Thesis: Terra AI is building the AI-native subsurface modeling stack for critical mineral discovery, enhanced geothermal, and carbon storage — three verticals fed by the same probabilistic 3D Earth model. The wedge is technical: two Stanford Intelligent Systems Lab PhDs fusing geophysics, geochemistry, and drilling data into uncertainty-quantified models that pure-ML shops (KoBold, Earth AI) can approximate but not derive from first principles, and one codebase serves all three end markets.
Validation runs deep. Signed a 2-year $550K milestone contract with OMV (note 2024-07-25), a California Geological Survey contract, and a copper miner engagement with a 1% royalty kicker. Series A closed September 2025 at a $50M post led by Khosla, converting our SAFE at $1.15969 into 172,459 shares (per Airtable), and BHP Ventures signed a $4M follow-on convertible termsheet at a $100M cap on 2/25/26. StoryHouse Fund I wrote a $200K check on the $3.3M seed at a $15M post — on Series A share pricing that is a ~2.18× paper markup ($200K → ~$435K); the BHP CN cap of $100M sets a subsequent 6.7× company-valuation marker if that CN converts.
The addressable prize is bigger than the "AI exploration tech" line item alone: Terra's platform sits at the intersection of critical mineral discovery, enhanced geothermal, and CCS — each a multi-billion-dollar downstream market being pulled forward by the IEA-flagged critical mineral undersupply and the global energy-transition CapEx wave.
| Player | Positioning | Funding | Edge |
|---|---|---|---|
| Terra AI | Probabilistic 3D subsurface model fusing geophysics + geochem + drilling; multi-vertical (minerals, geothermal, CCS) | ~$22.4M raised through Series A | — |
| KoBold Metals | Full-stack AI prospecting; TerraShed + Machine Prospector; owns/operates discoveries | $537M+ from Gates, Bezos, a16z, BHP | Capital moat, portfolio of discovered assets |
| Earth AI | Proprietary drilling hardware + predictive analytics; Australia-focused; claims ~75% hit rate | Series A-stage (PitchBook) | Vertically integrated to drill |
| VerAI | Concealed deposits under covered terrains; asset-portfolio model | Undisclosed early-stage | Covered-terrain specialization |
Moat: Stanford Intelligent Systems Lab peer-reviewed origination plus a probabilistic-uncertainty-first design lets Terra cross from minerals into reservoirs (OMV, geothermal, CCS) on the same codebase — something the pure-mining AI shops cannot follow without a full rebuild.
Most plausible outcome is strategic acquisition by one of the mining supermajors already inside the tent — BHP is now on the cap table via BHP Ventures and Rio Tinto is a live prospect (note 2024-08-23). KoBold's $537M+ war chest from Gates Frontier, Bezos, and a16z (Chief AI Officer) sets the appetite ceiling in the category. Post-Series-A ownership is ~0.87% (172,459 shares of ~19.8M outstanding); post-dilution assume ~0.6% at exit. A $500M-$1B strategic outcome returns roughly 15-30× on Fund I's $200K.
Completed milestone with OMV, some basic quantitative things that they hadn't been able to.
Delivering target to their copper mining client for brown field extension.
Prepping for large Series A this summer/fall; will need a lot of work on the pitch.
JV with Khosla — Vinod Khosla has a strong view on the value of geophysical data for greenfield exploration. Xcalibur Smart Mapping (multinational with ~$150M in revenue) has a massive amount of geophysical data, perhaps the most of any company in the world.
Khosla is structuring a JV that would receive exclusive and first access to Xcalibur's data specific to greenfield exploration in under-explored geographies. Occurred because Xcalibur is doing a Series F and approached Khosla about the financing; some urgency because following the Series F there may be new Xcalibur leadership which could impact whether the deal happens. Raj is driving the deal as Vinod is on vacation.
Structure: Khosla $8M, Xcalibur $2M. 30% Xcalibur / 40% Khosla / 10% Terra AI (in exchange for aspects of the 3D probabilistic AI/ML visualization technology) / remainder option pool. JV is an incubation and Khosla is recruiting the founding team.
Strategy: JV entity to do model development in-house and staff up an AI team. This is not ideal for Terra AI; this is what Terra is commercializing. The entity might also compete for the support of Khosla, for AI/ML talent, and for business. Focus greenfield initially (Terra AI is brownfield initially).
Summary: Mixed development for Terra AI prospects. 10% of this new entity and access to this data is potentially tremendously valuable; Khosla is now also very conflicted. Reading between the lines, Khosla values the technical expertise of the Terra AI team, but not their leadership. Khosla has not seen Terra's code beyond the level of a PPT and whiteboard summary.
Next steps: given the potential value of this asset and the deal still in negotiation, would be good to have Matt speak with the team soon to offer his perspective on how to navigate this new development.
What they're working on: figuring out the right contract structure for prospectors, junior exploration companies, and majors, at different stages. Sent out 112 emails; had 22 discovery calls from those emails.
They have an early-lifecycle (though post-prospecting) project with Rio Tinto — priority. Iro Copper, small-cap $2B company prospect — priority. Aluminum mine wants to look at data around extension rates.
Structures: Iro — cash bonus for successful extension; likely to go to an existing mine site with ~5 years lifecycle remaining plus explored/unexplored land nearby. Initial cash for service fee, cash bonus for discovery/extension, then some portion of total value over several years. Rio Tinto moving very slowly but could be very high impact.
Value-add: drilling is really difficult; figuring out how or when or if to drill is the most impactful question. Brownfield work helps decide how to drill. A lot of people drill in a grid pattern — Terra helps folks drill more dynamically. Even a miss is very informative.
Team: 2.5 technical hires. Bottleneck has been hiring the AI folks. Wants a chief of staff from industry and a geoscientist from industry who can do more traditional/logistical sides.
Training data: some open-source geophysics data; drilling data is unusable.