The constraint nobody wanted

In 2018, the founders couldn't earn trust the normal. So they built a substitute: computational model. A model so grounded in 100 years of geological data. Most of it sitting in filing cabinets, handwritten, undigitized, treated as worthless by the industry that produced it. That BHP, a mining giant, handed them 500,000 square kilometers of Western Australia and said: tell us where to drill.

That outcome is not normal. It only happens when the outsider's substitute is more convincing than the currency the industry spent a century accumulating.

Being locked out of the normal path forced them to invent a path incumbents cannot copy. BHP already has geological credibility. They had no reason to build what KoBold built. And now they can't. Because KoBold built it first, from data BHP didn't think to rescue.

The moat, fully mapped

What makes KoBold difficult to analyze is that their moat is not one thing. It is five things, stacked, each one reinforcing the others. Pulling any single layer out and calling it "the advantage" is how the standard narrative gets it wrong.

Layer one: the dark data archive

A century of geological records, handwritten field reports, legacy drilling logs, regional maps from failed exploration companies, was sitting in physical storage, undigitized, legally orphaned. KoBold built TerraShed to rescue it before anyone understood what it was worth. That archive cannot be regenerated. The physical conditions that produced it are gone. The companies that produced it are gone. The data exists once. This is not a data moat in the software sense. It is a Disappearing Data Moat (you can read about it here) scarce not because someone locked it up, but because the world that produced it no longer exists.

Layer two: the cost structure nobody else has

Traditional exploration scales linearly. So every new project needs a new team of geologists, new equipment, new flights to Zambia. KoBold built Machine Prospector. It takes everything TerraShed digitized from old data and runs physics simulations of what the subsurface looks like across thousands of scenarios simultaneously, ranking each by probability. It tells you where the minerals are, at what depth, and at what angle to drill. A human geologist does this for one project at a time, from experience and pattern recognition. Machine Prospector does it for sixty projects across four continents, at the same time, for nearly zero additional cost.

Layer three: the real-time feedback loop

Conventional exploration drills, ships samples to a lab, waits weeks, then plans manually. KoBold updates its subsurface model in hours as live field data arrives. 

Layer four: Incumbent lock-in

Once BHP trusts KoBold's platform to direct where it sinks millions in drilling capital, switching becomes structurally impossible. The geological data generated through the partnership feeds back into KoBold's model. Leaving means severing access to data your own drilling produced. No contract clause needs to enforce the lock-in. The architecture enforces itself.

Layer five: intelligence for equity

This is the layer that ties everything together, and the one most analyses miss entirely. At the moment KoBold's AI identifies a deposit, they face the choice every data company faces: sell the insight, or act on it. A SaaS company sells. KoBold acts. They take equity in the ground before anyone else knows what's there. Intelligence is traded for ownership stakes, not licensing fees. The business model and the moat are the same architectural decision.

Mingomba is the proof. A copper deposit in Zambia's Copperbelt ¡, a region explored by traditional methods for a century. Traditional geologists had studied that land for a hundred years without seeing it. KoBold's pattern-agnostic AI found it by analyzing data without human preconception. They earned into the project before the discovery was public. That is information asymmetry converted directly into irreversible asset ownership.

Why incumbents cannot copy it

Rio Tinto and BHP are built to deploy heavy capital to move dirt. Their entire corporate structure exists to justify large physical deployments. Pivoting to an asset-light model where code acquires earth requires them to become a different company. Not a better version of themselves. A different one. That is the lock. Not a patent. Not a trade secret. A structural impossibility built into the incumbent's own success.

The pattern

ASML couldn't fund EUV lithography alone. So they invented customer co-investmen to raise funds. Today ASML holds a monopoly on the machines that make every advanced chip on earth. The capital constraint forced the architecture. The architecture became the moat.

NorthVolt couldn't manufacture outside EU standards. So they built inside the regulatory constraint  making European compliance the product. The constraint became the differentiation. They build a system that could raise as much as fund they wanted. 

KoBold couldn't earn credibility through track record. So they made their analytical case airtight enough to substitute for it, then used that substitute to earn equity in projects before the market priced the discovery.

In each case: the company lacked the industry's standard currency. It couldn't buy its way in. So it invented a substitute. And the substitute, built under pressure, turned out to be harder to copy than the original currency would have been.

The framework, for founders

Stop leading with what you built. Lead with what stopped you from doing it the normal way. Those are not the same story. Only one of them is defensible.

Three questions. Answer them about your own company.

What didn't you have that the industry treats as table stakes? The actual entry requirement that every incumbent assumed you needed before you could compete. Name it. 

What did that force you to build? Not a workaround. An architectural invention. Something that doesn't exist anywhere in your industry because no one who had the standard currency ever needed to build it. If the answer is a product feature, look deeper. The invention is usually structural: a business model, a partnership architecture, a data acquisition mechanism disguised as something else.

Can an incumbent copy it without dismantling themselves? If a well-funded competitor can replicate your invention by writing a larger check, it is a head start, not a moat. The real test: does copying your architecture require them to give up the thing that makes them an incumbent? If yes, you are structurally protected. Not by law. By the weight of their own success.

If you have clean answers to all three, you have a constraint-built moat. Now rewrite your pitch around the absence, that’s exactly how I start each project. The constraint is the argument. That is what defensibility looks like when capability is no longer scarce.

One question stays open. Does the moat outlive the constraint? KoBold post-Series D, $537M raised, $2.96B valuation is no longer capital-constrained. The pressure that generated the architecture is gone. Whether the architecture survives abundance is the question this newsletter returns to. The early signal: Manono was not found by TerraShed. It was purchased with a checkbook. A different company is emerging inside the same name. We are watching to see which one wins.

References 

Daigle, C. (2025, October 23). How Kobold Metals found 300,000 tonnes of copper using AI (that traditional geologists missed). Chief AI Officer.

Dotson, K. (2025, January 1). AI-powered mineral exploration company KoBold Metals raises $527M. SiliconANGLE.

Harper, G. D. J. (2021, November 20). The geopolitics of cobalt. American Affairs Journal, 5(4).

KPMG International. (2025, April 16). Venture pulse Q1 2025: Global analysis of venture funding.

Mining Engineering Online. (2021, September 8). BHP teams up with KoBold Metals to search for critical minerals. Society for Mining, Metallurgy & Exploration (SME).

Mongabay. (2025, May). US firm KoBold Metals buys stake in contested Manono lithium project.

Porter, S., Sanders, I., Diasselliss, J., Johnson, B., Hardin, K., & Purohit, A. (2024, May 13). Promoting metals and mining sustainability in critical supply chains. Deloitte Insights.

Presidential Delivery Unit Zambia. (2025, June 26). Mingomba mine project takes shape. Government of the Republic of Zambia.

The Economist. (2023, November 1). Ground truths: Could AI help find valuable mineral deposits?

The Economist. (2024, October 31). Greenland faces one of history's great resource rushes—and curses.

The Economist. (2025, July 10). Want to be a good explorer? Study economics: The battle to reduce risk has shaped centuries of ventures.

The Economist. (2025, December 18). Minefield: Meet the American investors rushing into Congo.

Yee, L., Chui, M., Roberts, R., & Smit, S. (2025, July). Technology trends outlook 2025. McKinsey & Company.

Yu, A. (2023, February 1). Report: KoBold Metals business breakdown & founding story. Contrary Research.