The mineral exploration business is a numbers game, and for decades the numbers have been terrible. The industry standard is a 0.5% discovery success rate, a figure that makes venture capital look like a safe bet. For Roman Teslyuk, a geoscientist who founded Earth AI in 2017, the problem wasn't just the geology. It was the economics of digging random holes in the ground, a process he describes as wasteful and expensive. His company's pitch is a simple unit conversion: turn that 0.5% into something a rational investor could love.
Earth AI is a vertically integrated explorer. It uses a proprietary AI platform to analyze public geological data and identify promising sites for critical minerals like cobalt, tungsten, and lithium. Then, instead of selling the data, it sends its own drilling rigs to prove the deposits. The final product is a validated mineral prospect, which it aims to sell to large mining companies like BHP or Rio Tinto [Latitude Media, Jan 2025]. This end-to-end control, from algorithm to drill core, is the company's core wedge. It claims this approach has yielded a discovery success rate of 75% while slashing exploration costs by up to 80% [LinkedIn, Ben Werdegar, Unknown].
The bet on vertical integration
Traditional mineral exploration is a fragmented, high-risk supply chain. One firm does the geophysical surveys, another handles the permitting, a third does the drilling, and the data often sits in silos. Earth AI's bet is that owning the entire stack,from the AI targeting software to the physical drilling,creates a feedback loop that dramatically improves efficiency. The software learns from the drilling results, and the drilling is guided by ever-sharper software predictions. The company says its Mineral Targeting Platform has already identified six new mineralized prospects in Australia [miningstockeducation.com, 2025]. By developing these prospects itself, Earth AI captures the value uplift between raw data and a bankable resource, aiming to sell a de-risked asset.
Why the check from Tamarack Global cleared
In January 2025, Earth AI closed a $20 million Series B round led by Tamarack Global and Cantos Ventures, bringing its total disclosed funding to approximately $41 million [PR Newswire, Jan 2025]. The round was oversubscribed, a signal that investors are buying the integrated model. The tailwinds are clear: the energy transition is creating unprecedented demand for critical minerals, and existing discovery methods are too slow and inefficient. For a fund like Tamarack, which focuses on resource and technology investments, Earth AI sits at a rare intersection. It's a tech play that produces a physical, scarce asset. The company reported $7 million in annual revenue for 2025 [RocketReach, 2026], suggesting it has moved beyond pure R&D and into commercial transactions.
Seed (2021) | 5.8 | M USD
Seed+ (2021) | 5.5 | M USD
Series B (2025) | 20.0 | M USD
The team and the traction
Founder and CEO Roman Teslyuk is a Y Combinator alumnus and was working toward a doctorate in mineral exploration at the University of Sydney when he started the company [TechCrunch, Mar 2025]. His background as a geoscientist with eight years of field experience is central to the company's credibility; this isn't a pure software team applying AI to a domain they don't understand. The team includes other geologists and operates primarily in New South Wales, Australia, with a corporate presence in San Mateo, California [CB Insights, 2026]. While the public employee count is low (Crunchbase lists 3 profiles, Y Combinator suggests 12), the operational model is asset-light on the personnel side, relying on proprietary technology and contracted drilling.
The company's disclosed metrics paint a picture of early, audacious traction:
- Discovery rate. Claims a 75% success rate versus a 0.5% industry average [LinkedIn, Ben Werdegar, Unknown].
- Cost reduction. Says it can explore at 20% of standard costs [LinkedIn, Ben Werdegar, Unknown].
- Revenue generation. Reported $7 million in annual revenue for 2025 [ZoomInfo.com, Unknown].
- Asset pipeline. Has confirmed six new mineral prospects containing tungsten, cobalt, and gold [theaiinsider.tech, 2025].
Where the model faces pressure
The most immediate question is about those headline metrics. A 75% discovery rate is so far outside industry norms that it demands extraordinary evidence. While the company cites its own drilling results, independent verification from a major mining house on a commercial sale would be the ultimate proof point. The go-to-market motion is also unproven at scale. Selling a mineral deposit to a mining major is a complex, relationship-driven enterprise sale that can take years. Earth AI's public record does not yet name a specific buyer like BHP or Rio Tinto as a customer, only as the target archetype [Latitude Media, Jan 2025]. The company's answer is likely that its validated prospects, with drill cores in hand, speak for themselves and will attract bids in a supply-constrained market.
Another risk is operational. Managing a drilling campaign, even with proprietary tech, involves permitting, environmental assessments, and local logistics. Scaling this in multiple jurisdictions adds complexity that pure software companies avoid. The capital intensity of drilling is also why the $20 million Series B was necessary; you can't prove a deposit with just server credits.
The next twelve months
The coming year will be about converting prospects into partnerships. The key milestone to watch is the announcement of a first major offtake agreement or joint venture with a named mining company. That would validate not just the geology but the commercial model. The $20 million in new capital should fund several more drilling campaigns to expand the asset portfolio. Geographically, the company may look beyond its Australian base to other mineral-rich, stable jurisdictions like Canada or the United States.
On a back-of-the-envelope basis, if the traditional exploration cost for a prospect is $5 million with a 0.5% chance of success, the expected cost per discovery is $1 billion. Earth AI claims it can explore at 20% of the cost ($1 million) with a 75% success rate, making the expected cost per discovery roughly $1.33 million. That's a three-order-of-magnitude improvement in unit economics, if the claims hold. For Earth AI to prove its model definitively, it must beat not just the statistical averages, but the incumbent service model of consultancies like SRK Consulting or Snowden Group. It's not selling advice; it's selling the mine itself.
Sources
- [TechCrunch, Mar 2025] Earth AI's play in the hunt for critical minerals | https://techcrunch.com/2025/03/25/earth-ais-algorithms-found-critical-minerals-in-places-everyone-else-ignored/
- [PR Newswire, Jan 2025] Earth AI Closes Oversubscribed Round; Raising $20M for AI Driven Mineral Exploration | https://www.prnewswire.com/news-releases/earth-ai-closes-oversubscribed-round-raising-20m-for-ai-driven-mineral-exploration-302360289.html
- [LinkedIn, Ben Werdegar, Unknown] Ben Werdegar - Sparkwave Capital | https://www.linkedin.com/in/ben-werdegar/
- [Latitude Media, Jan 2025] Earth AI's play in the hunt for critical minerals | https://www.latitudemedia.com/news/earth-ai-series-b
- [CB Insights, 2026] EARTH AI - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/earth-ai
- [RocketReach, 2026] Earth AI Company Profile | https://rocketreach.co/earth-ai-profile_b5c4c4f5f5e6af4d
- [miningstockeducation.com, 2025] Earth AI confirms six new mineral prospects using AI | https://miningstockeducation.com/earth-ai-confirms-six-new-mineral-prospects-using-ai/
- [theaiinsider.tech, 2025] Earth AI Discovers Six New Mineralized Prospects in Australia Using AI-Powered Exploration | https://theaiinsider.tech/2025/01/earth-ai-discovers-six-new-mineralized-prospects-in-australia-using-ai-powered-exploration/
- [ZoomInfo.com, Unknown] Earth AI Company Revenue | https://www.zoominfo.com/c/earth-ai/526842789
- [Y Combinator, Unknown] EARTH AI: Predictive Explorer and Driller for Critical Metal Deposits | https://www.ycombinator.com/companies/earth-ai