In parallel, The company details energy and water usage figures for the first time associated with Gemini's responses, as the ecosystem reacts: new AI-powered features in Photos and Apple talks about using Gemini on Siri They draw a map where this model gains prominence.
Gemini for the home: from speaker to home center
Google has announced Gemini for Home, a version of the AI assistant that will replace Google Assistant on compatible home devices. The proposal is based on two pillars: understanding requests in natural language and manage chained actions without the need for rigid commands.
With this approach, the user can formulate compound requests such as "turn on the lights and set the temperature to 22 degrees" and expect all actions to be executed. Contextual understanding also allows for exceptions, for example, "turn off all the lights except the one in my room."
Gemini will be integrated with Google's leading applications for solve everyday tasksA typical example is cooking: describe what's in the fridge to get recipes and, if necessary, add ingredients to the shopping list in one step.
During its hardware event, Google showed off a unprecedented smart speaker which has yet to be officially announced. The company hasn't confirmed any details, but everything points to a "Made by Google" device ready for the new era of its assistant.
The deployment will be progressive: The first home users will be able to access Gemini starting in October., and throughout the year it will replace Assistant on compatible devices.
Consumption and footprint: what figures does Google provide and what doubts persist?
Google has published an analysis of the environmental impact of its AI inference. According to its internal measurements, an average text query on Gemini requires 0,24 watt-hours, emits 0,03 grams of CO2e and use 0,26 milliliters of water, an order of magnitude comparable, they say, to “turning on the TV for less than nine seconds.”
The report breaks down consumption: around 58% is concentrated in specialized chips that process the request; near the 25% falls on CPU and memory that hold the load; and approximately one 10% is reserved for backup machines which only come into play if something goes wrong. The rest is attributed to cooling and power conversion in data centers.
The company claims that its methodology is broader than third-party estimates and, as a result, increases estimated consumption by 2,4 times compared to previous calculations by including components not usually considered. It also claims to have achieved a Up to 33x reduction in energy per query between May 2024 and May 2025 thanks to model, hardware and software optimizations.
There are relevant nuances: the Google study focuses on the text inference and does not cover model training, data storage, or more demanding multimodal loads such as image or video generation. It also doesn't detail the daily volume of requests or the average size of queries, leading some industry voices to urge caution when extrapolating these figures.
Gemini on the Pixel 10: Voice Editing and More Transparency
In mobile, Gemini powers “Edit Asking” in Google Photos: a feature that allows modify images with spoken instructions or written in natural language. The system offers contextual suggestions and makes adjustments such as cleaning backgrounds, retouching lighting, or highlighting subjects without the need for advanced tools.
For now, this new product is launched in exclusive for the Pixel 10, initially in the United States and in English. Google usually opens these features in stages, with the intention of bringing them to more languages and devices later.
In addition, the company will implement C2PA content credentials on the Pixel 10 to improve traceability: the origin and history of modifications are recorded, including the application and the AI model Used if the image was generated or edited with Gemini-based tools. These credentials will apply to photos taken with the camera app, even without further editing.
Another front on which Google insists is the local execution of AI capabilities on the Pixels, relying on its hardware platform to reduce latency, improve privacy, and, where appropriate, offload the cloud.
Moving Pieces: Apple Considers Using Gemini in Siri
Industry sources indicate that Apple maintains talks with Google to integrate Gemini as the backbone of a revamped version of Siri, planned for 2026. The idea would involve a custom model running on Apple servers, within your private cloud infrastructure.
In Cupertino they continue two parallel paths: an iteration of Siri based on its own AI (Linwood project) and another that evaluates external technology (Glenwood). Apple has also recently explored options with OpenAI and Anthropic; there's no final decision on which approach will prevail.
If materialized, the agreement would expand the list of collaborations between Apple and Google beyond the default search in Safari and would strengthen Gemini's presence among the industry's major players.
Gemini faces a key phase: It is installed in the living room with a more natural assistant, gains useful features on mobile and brings more transparent, though controversial, consumption figures to the table. With rival manufacturers considering its integration and a timeline pointing to October for the home, the role of Google AI in everyday life promises to gain ground without losing sight of its impact.
