Every founder building in B2B technology right now has added AI to their story. The pitch decks mention it. The product roadmaps include it. The website leads with it.
Buyers know this. And the first thing a sophisticated acquirer does when they see an AI-forward company is ask a question they rarely ask out loud: is this actually proprietary, or is this a thin layer over someone else’s infrastructure?
The answer to that question determines whether your AI strategy adds meaningfully to your valuation or quietly dilutes it. And in 2026, with AI capabilities becoming increasingly commoditized at the infrastructure level, the gap between a genuine AI asset and an AI-branded IT services business is wider than most founders realize, and more consequential than most expect when they enter a process.
Why This Has Become the Central Question in AI M&A
Three years ago, having any credible AI capability in your product was a differentiator. Buyers were early in their understanding of what AI actually meant for software businesses, and the presence of machine learning features, intelligent automation, or natural language interfaces commanded attention simply by existing.
That is no longer true. The explosion of foundation models, the widespread availability of APIs from OpenAI, Anthropic, Google, and others, and the rapid commoditization of AI infrastructure has fundamentally changed how sophisticated acquirers evaluate AI claims. Building a product that uses AI is now trivially easy. Building a product where AI creates defensible, durable, compounding value is still hard, and buyers know the difference.
The result is a bifurcated market. Companies with genuinely proprietary AI capabilities, those where the AI is deeply embedded in a unique dataset, a defensible technical architecture, or a workflow that creates compounding advantage over time, are commanding premium valuations in acquisition conversations. Companies that have integrated third-party AI capabilities without meaningful differentiation are being priced like traditional software businesses, or in some cases, lower, because the AI dependency introduces questions about cost structure, vendor risk, and replicability that a clean software business does not carry.
What Buyers Mean When They Talk About “Wrapper Risk”
The term that has become common in M&A diligence conversations is wrapper risk. A wrapper, in this context, is a product built primarily as an interface or workflow layer on top of third-party AI infrastructure, without meaningful proprietary technology underneath.
Wrappers are not inherently bad products. Many of them are genuinely useful, well-designed, and generating real revenue. The problem is that they are, by definition, vulnerable in ways that proprietary technology is not.
If your product’s core AI capabilities depend on an API from a provider who can raise prices, change terms, or build a competing product themselves, a buyer is underwriting that dependency alongside the business. They are also underwriting the question of how long it would take a well-funded competitor, starting today, to build a functionally equivalent product using the same publicly available infrastructure. If the honest answer is months, not years, the business has a structural replicability problem that affects what any rational buyer should be willing to pay.
This plays out in valuation in a specific way. A wrapper business with strong revenue and growth may still command a decent multiple, but it will be valued on a revenue or EBITDA multiple appropriate for a services or distribution business, not on the higher multiples that software companies with defensible IP typically achieve. The AI branding does not help, and may actually hurt if the due diligence conversation reveals a gap between the narrative and the underlying technical reality.
What Constitutes a Genuine AI Asset
The counterpoint to wrapper risk is what actually creates defensible AI value in the eyes of acquirers. There are several distinct sources of this value, and most companies that achieve premium outcomes in AI M&A have at least two of them in meaningful combination.
Proprietary Training Data
The single most powerful source of AI defensibility is data that a competitor cannot access or replicate. This can come from many places: years of customer interactions that have been used to fine-tune models to a specific domain, proprietary datasets generated as a byproduct of operating the business, exclusive data partnerships or licensing arrangements, or the accumulated behavioral data of a large user base trained on a specific workflow.
What makes this defensible is time and scale. A competitor starting today cannot acquire five years of your customers’ domain-specific interaction data simply by licensing the same foundation model you are using. The moat is not the model, which is increasingly available to everyone, but the data that shapes how the model performs in your specific context.
Proprietary Model Architecture or Fine-Tuning
Beyond data, companies that have invested meaningfully in customizing, fine-tuning, or building on top of foundation models in ways that create demonstrably superior performance in their specific domain carry genuine IP. This is different from simply calling an API. It represents accumulated technical investment that a competitor would have to replicate from scratch, and it is the kind of investment that shows up in model performance benchmarks, in customer retention data, and in the quality of outputs that users and customers can see and measure.
Deep Workflow Integration
AI that is embedded deeply in a business workflow, where removing it would fundamentally break the product, is structurally different from AI that has been added as a feature. When buyers evaluate workflow integration, they are looking for two things: how central the AI is to the core value proposition of the product, and how much switching cost the integration creates for customers. Both of these factors affect valuation, because they determine how defensible the revenue is against a competitor who builds a functionally similar AI capability.
