How KoBold Metals transformed mineral exploration
Leveraging Coiled to accelerate the discovery of critical battery metals
Company Background: KoBold's Unique Approach#
KoBold Metals isn't a traditional mining company. Instead of relying solely on geological intuition and luck, they're revolutionizing mineral exploration through data science and artificial intelligence.
We are built around the idea of generating as many hypotheses as we can and falsifying them as quickly as we can. And to do that, we really need large-scale geospatial computational capabilities.
Jerry Vinokurov
Senior Software Engineer, KoBold Metals
Focused on four critical minerals essential for modern batteries—nickel, copper, cobalt, and lithium—KoBold analyzes vast amounts of geospatial data to determine which areas are most likely to contain valuable mineral deposits before committing significant exploration resources.
This computational approach to mining transforms what was traditionally an intuition-driven process into a repeatable, data-driven science. The goal: identify promising exploration targets faster and with greater precision than conventional methods.

The Data Challenge: From Surface Samples to Continental Scale#
KoBold's analysis starts with two fundamental data types:
- Point-sampled surface data - Geochemical samples revealing mineral composition at specific locations
- Continuous field data - Magnetic and gravitational fields measured across entire regions
As exploration progresses, the team incorporates increasingly complex data sources:
- Electromagnetic responses from subsurface structures
- Hyperspectral imagery with hundreds of spectral bands
- Drilling results when a project advances to direct sampling
Depending on the size of the area that we're looking at, data sets can very quickly balloon. As the area of interest expands to subcontinental or even continental size, the resources required scale quadratically. Not something you can really do very well on a single machine.
The computational demands become particularly challenging when processing magnetic field data, which requires taking derivatives of potential fields in various directions. These calculations involve Fourier transforms of entire regions—a process that quickly exceeds the capabilities of even the most powerful workstations.
In order to do that efficiently, you have to take the Fourier transform of the magnetic field. And that becomes really prohibitively expensive on a single machine real fast.
As the team sought to analyze more extensive areas with finer resolution, they needed a solution that could scale with their ambitions.
Technical Implementation: Scientific Python at Scale#
KoBold built their technical stack on Python's scientific computing ecosystem, leveraging a combination of standard libraries and custom code:
- NumPy and SciPy power their core numerical computations
- X-Array manages multidimensional geospatial data
- Simpeg simulates electromagnetic responses
- Dask coordinates parallel computing tasks
- Custom algorithms handle specialized geostatistical analysis
When processing demands outgrew single-machine capabilities, KoBold evaluated several cloud solutions. The decisive factor in choosing Coiled was its seamless integration with their existing workflow:
For us, the ability to just point Coiled at a Docker image and have it just work is amazing. The software that we write, the final product is a Docker image that is used by any of our developers. To have that and to be able to mount that and just have it run without doing anything else is just so convenient.
This integration eliminated deployment friction and allowed KoBold's scientists to focus on what they do best—analyzing geospatial data to find promising mineral deposits—rather than managing cloud infrastructure.
Results: Processing the Previously Impossible#
With Coiled, KoBold transformed their analytical capabilities, tackling problems that were previously out of reach:
We scaled it up from rather small rasters to things that are country-sized or larger. There's practically no real limit to what we can process now. We can process magnetic surveys over pretty much any area of interest that we care to.
The success of their initial magnetic field processing pipeline catalyzed adoption across the organization. Today, the team uses Coiled for diverse applications:
- Remote sensing analysis across multiple spectral regimes
- Hyperspectral processing with hundreds of bands from aircraft surveys
- Stochastic inversions that generate subsurface models from surface data
From a small group of early adopters, Coiled usage expanded to several team members across departments. The platform's observability features proved particularly valuable for optimization:
I usually have the Dask dashboard running on one screen and then the Coiled dashboard on the other screen. I'm comparing them and trying to understand, okay, here is where your memory usage blew up and here is probably why.
For computationally intensive tasks like hyperspectral image processing, KoBold now leverages clusters that would be impractical to maintain in-house.
Business Impact: Data-Driven Decisions#
The computational capabilities enabled by Coiled transformed KoBold's ability to evaluate potential mining sites quickly and accurately:
When we process this data, you have to take derivatives of the potential field in various directions. And that becomes really prohibitively expensive on a single machine real fast. With Coiled, we can handle areas that might expand to subcontinental or even continental size.
This acceleration directly impacts the company's core business strategy: generating and falsifying hypotheses about mineral deposits as efficiently as possible. By rapidly evaluating the potential of different areas, KoBold makes smarter decisions about resource allocation—avoiding wasteful expenditure on less promising sites and focusing efforts where data suggests the highest probability of success.
The ability to process continental-scale datasets gives KoBold a competitive edge in identifying promising regions that traditional exploration methods might miss. As the company expands operations, the computational foundation provided by Coiled allows them to evaluate more sites with higher resolution and greater accuracy, without proportional infrastructure costs.
Looking Forward: Data-Driven Mining Exploration#
As demand for battery metals grows with the global transition to electric vehicles and renewable energy, efficiently identifying new mineral deposits becomes increasingly critical. KoBold's computational approach, powered by Coiled, positions them at the forefront of this essential industry.
By turning mineral exploration into a repeatable science rather than a hit-or-miss endeavor, KoBold isn't just finding minerals more efficiently—they're redefining what's possible in an industry that has remained largely unchanged for generations.
Coiled has made it possible for us to process huge volumes of data, and that's been really key to getting fast turnarounds on our analysis. The faster we can falsify our hypotheses, the sooner we can decide whether or not to commit resources to exploration.
Jerry Vinokurov
Senior Software Engineer, KoBold Metals
For a modern mining company looking to discover the materials that will power our future, the combination of advanced geoscience, comprehensive data, and scalable computing provides a foundation for success in an increasingly competitive global market.