TechRadar analysis warns AI wrappers face moat erosion as model providers expand

The analysis defines an AI wrapper as a product whose core function is an interface over a third-party foundation model’s API, with little or no proprietary technology beneath that layer. It points to Jasper as an example. Once a standout in AI writing with a peak valuation of $1.5 billion, Jasper saw its core market undercut when ChatGPT introduced free, native writing features, according to the article. Its revenue fell from about $120 million in 2023 to about $55 million in 2024. Jasper survived by shifting toward enterprise workflows and brand-specific training.

The article also cites analysis showing that 81% of AI startups that wound down in the United States in 2025 were “wrappers and apps.” It describes that category as copilots and assistants layered on foundation models, AI productivity tools, content generators, and vertical SaaS products whose main differentiation was an LLM front end rather than proprietary data or infrastructure. In Jasper’s case and many others, the analysis says, the pressure came not from a competitor out-executing the company but from the platform it built on absorbing the feature the company had built around.

The analysis cautions against treating all wrappers as equal. A thin prompt-and-display wrapper differs from a product that does heavy retrieval, tool orchestration, agentic workflows, or domain-specific fine-tuning, and the latter can have real defensibility even while using a third-party base model. Some companies use multiple foundation models or fine-tune and distill smaller models for specific tasks after prototyping on a frontier API, and open-weight models add further variety.

Even so, the article says AI products across the wrapper-complexity scale share a problem: when a company’s core value proposition depends entirely on an API call to a third party, the company does not fully control its product. It cannot control pricing changes, rate limits, deprecations, shifts in model behavior, or the possibility that the platform provider decides to compete directly in the same vertical. The analysis says this is not a theoretical risk, though the effect on a given company likely depends on how much of its value sits above the model layer.

On defensibility, the article argues the answer comes down to intellectual property. Where a company has an existing distribution advantage, established customer relationships, regulatory licensing, network effects, or years of brand trust, model-layer differentiation may matter less. But if what makes a wrapper distinctive is observable from outside the organization and replicable by a competent competitor within a limited period, the analysis expects it will eventually be replicated. The cost of replication is near zero for the platform that controls the underlying model and near total for the startup that does not, according to the article. A wrapper’s moat, if it has one, must come from something other than the model call itself; absent that, the analysis says a lasting competitive advantage is unlikely.

The article acknowledges that model ownership is one solution but says it is easier said than done. It attributes the small number of companies owning their models less to a lack of desire than to financial constraints. Training a frontier-competitive model requires large compute budgets, specialized talent, proprietary or expensively licensed data pipelines, and infrastructure to run reinforcement learning at scale, putting model creation out of reach for all but the best-funded companies, according to the analysis. Open-weight models are one force pushing the barrier down, but the article says hosting and fine-tuning an open-weight model is still no small task and demands technical work.