KoBold keeps discovery in house
KoBold Metals
KoBold is built to capture the full economics of a mineral discovery, not just charge a tool fee for improving someone else's odds. That means the software is only one layer of the machine. It feeds a workflow where KoBold picks targets, secures mineral rights, funds drilling, and in cases like Mingomba carries a discovery into mine construction, so the payoff can be a multibillion dollar asset instead of a six or seven figure software contract.
-
This changes who uses the product and what success looks like. The core users are KoBold's own geologists, geophysicists, data scientists, drill engineers, and lab teams, using ranked targets and uncertainty estimates to decide where to sample and drill next, then feeding new results back into the system.
-
The closest contrast is Terra AI. Terra sells decision software to miners through enterprise contracts, while KoBold keeps the stack in house and monetizes through ownership, earn ins, joint ventures, royalties, project sales, or future mine cash flow. Same technical problem, very different money model.
-
Owning the upside also makes the model far more capital intensive. KoBold was effectively pre revenue in 2025 despite a roughly $2.96B valuation and about $1.21B of total equity funding, because revenue starts when mines produce. Mingomba moved into shaft construction in Zambia in April 2026.
The next phase is a shift from proving the models can find ore to proving they can repeatedly turn discoveries into producing mines. If Mingomba and projects like Manono advance, KoBold looks less like a mining software company and more like a new kind of exploration major, one whose edge starts in data but compounds through asset ownership and development execution.