TL;DR: Preferential access to a coveted asset that independently enhances value. The bar is high — five tests must pass — and most claims fail at least one of them. When you actually have it, it’s one of the most durable forms of Power because it locks in at origination, before anyone is paying attention.

What it means

A Cornered Resource isn’t about owning something valuable. It’s about having preferential access at attractive terms to something competitors can’t easily replicate, acquire, or substitute. The resource must independently drive value — not just be nice to have, but actually change the economics of the business in a way that a competitor literally cannot duplicate (seven-powers).

The canonical examples: De Beers’ supply agreements for rough diamonds. Qualcomm’s cellular patents. Pixar’s concentration of animation talent in the 2000s. In startups, the most common cornered resources are founding teams with rare domain expertise, proprietary datasets, and early regulatory positions that closed behind the founders.

The argument

Five tests, and most claims fail. Helmer’s bar is deliberately high. The resource must be:

  1. Scarce — others can’t get it.
  2. Relevant to value creation — not just impressive.
  3. Difficult to replicate — not just expensive.
  4. Durable in advantage — not just temporary.
  5. Maintainable in preferential terms — you keep your access on good terms over time.

Most “cornered resources” founders cite — “our team is amazing,” “we have unique data,” “we have deep partnerships” — fail one or more tests under scrutiny. Talent is liquid (people quit). Data can be scraped or purchased or synthetically generated. Regulatory moats get legislated away. Partnerships get unwound when an acquisition happens at the partner. If you can’t pass all five, you don’t have this Power. You have a head start. (seven-powers)

Locked in at origination. Along with counter-positioning, Cornered Resource is an origination-stage Power. You either start with preferential access or you don’t. You cannot retrofit it. This is why founding decisions matter so much more than most founders realize — the choice of co-founders, the initial dataset you train on, the regulatory filing made in week one — these early moves determine whether Cornered Resource is even available to you. If you skip them, the door closes silently and you spend years wondering why you can’t seem to build a moat (power-progression).

The Paul Graham connection. Graham’s emphasis on technical talent as the decisive startup advantage maps directly onto Cornered Resource. “In business, nothing is more valuable than a technical advantage your competitors don’t understand” (hackers-and-painters). A founding team with deep, rare technical ability is a cornered resource — if it passes the five tests. The key question almost no one asks: is the talent advantage durable (do these specific people create value nobody else can?) or temporary (will equally good people be available in 18 months when capability commoditizes)?

Data as the new cornered resource. In the AI era, proprietary data that populates the context window is emerging as the most important cornered resource (cheating-is-all-you-need). But the “We Have No Moat” memo warns that static data advantages erode fast as open-source models and synthetic data improve. The durable version isn’t a static dataset — it’s a pipeline that continuously generates data competitors can’t access. (data-moat) The moat is the conveyor belt, not the warehouse.

Temporal and geographic boundaries. Cornered resources have expiration dates. Patents expire. Supply agreements end. Talent gets poached. The Power is only durable if you reinvest in maintaining access or convert it into a different Power — like network-effects or process-power that compound from the initial advantage. Cornered Resource is the seed, not the harvest.

The OpenAI version

OpenAI’s most underappreciated Cornered Resource isn’t the model. It’s the researcher talent. For most of 2021–2024, a small group of frontier-AI researchers had genuine ten-figure expected value impact on whichever company employed them, and OpenAI had managed to concentrate a disproportionate share of them in one place (chatgpt-pmf). That talent concentration passes all five tests: scarce (a few hundred people in the world can lead frontier model training), relevant, hard to replicate (couldn’t be acquired even with billions), durable (so far), and maintained (heavy compensation, mission, equity). It’s the kind of cornered resource that predates the product and shaped which products were possible.

Loose threads

  • How does cornered resource interact with moats in the AI era? If models commoditize (google-no-moat) but data doesn’t, does “proprietary data” become the dominant cornered resource of this decade?
  • The founding team as cornered resource is underexplored. Rabois argues “the team you build is the company you build” (keith-rabois-lessons). Netflix’s talent density model (netflix-culture) suggests that maintaining a talent cornered resource requires continuous investment, not just initial hiring.

What links here