Compounding Advantage Through Network Effects or Feedback Loops
The most valuable AI assets are those where the technology improves specifically because of customer usage, creating a compounding advantage that widens over time. Every customer interaction that generates data, which is used to improve the model, which creates better outcomes for all customers, builds a widening lead over any competitor starting from the same foundation model baseline. Buyers pay for this dynamic because it is the closest thing to a durable moat that AI companies can demonstrate.
How to Diagnose Where You Actually Stand
Before entering an M&A process, founders should be able to answer these questions clearly and specifically. These are exactly the questions sophisticated buyers will ask during diligence.
If your primary AI provider changed their API pricing by 50% tomorrow, what would happen to your margins and your product? A business that would face serious margin compression or a significant product rebuild has meaningful vendor dependency that buyers will price as risk.
If a well-funded team started building a competitor today, using the same publicly available AI infrastructure you use, how long until they could offer a functionally equivalent product? If the honest answer is under 12 months, you have a replicability problem. If the honest answer is three to five years, because of proprietary data, technical depth, or accumulated domain expertise, you have a genuine moat.
Can you demonstrate that your AI performs measurably better in your specific domain than a general-purpose implementation of the same foundation model? Buyers want to see benchmarks, customer outcomes, and retention data that prove your AI is not just using the same tools as everyone else, but using them in a way that creates outcomes competitors cannot easily match.
Is the AI embedded in the core workflow, or is it additive? Products where removing the AI would make the core value proposition stop working are fundamentally more defensible than products where the AI is an enhancement layer on top of a conventional software foundation.
How to Position Your AI Strategy Before Going to Market
If the diagnostic above reveals genuine defensibility, the work before an M&A process is largely about documentation and narrative. Buyers cannot value what they cannot see, and many companies with genuinely strong AI assets underperform in M&A conversations simply because they have not told the story in the terms that matter to acquirers.
Document your data assets specifically. What data do you have, how was it generated, what does it enable that a competitor starting fresh could not replicate, and how does it continue to accumulate and compound as the business grows? This documentation belongs in your data room and in your management presentation, not just your pitch deck.
Demonstrate AI performance with measurable outcomes. Customer retention rates, output quality benchmarks, time-to-value metrics, and head-to-head comparisons with alternatives are all evidence that your AI creates real, measurable value. Qualitative claims about AI capabilities are common. Quantified, verified outcomes are relatively rare and carry significant weight in diligence conversations.
Separate your AI IP clearly from your general technology stack. Buyers who cannot easily see what is proprietary versus what is licensed or commoditized will assume the worst. A clear technical architecture document that explains what you have built versus what you are using from third parties is worth preparing well in advance of any formal process.
If the diagnostic reveals significant wrapper risk, the more important question is whether the risk can be reduced before going to market. In some cases, the answer is investing in proprietary data collection or model fine-tuning over the 12 to 18 months before initiating a process. In other cases, the right positioning is to be honest about what the business is, price it accordingly, and find buyers who are acquiring for commercial traction and customer relationships rather than for technical defensibility.
Frequently Asked Questions
What is the difference between a wrapper AI company and a proprietary AI company? A wrapper AI company builds primarily on top of publicly available foundation models and APIs without meaningful differentiation in the underlying technology. A proprietary AI company has unique data assets, custom model architectures, or deeply embedded AI workflows that a competitor cannot replicate simply by using the same public infrastructure.
Does using foundation models like GPT or Claude automatically make my company a wrapper? Not necessarily. Many companies that build on foundation models have created genuine defensibility through proprietary fine-tuning, unique training data, or deep workflow integration. The question is not which infrastructure you use, but what you have built on top of it that creates durable differentiation.
How does wrapper risk affect valuation in an M&A process? Businesses with significant wrapper risk are typically valued on revenue or EBITDA multiples appropriate for services or distribution companies, rather than on the higher software multiples that genuinely proprietary technology commands. The AI branding does not close this gap if diligence reveals underlying replicability.
What is the most defensible form of AI IP in 2026? Proprietary training data generated as a byproduct of operating the business over time is generally the most defensible, because it cannot be acquired simply by licensing infrastructure. Model fine-tuning, deep workflow integration, and feedback loops that improve performance through customer usage are also strong defensibility factors.
How should I present my AI strategy in an M&A process? Lead with documented evidence rather than narrative claims. Proprietary data assets with specific descriptions of what they enable, measurable AI performance outcomes, a clear technical architecture showing what is proprietary versus licensed, and customer retention data that demonstrates the value of the AI integration are far more compelling than general statements about AI capabilities.
Telegraph Hill Advisors is a boutique investment bank based in San Francisco that specializes in technology M&A, including AI, machine learning, and enterprise software transactions. We have advised on 250+ transactions for founder-led technology companies and understand exactly how acquirers evaluate AI assets in the current market. If you want to understand how your technology stack positions you in an M&A process, we are happy to have that conversation.