Google's new Pixel 9 lineup goes all-in on on-device AI

For years, phone launch events followed a tired script of megapixel counts, peak brightness specs, and benchmark scores. Google’s Made by Google event on August 13 threw that script out the window.

Hardware chief Rick Osterloh spent the first 25 minutes barely acknowledging the new hardware. Instead, he walked through Google’s AI stack, showing off features on Samsung and Motorola phones before even pointing a camera at the new Pixel.

When the Pixel 9 family finally showed up, one message was crystal clear: every piece of silicon is now built to serve Gemini.

Hardware rebuilt for local intelligence

The lineup splits into four devices: the standard Pixel 9 ($799), the Pixel 9 Pro in a compact 6.3-inch size ($999), the 6.8-inch Pixel 9 Pro XL ($1,099), and the Pixel 9 Pro Fold.

Under the hood sits Google’s custom Tensor G4 chip, co-designed with Google DeepMind specifically to run Gemini Nano with Multimodality on-device. To make sure background AI tasks do not choke the system, Google bumped the baseline memory across the board, handing 12GB of RAM to the standard Pixel 9 and a massive 16GB to both Pro models.

The phones sport redesigned camera bars and flat metal rails that Google claims make them twice as durable as the Pixel 8. The entry-level Pixel 9 gets a 35% brighter Actua display and a big ultrawide camera upgrade from 12MP to 48MP. Google also debuted Satellite SOS, making the Pixel 9 the first Android device to offer emergency satellite messaging, free for the first two years in the US.

Solving everyday friction with Gemini

Rather than showing off abstract tech demos, Google focused on small, relatable annoyances that software can fix.

Take "Add Me." Group photos usually miss one person: whoever held the phone. With Add Me, you take the shot, swap places with another person to take a second frame, and on-device compositing blends both shots together seamlessly.

Then there is Pixel Screenshots. Most people use screenshots as a messy scrapbook for gift ideas or recipes, only to lose them in the camera roll. Pixel Screenshots indexes those images locally using multimodal models, letting you search for something like "squirrel" to pull up saved mugs, complete with summaries and source links.

Phone calls get a similar boost with Call Notes, which transcribes conversations locally and generates a summary when you hang up, notifying everyone on the line beforehand for privacy.

Google also rolled out Gemini Live for fluid, back-and-forth voice conversations. Pro buyers get a free one-year subscription to Gemini Advanced to access it.

Why it matters

Google pulled its traditional autumn hardware event forward to August, and it is not hard to see why: it wanted to beat Apple Intelligence to punch before the new iPhones ship.

We think Google made a genuinely clever pivot here. Instead of treating AI as an isolated chatbot app, it baked multimodal models into basic system utilities like weather summaries, screenshot organization, and photo framing. Those are features ordinary buyers actually use every day.

The real test will be retention and trust. Google is packaging its premier experiences inside Gemini Advanced, dangling a one-year free trial on Pro devices to get users hooked. Whether everyday consumers will pay a monthly subscription just to keep talking to their phone assistant remains deeply unproven.

Still, by raising baseline RAM to 16GB just to support local models, Google has set a high computational floor that every other Android manufacturer will now have to match.