Toronto and Adelaide, Australia, November 10, 2026 / PRNewswire / – Obtaining In-Q-Tel and Teachers, supported by Venture Growth ( TVG ), AI Company Fleet Space Technologies (referred to as Fleet ) has today launched Tau 3.8 in the United States and Canada. This is a AI “world model” designed for mineral systems.
Fleet indicates that Tau is trained based on 10 million synthetic geological worlds, combining the company's 3D seismic intelligence with drilling, geological, geophysical, and mineralogical data. It can reconstruct hidden geological structures, prioritize resource exploration targets, and identify which new measurements can best reduce uncertainty, thereby helping mining companies to make more informed decisions regarding the location of their next drill holes.
The company stated that the infrastructure construction of AI in North America is facing a physical limitation: copper. Every data center, every power transmission line, and every upgrade to the power grid requires copper. S&P Global predicts that by 2040, the demand for copper by data centers will increase from 1.1 million metric tons in 2025 to over 2.5 million metric tons, with AI training accounting for 58% of this demand by 2030. Canada's "Critical Minerals Strategy" lists copper as one of six priority minerals, but according to data from the Canadian Ministry of Natural Resources, Canada's copper production decreased by nearly 25% between 2014 and 2023. Meanwhile, data from the U.S. Geological Survey shows that 57% of the copper consumed in the United States currently relies on imports, up from 45% a year ago.
Fleet believes that discussions regarding the restrictions imposed by AI mainly focus on chips and power supply, but the more fundamental constraints lie in the discovery of mineral deposits. The company states that finding and expanding mineral deposits that can supply the industry remains a slow, costly process fraught with uncertainties, and decisions involving billions of dollars are often based on drilling samples from only a very small portion of the underground area.
Fleet, co-founder and CEO, Flavia Tata Nardini, stated: "North America is building an AI economy, but the mineral foundation upon which it relies has not yet been found. Ultra-large-scale cloud service providers can order more GPU, and they can also sign new power agreements, but they cannot order a copper deposit out of thin air. The gap between what the economy needs and what the industry can produce is widening, and the discovery of minerals itself has become a constraint on growth. For years, various companies have been collecting drilling and survey data. Tau integrates this data into a model, allowing teams to understand mineral bodies earlier and making every foot of drilling more valuable."
Flavia also indicates: "Toronto is one of the important birthplaces of modern AI, and it is also a global mining financing center. The exchange between these two worlds is far from sufficient. Tau is the result of their beginning to engage in dialogue. The capital provided for exploration can obtain clearer answers regarding where to drill before the funds are actually invested underground. Faster and more capital-efficient exploration is the way to ensure that supply keeps up with the economic volume created by AI."
Reduce ineffective drilling
Fleet indicates that drilling is typically the largest cost in mineral exploration, with the cost of a deep diamond drill hole potentially reaching hundreds of thousands of dollars. By more accurately utilizing existing drilling data to predict copper grades, Tau can show which areas are likely to continue having high grades and which are not, thereby reducing the number of drill holes that hit depleted strata. As a result, companies can prioritize advancing the most valuable targets and define and expand resource volumes with less drilling, in less time, and at lower costs.
In the Valeriano copper-gold project located in Chile, Tau combines the environmental noise tomography (ANT) and magnetotelluric (MT) measurements from Fleet with Comet and machine learning. Fleet states that when testing on reserved boreholes not provided to the model, the model's prediction error for copper grade was 38% lower than that of traditional geological statistical estimation methods (Kriging method) based on the same drilling data; on the deepest extended boreholes, the error was as much as 67% lower.
Fleet indicates that its technology has been used in multiple on-site projects, covering areas from Pakistan to the Athabasca Basin in Canada. Related projects include Barrick's Reko Diq copper-gold project in Pakistan, Lundin Gold's Fruta gold mine in Ecuador, del, Norte gold mines, Fermi Exploration's Perch River project in the Athabasca Basin, Southern Cross Gold's Sunday Creek gold-antimony project in Victoria, Australia, and Alkane Resources's Boda-Kaiser copper-gold project in New South Wales. Fleet also includes a wider range of clients such as Gold Fields, Rio Tinto, Ma, aden, Standard Uranium, IsoEnergy, Q2 Metals, as well as the French Geological Survey's BRGM.
"World Model" Applied to the Earth
Fleet indicates that "world models," which are systems that learn how the physical world operates rather than predicting the next word, have become one of the most focal and cutting-edge directions in the AI field. Previously, most of the attention was focused on robots, videos, and autonomous vehicles. Fleet applies this approach to an environment that is even more difficult to observe than any road or factory workshop: the Earth's underground.
Fleet Chief Science Officer Gerrit Olivier Dr. stated: "Simply putting a shell around a large language model does not make one a world-class geologist. The mineral system is three-dimensional and buried underground, and our understanding of it relies on sparse observations that are also costly to obtain. Fleet's approach is to build 10 million simulated geological environments, simulate what earthquake sensors and boreholes would detect in each environment, and train Tau to infer the geological conditions in between based on these incomplete measurements."
He also stated, "World models learn about the structure, geometric forms, and composition of the physical world itself. It seems that we have now crossed a critical threshold: a model trained in a synthetic world can use real-world data to make predictions and guide exploration decisions. I believe this marks a 'AlphaGo moment' in the field of geology."
Why are earthquake data important?
Fleet indicates that AI can generate millions of plausible versions regarding the underground conditions, but many of these may coincide with similarly sparse data. Seismic data, however, provides the physical evidence needed to distinguish between these versions, enabling the determination of the depth, geometric shape, direction, and continuity of the geological structure over large areas between boreholes.
Olivier indicates that seismic data narrows down the range of geological possibilities between boreholes, allowing the team to answer questions that drive capital decisions: where do the mineral deposits extend, how deep do they extend downward, and where is drilling another foot most likely to add value?
According to Fleet, this integration capability is where its strength lies: Tau compares competing predictions with physically measurable underground properties, rather than relying solely on geological plausibility. As new data becomes available, the model updates its understanding and suggests the next set of measurements to be taken. Fleet refers to this method as “Agentic Geoscience (Proxy Earth Science)”.
Flavia indicates that: “Tau is not about excluding geologists from the decision-making process. It provides them with a quantified view of uncertainty before the drill bit gives an answer.”
Developed relying on the North American AI ecosystem
Fleet indicates that Tau was developed jointly by experts in seismology, geology, geophysics, applied mathematics, and machine learning, including earth scientists with a combined experience of over 200 years across more than 400 projects, as well as leaders in machine learning from the California Institute of Technology and the University of Oxford. Fleet is currently collaborating with NVIDIA to build the accelerated computing infrastructure necessary for training the next generation of high-resolution underground models.
Fleet is also an industrial partner of the Mineral-X project at Stanford University. This project connects researchers with the industry, focusing on AI, earth sciences, and decision-making under conditions of uncertainty. Tata Nardini serves as a member of the Mineral-X advisory committee.
About Fleet
Fleet Space Technologies is a company in the space and AI sector, dedicated to developing technologies for finding the minerals that the global economy relies on. Its flagship platform, ExoSphere, integrates Fleet's satellite network, rapid 3D seismic data collection, and AI into one system, enabling nearly real-time imaging of mineral systems while minimizing environmental impact. As the world model for mineral systems from Fleet, Tau expands this platform to "proxy earth sciences." Fleet collaborates with leading exploration and mining companies, including Rio Tinto and Barrick, with investors such as In-Q-Tel, Blackbird, and TVG. Founded in Australia, the company operates in the United States, Canada, Chile, Australia, and Africa.
