Google launches Android 17: but AI flagship Gemini Intelligence has to wait until summer and is very selective about phones

Google officially rolls out Android 17 to Pixel devices, with multi-tasking, community video features, and comprehensive security upgrades. However, the Gemini Intelligence agent AI suite—treated as the core selling point—will only be gradually opened up starting this summer, and it requires Gemini Nano v3 and at least 12GB RAM, directly excluding most Android phones released before 2026.
(Background: Google Translate upgrade: Gemini 3.5 makes real-time voice interpretation no longer awkward)
(Background: Bloomberg: Apple will launch a “Camera version AirPods” in 2027, with foldable phones and an iPhone 20th anniversary edition also coming)

Table of Contents

Toggle

• What is launching first
• Which models can upgrade now
• Why the AI core still needs to wait
• The pace of the arms race

On the 16th, Google officially pushed Android 17 to Pixel devices, calling it the most AI-centric Android major update in recent years. However, this “AI-centric” promise is still only half-fulfilled. The Gemini Intelligence—which Google effectively treats as its headline—won’t begin phased rollout until this summer, and it has strict hardware requirements, keeping most Android phones out.

The first to arrive is the “shell”: a multi-tasking interface, community video features, and enhanced security and privacy. Next is the “soul”: an agentic AI suite that can automate multi-step tasks.

What is launching first

The initial wave of Android 17 updates focuses on three major areas.

On the multi-tasking side, the new “Bubbles” feature lets users shrink individual apps into bubble form, which can be freely stacked and dragged to any position on the screen. For foldable phones, a dedicated Bubbles section is added to the taskbar, making multi-window multi-tasking on large screens more intuitive.

On the community video side, Google directly builds “Screen Reactions” into the system. Users can record videos while overlaying their own face on top of what the phone is currently displaying—this “camera + screen in the same frame” format has become popular on social platforms in recent years. Google’s intention is clear: to reduce incentives for users to switch to third-party apps such as CapCut and Instagram Edits by integrating it natively into Android.

On the security side, Android 17 adds “one-time location access,” so users no longer have to choose all-or-nothing location permissions for apps. In low-population areas, “approximate location” will intelligently expand the displayed coverage range, avoiding precise location tracking of users in remote areas. The lost mode also tightens restrictions: unlocking the device now requires dual verification via both biometric authentication and a PIN code, making it harder for thieves to resell stolen devices.

Which models can upgrade now

The first wave covers 21 Pixel models, from the Pixel 6 series of 2021 all the way to the latest Pixel 10 series. Pixel Tablet and Pixel Fold are also included. Updates for other Android brands will be rolled out gradually throughout 2026.

The complete list of Pixel devices being pushed starting today is as follows:

• Pixel 6, Pixel 6 Pro, Pixel 6a
• Pixel 7, Pixel 7 Pro, Pixel 7a
• Pixel 8, Pixel 8 Pro, Pixel 8a
• Pixel 9, Pixel 9 Pro, Pixel 9 Pro XL, Pixel 9 Pro Fold, Pixel 9a
• Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, Pixel 10 Pro Fold, Pixel 10a
• Pixel Tablet
• Pixel Fold

Why the AI core still needs to wait

The most talked-about feature of Android 17 is Gemini Intelligence, positioned as an “agentic AI” suite. In simple terms, it’s not just a chatbot that answers questions—it can take the user’s place to complete chains of multi-step tasks on the phone, such as automatically organizing calendars, summarizing web pages in Chrome, creating custom tools using natural language, and polishing voice messages with the new tool “Rambler.”

This set of features will start being rolled out in batches this summer to the Google Pixel 11 series and Samsung’s upcoming flagship devices, with broader availability expected later this year.

The root reason for the delay is the hardware threshold, not software progress. Gemini Intelligence’s on-device AI models require the device to have Gemini Nano v3, and it must be equipped with a qualifying flagship SoC (system-on-a-chip), with RAM of at least 12GB. Gemini Nano v3 is the newest generation of Google’s on-device model. Put plainly, it’s a small language model that runs directly on the phone’s chip and doesn’t require sending data to cloud servers for processing.

This design protects privacy and speed, but it also means that only flagship phones released in the past one or two years meet the requirements. Most Android phones released before 2026—including many mid-to-high-end models—are excluded because their RAM isn’t sufficient or their SoC generation is too old.

Android 17’s existing features vs. Gemini Intelligence: the former is usable today, while the latter will arrive this summer and only be opened to a limited number of new flagship models. This timeline and device threshold are the parts of Google’s rollout that truly need to be examined separately.

The pace of the arms race

Last week, Apple just announced a long list of AI and Siri improvements coming this fall, covering deeper system integration and multi-step automation capabilities. The two companies essentially previewed their AI plans for the second half of the year within the same month, but both announcements share a common characteristic: the real AI features still aren’t here yet—only the direction is being made known to the outside world.

Google’s strategy is to roll out the underlying OS infrastructure and user experience first, laying the groundwork for agentic AI; Apple, meanwhile, previewed its fall update at its developer conference. Both are doing the same thing: managing external expectations with a release schedule rather than delivering finished products.

This is a noteworthy turning point in the smartphone AIization wave: the speed of software updates has begun to be constrained by the physical requirements of on-device AI models for memory and chips. The release cadence is no longer determined solely by engineering progress—it also depends on the specifications of your previous phone.