As the competition in large models shifts from parameter scale to commercial deployment, Google is quietly reversing the profitability trajectory of its AI cloud business through the rapid iteration of Gemini. According to internal data cited by The Information, the number of Gemini API calls via Google Cloud surged from 35 billion at the release of Gemini 2.5 in March last year to 85 billion in August, a growth of over 140% in half a year; the launch of Gemini 3 in November has triggered another wave of usage, accelerating enterprise adoption.

This growth is not just a victory in quantity but also a turning point in the business model. Early versions of Gemini 1.0 and 1.5 captured the market with significant discounts, resulting in long-term negative profit margins. However, with the significant improvements in reasoning, multimodal capabilities, and contextual understanding in versions 2.5 and 3.0, Google has escaped the "price war" quagmire and shifted to a positive marginal profit driven by performance premiums. A source revealed that the unit economics of the new version API have turned positive, marking the beginning of real returns from AI investments.

The leap is supported by unprecedented capital investment—Google's capital expenditures for 2025 are expected to reach $91 to $93 billion, nearly double that of 2024, mainly used for building AI data centers and custom chips. Investors are closely watching the upcoming Q4 earnings report, seeking clear signals that the massive spending is translating into revenue.

In the enterprise market, Google is pushing Gemini Enterprise, which has already attracted 1,500 enterprise customers, 8 million subscription users, and over 1 million online registered users. Google emphasized that its cloud business is showing strong overall momentum, especially in the AI application layer.

However, market feedback has shown clear polarization. Simon Margolis from consulting firm Sada pointed out that customer evaluations of Gemini Enterprise are "almost split evenly"—those who like it praise its fast response and deep integration, while those dissatisfied complain about its weakness in specific business scenarios (such as financial compliance and supply chain optimization). He said directly: "Google is more of a 'builder's' cloud than a 'buying product' cloud." Many customers prefer to call the base model and develop their own agents rather than use pre-packaged solutions.

Analyst Chirag Mehta from Constellation Research also observed similar phenomena: although Gemini performs well in answering general questions based on corporate knowledge bases, it still has limitations in executing complex workflows. However, unlike some competitors who face a "use and discard" situation, customers generally adopt a "try again" attitude, showing long-term trust in Google's technical foundation.

The AI cloud transformation driven by Gemini is standing at a critical crossroads: one side is the dual upward curve of API call volume and profit margin, and the other is the ongoing challenges of enterprise implementation depth and product usability. As the AI competition enters the threefold test of "needing scale, profit, and experience," whether Google can truly transform its technological advantages into an irreplaceable commercial moat remains to be seen in the next report and the next version update.