AI businesses must be judged on the cost of running models and on what protects them from competitors using the same models. Deep tech companies face technical risk and long timelines before revenue, in exchange for harder-to-copy advantages.
12 min read · Level 4
I think the most common error in AI investing is confusing access to a powerful capability with owning an advantage. When everyone can use the same models, the lasting value tends to sit in data, workflow and distribution rather than in the model itself, and those are the things worth diligencing.
AI applications. Many AI products are sold like software but cost more to deliver, because each customer request uses computing power.
Revenue 10,000,000 Model and computing costs 4,000,000 Other delivery costs 1,000,000 Gross profit 5,000,000 Gross margin 50%
A traditional SaaS company might have a gross margin of 75% or more. Lower margins leave less to fund growth and change how the business should be valued.
Defensibility is the other central question. If a product is built mainly on models that competitors can also access, advantages must come from elsewhere:
Source of advantage Example Proprietary data data others cannot easily obtain Workflow integration deeply embedded in how customers work Distribution existing access to customers Specialised performance results general models cannot match
Deep tech. Companies built on scientific or engineering breakthroughs - new materials, advanced computing, robotics - face different questions:
Technical risk will the technology work at scale Timeline often many years to revenue Capital needs large investment before commercial proof Milestones funding usually tied to technical proof points
The reward for surviving these risks can be advantages that are genuinely hard to copy.
Because applying software assumptions to AI companies can overstate margins and defensibility, and applying software timelines to deep tech can make viable companies look like failures.
Both sectors also move quickly in public attention, which makes it easy to pay for a narrative rather than the specific economics and risks.
Educational, not personalised financial advice. Nothing in the Investor Academy is a recommendation to buy, sell or hold anything — what is right for you depends on your own circumstances and the rules where you live.
Two AI startups with 5,000,000 of revenue growing quickly.
Measure Company A Company B Gross margin 35% 68% Main advantage interface on a public model proprietary medical imaging data Price change when model costs fall competitors cut prices too maintains pricing
Company A's margin is thin and its product could be copied by others using the same model. Company B's data creates an advantage competitors cannot easily replicate. Their revenue growth looks similar; their businesses are very different.
Valuing AI companies with software margins and defensibility assumptions without checking computing costs and what prevents competitors from building the same product.
Ask what share of revenue goes to model usage and infrastructure, and how that changes with usage.
Ask what the company has that a competitor using the same models could not quickly reproduce.
List the technical proof points needed before commercial revenue, and the capital required to reach each.
For deep tech, assume milestones take longer than planned and check whether funding would survive delays.
When evaluating AI application companies or science and engineering-based startups.
Do not assume all AI companies have thin margins or weak defensibility. Some have strong data advantages and improving cost structures; the analysis must be specific.
Commentary from venture investors and reporting on AI startups during 2023 and 2024 widely discussed lower gross margins at many AI application companies compared with traditional software, driven by model usage and computing costs, and noted pricing pressure as underlying model capabilities became widely available.
Examine gross margin and pricing power carefully rather than assuming AI products behave like traditional software.
Create a free account to read the rest and apply it to your own investing.