Gemini has become a hit with developers, but Google is stalling with its AI products

Over five months—from March to August 2025—the number of Gemini API calls grew from 35 billion to 85 billion, representing a 143% increase. The company attributes this surge to the release of the Gemini 2.5 model and its quality improvements, which led to a noticeable shift in developer preferences toward Google’s solutions.

According to sources, demand for the API was so high that Google had to optimize model delivery and redistribute computing resources to free up capacity. Within the company, this is viewed as a «good problem,» indicating real market adoption of the technology.

The economics of the models have also changed. Earlier versions of Gemini 1.0 and 1.5 operated at negative margins and were used as unprofitable growth drivers. Gemini 2.0 only occasionally broke even, whereas Gemini 2.5 has shown consistent positive margins for the first time by competing on quality rather than price. These calculations don’t yet account for research and development expenses, but the overall trajectory points to the formation of a viable business model.

At the enterprise software level, the picture is less clear. Gemini Enterprise adoption has reached approximately 8 million seats, though active daily usage and satisfaction vary. According to KPMG, 83% of customers report being satisfied with the product, while consulting partners report a pattern that is «almost 50–50.» This suggests the solution works well in specific scenarios (such as deep research and document generation) but falls short of expectations in more specialized tasks and when creating custom AI agents.

Google’s strategy itself creates additional tension. The company’s strength—a «cloud platform for developers» where building custom solutions is easy—simultaneously undermines sales of ready-made enterprise products. Many clients prefer to build their own tools using the API instead of purchasing Gemini Enterprise.

Google expects that investments will still pay off through the flywheel effect: API expenses stimulate growth in complementary cloud services like data storage, databases, and computation. Even with thin margins on the models themselves, the additional cloud revenue justifies the investment.