Gemini Intelligence needs about 12GB of RAM — limited to high‑end phones

Google said on May 12 that Gemini Intelligence will begin rolling out this summer on select Samsung and Google phones, pitching the new Android layer as a more proactive set of AI features built into the operating system. The company said the tools will automate multi-step tasks, summarize web pages, help fill out forms and create widgets from natural-language prompts. Google did not spell out the hardware bar in that announcement, but its developer materials and related product pages point to a narrow class of devices with top-tier chips and on-device AI support. That means the launch wave is likely to land first on recent flagship phones rather than the broader Android market. ### Which phones are Google targeting first? Google said the first wave will reach “select Samsung and Google phones” this summer. The company named the latest Samsung Galaxy and Google Pixel phones in its May 12 post, and said broader availability across watches, cars, glasses and laptops would follow later in 2026. (blog.google) Mindy Brooks, Google’s vice president of product management, wrote that Gemini Intelligence is meant for Google’s “most advanced devices.” That wording matters because Google is tying the experience to both software features and a specific class of hardware capable of running more AI tasks locally. ### What hardware does Gemini Intelligence appear to require? (blog.google) Google’s requirement list, as cited by 9to5Google from a footnote on the Gemini Intelligence page, calls for a “flagship chip,” 12GB or more of RAM, support for AICore, and Gemini Nano v3 or higher. The same report said devices also need at least five Android OS upgrades over their lifespan, six years of security updates and baseline quality thresholds such as low crash rates. (blog.google) Android Developers documentation describes AICore as the system layer that runs Gemini Nano on-device, manages model updates and applies safety features while using the phone’s local hardware. Google says that approach keeps prompts on the device, reduces network dependence and can support offline use, though performance still depends on the hardware. (9to5google.com) ### Why does the 12GB RAM number matter? The 12GB figure is the clearest sign that Gemini Intelligence is not being positioned as a standard Android feature that will simply trickle down through software updates. On-device models compete for memory with the rest of the system, and Google’s own developer pages say inference speed depends on device hardware. (developer.android.com) Industry reports published on May 15 said that threshold could exclude many midrange Android phones and even some recent premium devices. 9to5Google reported that Google’s Pixel 9 line and Samsung’s Galaxy Z Fold 7 remain on Gemini Nano v2, which would leave them short of the Nano v3 requirement. That assessment is based on Google’s published model-support information, as interpreted by the outlet. (developer.android.com) ### What exactly is Gemini Nano v3 doing here? Google’s Android developer site says Gemini Nano is the company’s on-device foundation model for text, image and audio tasks exposed through ML Kit GenAI APIs and AICore. Supported tasks include summarization, rewriting, proofreading, image description and speech recognition. A recent Android Developers post said Gemini Nano v3 expands language understanding and multimodal capabilities on supported devices. (9to5google.com) Google has also begun previewing work tied to later on-device models, including Gemini Nano 4 through the AICore Developer Preview, which suggests the company is moving quickly on the local-model stack even as v3 becomes the gatekeeper for current Gemini Intelligence features. (developer.android.com) ### Is this an Android 17 story, a Gemini story, or both? Google is framing it as both. The company’s Android materials describe Android as shifting from an operating system toward an “intelligence system,” while the Gemini Intelligence branding packages those features under the Gemini name. Google I/O 2026 is scheduled for May 19-20, and the official agenda lists the main Google keynote for May 19 at 10:00 a.m. (developer.android.com) Pacific time, followed by a developer keynote later that day. Google’s event pages say the conference will include keynotes, sessions and product updates spanning Android and Gemini, giving the company its next formal venue to clarify device support, rollout timing and developer access. (io.google) (blog.google)