling, seismic surveys or other data collection.
Fleet calls this approach Agentic Geoscience. The intention is to make exploration an iterative process in which models update their geological interpretations as new measurements arrive.
Valeriano project provides a real-world test
One of the announcement’s most significant claims comes from ATEX Resources’ Valeriano copper-gold project in Chile.
Fleet says Tau combined ambient noise tomography (ANT) and magnetotelluric (MT) survey data with drilling information, its Comet platform and machine learning to predict copper grades. The model was tested against drillholes that had been withheld from the prediction process.
According to Fleet, Tau produced copper-grade predictions with 38% less error than a conventional geostatistical estimate using kriging and the same drilling data. On the deepest step-out holes, the reported error reduction reached as much as 67%.
The result is relevant because it offers a comparison against a conventional estimation method rather than relying solely on an internal AI performance score. However, the findings are company-reported and relate to a particular project and dataset. They should not be interpreted as proof that Tau will deliver equivalent improvements across different geological settings.
The distinction matters in mineral exploration, where deposits can vary substantially in geometry, mineral composition, structural complexity and data coverage.
Why seismic data matters
Drilling provides direct information about the rocks and mineral grades encountered at particular locations. But even a substantial drilling campaign samples only a small fraction of a deposit.
Seismic measurements can help fill some of the gaps by providing information about underground structures between drillholes. Fleet’s approach uses seismic evidence to constrain competing geological interpretations, helping the model estimate the depth, orientation and continuity of mineralised structures.
This is a key part of the company’s differentiation. Fleet is not presenting Tau as a general-purpose language model with a geological interface. Instead, it combines machine learning with physical measurements of the subsurface.
The approach could help exploration teams identify where additional drilling is most valuable, reduce unnecessary holes and improve resource expansion decisions. It does not remove the need for geological expertise or physical verification: predictions still need to be tested against new measurements and field results.
Building an AI layer for mining operations
Fleet says its technology is already being used across projects in several countries. Named examples include Barrick’s Reko Diq copper-gold project in Pakistan, Lundin Gold’s Fruta del Norte operation in Ecuador, and exploration projects involving Fermi Exploration, Southern Cross Gold and Alkane Resources.
The company also identifies Gold Fields, Rio Tinto, Ma’aden, Standard Uranium, IsoEnergy, Q2 Metals and France’s geological survey BRGM among its wider customer base.
Fleet is working with NVIDIA on the accelerated computing infrastructure required to train future high-resolution subsurface models. It is also an industrial affiliate of Stanford University’s Mineral-X program, which brings together researchers and industry participants working on AI, geoscience and decision-making under uncertainty.
The broader opportunity extends beyond copper. The same combination of seismic intelligence, geological data and machine learning could support exploration and resource expansion across other mineral systems, although performance will depend on the available data and the characteristics of each deposit.
For mining companies, the commercial case will ultimately rest on whether these systems can improve exploration decisions, reduce avoidable drilling expenditure and help teams identify viable resources earlier.
Tau 3.8 points to a different application of world models: using AI not simply to generate content or automate digital tasks, but to interpret complex physical environments. As North America’s demand for critical minerals grows, the ability to extract more insight from existing geological data could become an increasingly important part of the mining technology stack.
Market Landscape
AI in mining is expanding beyond equipment monitoring, predictive maintenance and autonomous vehicles into geological interpretation and exploration planning.
Three market developments are particularly relevant:
- AI for subsurface intelligence: Machine learning is being applied to seismic, drilling and geophysical datasets to improve geological models and identify potential mineral targets.
- Critical mineral supply pressure: Copper, lithium, nickel and other minerals are increasingly important to data centers, electrification, energy storage and industrial infrastructure.
- Decision-making under uncertainty: Mining technology providers are combining AI predictions with physical measurements to help teams decide where additional exploration capital is most valuable.
Fleet competes in an emerging space that includes geoscience software providers, geological modelling platforms, mineral exploration technology companies and AI-enabled mining services. Its approach combines proprietary seismic capabilities with synthetic-world training and iterative data collection.
The key competitive test will be whether Tau can consistently outperform established geological estimation methods across multiple projects, commodities and geological environments—not just in selected case studies.
Top Insights
- Fleet launched Tau 3.8 in the United States and Canada, applying AI world-model techniques to mineral exploration and underground geological reconstruction.
- The system was trained on 10 million synthetic geological environments and combines seismic, drilling, geology, geophysics and mineralogy data.
- At Chile’s Valeriano project, Fleet reported 38% less copper-grade prediction error than a conventional kriging estimate using the same drilling data.
- Tau identifies additional measurements that could reduce geological uncertainty, helping exploration teams prioritise drilling and other data collection.
- Rising copper demand from AI infrastructure and electrification creates a broader commercial opportunity for AI-assisted mineral exploration.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
