Earth AI

At roughly $20M raised, Earth AI is minnow-scale against a task — proving up economic mineral deposits — that consumes hundreds of millions and a decade before anything is mined. The business model depends on a handoff it does not control: it must convince mining majors to buy its rights at prices that reward the discovery, and majors are notoriously slow, skeptical of junior-explorer claims, and inclined to wait until a deposit is fully de-risked (by someone else's capital) before paying up. Owning drill rigs cuts both ways — it is capital-intensive, operationally demanding, and exposes a software company to the margins, safety liabilities and permitting delays of a field-services business. AI targeting also remains scientifically unproven at scale; the industry is littered with geostatistical models that back-tested beautifully and found nothing new, and a run of dry holes would evaporate the credibility that the whole pitch rests on. Australia's exploration ground is picked-over and permitting is slow, so "untapped" is doing heavy lifting. Meanwhile KoBold Metals, capitalized at fifty times the level and backed by Gates and Bezos, is chasing the same AI-discovery prize with vastly more data, drilling and balance sheet. Earth AI's window to prove the loop actually closes — targeting to drilling to sold rights to royalty — is narrow, and it is racing far better-funded competitors to demonstrate that AI geology is real rather than a well-marketed correlation engine.