Is your AI startup a wrapper or a company, and how do investors tell?

What is an AI wrapper?

It is a product that is mostly a thin layer over a model someone else runs. That is our working definition, not a term either source defines. Troutman puts the problem in one line: "If a competitor can recreate your company or your AI product quickly using publicly available tools, you lack strong defensibility."

Are AI wrappers dead?

Neither source says so. ICONIQ's 2026 survey of over 300 executives at software companies building AI products found that close to two-thirds are working on horizontal or vertical AI applications. Building on the application layer is the common choice, so it is not the problem.

The test is what sits around the model. Troutman writes: "AI defensibility is not about the model, it's about what surrounds the model: data, users, workflows, and positioning." It names four places a moat can come from:

1. Proprietary data and algorithms: exclusive data, properly obtained.
2. Operational dependence and stickiness: AI embedded in customer workflows, so switching costs build.
3. Scale, distribution and network effects.
4. Industry knowledge and regulatory expertise, for example in healthcare, finance and data privacy.

ICONIQ points the same way on products. It says a product embedded in a real workflow appears to be harder to dislodge than a general-purpose model one layer up.

Are AI wrappers profitable?

We found no 2026 source that measures wrappers separately. The nearest primary data is ICONIQ's survey. It reports gross margins on AI products of 45% in 2025, a projected 53% in 2026 and 59% in 2027, and says two-thirds of companies report better per-query unit economics.

It also reports that companies run an average of about 3.3 models, with licensed third-party APIs the most used type. And it flags a cost surprise: one builder watched an agent workflow budgeted at ten cents a run drift past a dollar fifty as agents retried and corrected themselves.

These are survey averages across software companies, not a benchmark for your product. An investor will want your gross margin after model and inference costs, month by month.

What paper do investors read about your AI?

Troutman says AI-specific diligence and representations are rapidly becoming standard in deal documents. Among the representations it lists:

• Disclosure of all third-party foundation models, LLMs, model APIs and fine-tuning services used in your products, and compliance with their terms of use.
• That no such usage gives the provider ownership of your prompts, outputs, embeddings or confidential information.
• That your AI inputs were obtained and used in line with the applicable terms and laws.
• That written claims about what the AI can do are true and supported, with known limitations disclosed.

The article also says patents are hard to get for generic use of off-the-shelf models, and that trade secrets are where "the real moat resides". Its warning is that AI products without trade secrets or truly novel inventions are weakly defensible from a pure IP standpoint.

What does an investor want ready?

Put the two sources together and the diligence list has four items:

1. Which model providers the product depends on, and on what terms.
2. Where the product's own data comes from, and the right to use it.
3. How the product sits inside the customer's work: usage, retention, and how long a customer would take to switch.
4. Gross margin after model costs, and its trend.

The numbers an investor reads next are in The ten numbers an investor reads first. For the cash side, see how much runway a startup needs when it already has revenue.

Work through your answer with us

If a founder wants to say the answers out loud before an investor asks, the Brutally Honest Mentor AI is an AI voice mentor we built and trained on Ozan Sonmez's method. The first 20 minutes are free.

TBHM Growth, our four-week online programme for founders whose company has revenue, works through gross margin, burn and runway in its third week. The next cohort is announced on the Growth page.