The Wrapper Wave
In 2024 and 2025, it was the fastest path to a startup: take an OpenAI or Anthropic API, build a specialised interface around it, and sell access as SaaS. AI copywriting for marketers. AI summaries for lawyers. AI analysis for recruiters. The idea was always the same: take a large language model, wrap it nicely, and sell it to an audience that doesn't know how to write prompts.
Hundreds of startups did exactly that. Most of them will not survive 2026.
The Fundamental Problem: No Moat
A moat is what protects a business from competition. In traditional software, that might be a network effect (Facebook), proprietary data (Bloomberg), deep integration (Salesforce), or sheer technical complexity (AWS).
AI wrappers have none of these advantages. Their entire value proposition rests on an API call to a model that is available to everyone. The only differentiator is the interface – and an interface alone is not a business model.
The problem is compounded by the speed at which model providers themselves ship features. Claude has Projects and Artifacts. ChatGPT has Custom GPTs and Canvas. Every feature that a wrapper startup once claimed as its unique selling point can become a native function of the base model overnight.
The Margin Trap
The economics of AI wrappers are brutal. A typical setup:
Revenue: £29 per user per month.
API costs: Depending on usage, £5–15 per user per month. Significantly more for power users.
Gross margin: 50–80% sounds good – until you factor in the remaining costs.
Customer acquisition: In a market with hundreds of similar products, CAC explodes. Google Ads for "AI tool for X" are expensive because every wrapper provider is bidding on the same keywords.
Churn: When the only advantage is the interface, users switch quickly to the next tool – or straight to the base model.
The result: most wrapper startups burn more money on customer acquisition than they will ever recoup in customer lifetime value.
What Happens When the Base Model Improves
Every model update is a risk for wrapper startups. When Claude or GPT gains a new feature that covers the wrapper's core use case, the startup loses its reason to exist overnight.
We have already seen this play out: when OpenAI introduced Code Interpreter, dozens of data analysis wrappers became redundant. When Claude gained Artifacts, many content generation tools lost their edge.
This risk is not hypothetical – it is the default trajectory. Model providers have every incentive to integrate the most popular use cases of their API customers as native features.
What Works Instead
Not every company building on LLMs is a wrapper. There are clear patterns that work:
Proprietary data + AI: Companies that own unique datasets and make them accessible via AI have a real advantage. A legal advisory platform that has analysed thousands of real contract templates offers more than a ChatGPT prompt.
Deep workflow integration: Tools that embed into existing systems – ERP, CRM, accounting – and add AI capabilities there are hard to replace. The value lies not in the model, but in the integration.
Specialised agents: Instead of a chat interface, build a system that autonomously completes tasks: monitoring, data processing, quality assurance. Agents embedded in processes have a switching cost that a chat interface does not.
Infrastructure for AI: Sell the pickaxes, not the gold. Evaluation tools, prompt management, model routing, cost optimisation – these are products that grow with the AI market rather than being cannibalised by it.
The Exceptions
There are wrappers that will survive – but they have one thing in common: they are no longer really wrappers. Over time, they have accumulated proprietary data, fine-tuned their own models, or built such deep workflow integration that the API call is just one component among many.
The transition from wrapper to genuine product is possible, but it requires a deliberate strategy. If after a year you are still primarily forwarding API calls, you have missed the turn.
What This Means for Decision-Makers
For businesses evaluating AI tools, the wrapper question is an important filter: does this tool provide genuine, independent value – or am I paying for a pretty interface around something I could do myself with the API?
At nh labs, we advise our clients to evaluate AI solutions against three criteria: does the tool own proprietary data or logic? Is it deeply integrated into existing workflows? And would it still deliver value if you swapped out the underlying model? If the answer to all three is "no", it is a wrapper – and probably not a sound investment.
Conclusion
The AI wrapper wave was a natural first step. Every new technology initially produces a wave of simple applications before the truly valuable products emerge. In 2026, the wheat is separating from the chaff. The startups that survive are those that have created real value beyond the interface – through data, integration, or automation. The rest will be a footnote in the history of the AI boom.