According to internal Google data cited by The Information, API calls to the Gemini family via Google Cloud have surged rapidly. The number of API requests has jumped from around 35 billion in March last year, when Gemini 2.5 was released, to about 85 billion in August—more than doubling in just a few months.

The launch of Gemini 3 in November sparked another wave of adoption and received widespread positive feedback. This growth is not only visible in volume, but also in quality and profitability. Early versions, such as Gemini 1.0 and 1.5, were sold at heavy discounts and carried negative margins. With the rollout of Gemini 2.5 and later versions, improved model quality has allowed Google to move away from a pure “price war” toward a “quality-driven” strategy, turning marginal profitability positive.

Last autumn, Google guided that its capital expenditures would reach between USD 91 billion and USD 93 billion—almost double the USD 52.5 billion spent in 2024. Investors are now closely watching the upcoming Q4 earnings release for signs that this massive capex is starting to generate tangible returns.

On the software application side, Google is trying to lift margins through Gemini Enterprise. A Google spokesperson said Gemini Enterprise currently has 8 million subscription users across 1,500 companies, plus more than 1 million additional users who have signed up online.

However, market feedback has been mixed. Consulting firm Sada noted that client reactions are far from uniformly positive, with the ratio of customers who like versus dislike the product roughly split 50–50.

Part of the challenge lies in Google’s developer-first DNA. Google Cloud has historically felt more like a platform for “builders” than a “ready-made products” cloud. As a result, many customers prefer to use the underlying Gemini models directly to build their own custom agents, rather than buying Google’s pre-packaged software suites.

Market commentary:

While Gemini Enterprise performs strongly when answering general questions based on corporate data, it still struggles with more specialised or workflow-specific tasks. Even so, clients are not abandoning it the way some have with rival offerings; instead, many are adopting a “let’s keep testing it” attitude and continuing to experiment with Google’s AI stack.