Three-Layer AI Opportunity Scorecard
Choose the AI layer where capital barriers are low and proprietary value is high
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 97%
Bill Inman divides AI into three commercial layers: compute power, algorithms, and training data. Founders first classify an opportunity by the layer that supplies its core advantage. Compute includes chips and data centers, but Inman sees its capital requirements as too large for most entrepreneurs. Algorithms include GPTs and narrow AI products, where he sees a glut of competing offers. Training data is his preferred opening because a person or business can assemble specific, accurate knowledge that broad models do not uniquely own. The scorecard therefore compares capital barriers, crowding, and access to proprietary or legitimately collected data. Its output is a market-entry choice aligned with the founder's actual resources rather than the popularity of the AI label.
Origin
Inman offered the three-part breakdown when asked what an underdog founder should do in a hypercompetitive, oversaturated AI market.
Core principles
- 01AI businesses compete through compute, algorithms, or training data
- 02Compute carries a capital moat that excludes most new founders
- 03Algorithms face heavy competition and product saturation
- 04Distinct, accurate training data can create a defensible opening
How to run it
- 1
Map the Three Layers
Separate the market into compute power, algorithms, and training data. Place the proposed business in the layer that creates most of its value.
Pro tip Classify the source of advantage, not merely the product's user interface.
- 2
Test the Capital Moat
Estimate the infrastructure and funding required to compete. Reject a compute-led strategy when chip manufacturing or data-center scale is beyond the team's reach.
Watch out Do not mistake access to third-party compute for ownership of a compute moat.
- 3
Measure Saturation
Examine how many interchangeable algorithms or productivity offers already target the same need. Require a clear reason the proposed model or application will stand apart.
Pro tip Inman identifies algorithms as the layer where the current glut sits.
Watch out Adding an AI label does not create differentiation.
- 4
Inventory Ownable Data
List the knowledge, records, and content the founder can legitimately collect or control. Favor data that is specific enough to serve a person, business, profession, or location accurately.
Pro tip Start with information generated by your own work or assembled from genuinely public sources.
Watch out Do not treat other people's content as proprietary data.
- 5
Choose the Defensible Layer
Compare the three layers on feasibility, competition, and uniqueness. Enter where available resources can build a durable advantage rather than where the industry spends the most money.
Pro tip For a small founder, Inman's recommendation is the training-data layer.
In the wild
Inman proposes that an entrepreneur who wants to become the best AI real-estate agent in Newport Beach collect every relevant piece of publicly available local information and place it in a twin. Rather than building chips or another general model, the founder competes through narrow local knowledge that helps an international buyer understand the area's boats, events, and lifestyle.
→ The AI can educate prospective buyers and direct demand toward the entrepreneur's real-estate business.
Common mistakes
Entering Compute Without the Capital
Chip manufacturing and large data-center projects carry a moat that a typical new founder cannot cross. Classify the economics before choosing the layer.
Launching Another Interchangeable Algorithm
Inman describes algorithm products as the crowded part of AI. A generic productivity promise offers little protection from competitors.
Collecting Data Without Specificity
The opportunity comes from accurate, narrow knowledge for a person or business, not an undifferentiated pile of information.
Is it for you?
Best for
It is best for entrepreneurs evaluating where a small team can build a differentiated AI business.
Not ideal for
It is not ideal for well-capitalized semiconductor or infrastructure companies whose advantages already lie in compute.
From the transcript
“AI is three things. It's compute power.”
“the second part of AI is algorithms.”
“the third part of AI is training data and that's the part that I'm in that's the part I think is blue ocean”
From the episode
The Future of AI, Digital Twins & AGI ft. Bill Inman
Bill Inman