They don’t build any models or train them, but they are the ones selling products that are closely tailored to businessneeds. They demonstrate ROI where deep tech companies are still only selling promises.
But if we look closer, this raises some concerns. I have already said that wrappers are wedged between deep tech companies, whose profitability is difficult to envision, and customers, thus capturing most of the market’s added value (AI heading for an economic dead end?). I’m not going to say that they live like parasites on the back of deep tech, especially since they bring them substantial revenue, but there’s no denying that they are doing better financially than the latter without having to bear the burden of heavy investment.
But what would happen if this model ended up weakening deep tech to the point where it collapsed? In other words: can wrappers survive if the technology that supports them stops evolving or disappears?
In short:
- Wrappers adapt existing AI models to business needs and capture most of the value.
- They are entirely dependent on costly and unprofitable deeptechs, which weakens their model.
- A collapse of deeptechs would threaten the entire application chain.
- Solutions exist: vertical integration, open source, value sharing, or shared infrastructure.
- The AI ecosystem will need to rebalance itself to become sustainable.
A fragile model of dependency
Today, many AI startups, often described as “powered by AI,” simply integrate one or more models provided by OpenAI, Anthropic, Google, or Mistral. Their product relies on API access, and they sell perceived business value, but they are entirely dependent on infrastructure they do not control.
This results in an asymmetrical business model, where deep tech invests heavily in R&D, infrastructure, and compliance, while the wrapper capitalizes on usage and distribution and captures user feedback to improve its product. These are users or customers that deep tech does not know and will never speak to.
This is sustainable as long as deeptech companies survive, but to date, none have demonstrated their economic viability. Training costs are skyrocketing, models are piling up, but the promises of monetization are slow to materialize. Today, more and more voices are being raised to say that the question is no longer “if” but “when” the system will reach its limits.
Towards a chain reaction collapse?
If one or more players were to slow down, pivot or disappear, the entire downstream value chain would be affected.
And contrary to marketing rhetoric, very few wrappers today have a serious alternative to using these proprietary models. The scenario of a “deep tech crash” would therefore cause a massive collapse of business applications dependent on these infrastructures, paradoxically due to the fact that these same wrappers intercept a large part of the value but can saw off the branch they are sitting on at the same time as they build their profitability.
How can we escape this trap?
This systemic dependency is not inevitable, and there are several scenarios that could prevent the house of cards from collapsing, but they must be activated before it is too late.
First, there is vertical integration.
Some wrappers will seek to internalize their own models, even smaller or more specialized ones, to secure their stack, while others will forge strategic or exclusive partnerships with a deeptech company, but this is no guarantee if the deeptech company goes under.
In both cases, it is a matter of regaining control over the foundations to stabilize the business model.
Then there are value-sharing agreements.
A new economic model may emerge between model producers and those who exploit them. This could take the form of revenue-sharing agreements, cross-equity, or pricing that is more aligned with the value actually captured, so that the growth of one does not come at the expense of the other’s future.
Without this, deeptechs will continue to fuel, at a loss, uses over which they have no economic leverage.
There is also a transition towards more frugal open source models, such as Deepseek.
The emergence of powerful and lighter open source alternatives can breathe new life into the chain. They allow the wrapper to gain technical and economic independence, but at the cost of greater integration efforts.
Finally, there is coopetition or mutualization.
It may be necessary to devise collective infrastructures, financed by coalitions of players or public funds, to guarantee a stable, open, and independent technological base. After all, if states consider AI to be a strategic issue of sovereignty, it is conceivable that they would support deep tech in this way to ensure the survival of the entire ecosystem that depends on it.
Bottom line
The illusion of an ecosystem where everyone can build on AI APIs without worrying about their longevity is not sustainable in the medium term.
Wrappers capture value today, but are built on foundations that are too fragile to guarantee sustainable growth. Ultimately, a choice will have to be made between vertical integration to control the technology, rebalancing the value chain, or building a new shared infrastructure.
The AI economy cannot rely solely on outsourcing models and, like any industry, it will have to find a technical, economic, and even political balance if it is to survive.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
