Google and Accel selected 5 Indian startups from 4,000 applications — and none was an AI wrapper

Google and Accel reviewed more than four thousand applications from startups connected to India as part of the selection process for the Atoms accelerator — and came to a discouraging conclusion: approximately 70% of candidates were building products commonly referred to as AI-wrappers. Five teams made it to the final round, and none of them exploit external models without adding their own value. The term "wrapper" has firmly entered venture capital lexicon over the past two years.

It refers to startups that take the API of a large language model and build a minimal interface on top of it with little proprietary development. Such products are easy to copy, they lack technological moat, and at the slightest change in provider conditions, the entire business is threatened. This is precisely why investors increasingly reject wrappers — even if their early-stage revenue looks convincing.

Atoms accelerator — a joint initiative of Google and venture fund Accel, is focused on startups with a focus on India. This year the program received a record flow of applications. The selection team reviewed products, spoke with founders, and assessed whether there was real technical differentiation behind the polished pitch deck.

The result proved telling: seven out of ten teams were building products that were almost indistinguishable from each other. The five finalists are described as startups with genuine technological substance. The organizers plan to reveal details about each company as part of the program.

However, the very fact that the selectors openly discuss the problem of wrapper culture in the Indian AI segment speaks volumes about market maturity. India today is one of the most active markets in terms of launched AI startups. Cheap engineering resources, a large domestic market, and the growing presence of technology corporations create fertile ground for experimentation.

But this very same set of factors provokes a wave of clones: it's easier to take a ready-made model, wrap it in a product for local needs, and try to attract funding than to invest in fundamental development. A similar picture is observed by accelerators in the USA and Europe, but in the Indian context it is particularly noticeable due to the scale of the application flow. For Google, participation in Atoms is not simply ecosystem support.

The company is looking for partners who will build on the basis of its infrastructure: Gemini API, Google Cloud, Vertex AI. Startups with real technological depth are potential long-term customers and partners, not one-time API users. Accel, for its part, has been investing in the Indian market for more than twenty years and is interested in shaping the next generation of companies with global potential.

The selection of five startups from four thousand is a signal to the market: the era of easy money for wrapper projects is ending. Institutional investors, including the largest technology corporations, are increasingly distinguishing between real AI companies and companies that simply rent someone else's intelligence. For founders, this means one thing: if the product cannot be explained without reference to the base model — it has serious positioning problems in the next funding round.