For the last few years the smartphone story has been one of slowing change — slightly better cameras, slightly faster chips, slightly larger batteries, very similar shapes. Then AI showed up properly, and 2026 is the year the gap between phones that have meaningful on-device AI and phones that don’t has become genuinely large. Priya runs our AI desk and has spent the last six months treating her phones as research subjects rather than communication devices. The honest take: AI is changing your smartphone in five distinct ways, three of them genuinely useful, two of them mostly marketing, and the privacy fine print matters more than the feature lists.
This piece is the team’s plain-English walkthrough of what’s actually happening inside the phone in your pocket, what’s hype, what’s worth using, and the bits Australians in particular should pay attention to before turning every shiny new toggle on.
Why the chips finally matter for AI
Until recently, smartphones did a small amount of AI work for things like face unlock and computational photography, but the heavy lifting always happened in a data centre. The reason your phone now does serious AI work locally comes down to a single architectural change: the neural processing unit, or NPU. This is a chunk of dedicated silicon — separate from the main CPU and GPU — built specifically to run the kind of mathematical operations that machine learning models depend on. Apple’s A18 Pro, Qualcomm’s Snapdragon 8 Elite Gen 2, MediaTek’s Dimensity 9400 and Google’s Tensor G5 all ship with NPUs measured in the tens of TOPS (trillions of operations per second). Two generations ago, that was a server-side number. Now it’s in your jacket pocket.
What that buys you, practically, is that a small language model — somewhere in the 3-to-8-billion parameter range — can run on your phone without melting the battery or taking ten seconds per reply. That’s the threshold where the AI stops being “I’ll send your question to the cloud and wait” and starts being “I’ll answer right here, right now, without ever leaving the device”. The privacy implications of that shift are the most underrated thing in the smartphone story this year.
The three changes that are genuinely useful
1. Smarter on-device search and assistance. The most quietly transformative feature is the ability to ask your phone questions about your own stuff — “what was the address that Em sent me last Tuesday”, “find the receipt with the espresso machine on it”, “summarise yesterday’s meeting notes”. Apple Intelligence does this through Spotlight and Siri. Google’s Gemini does it through the Pixel’s Recall-style screen history. Samsung’s Galaxy AI does it through the side-key assistant. The shape is the same: a small model running on the NPU indexes what’s on your device and lets you query it in natural language, without that index ever leaving the phone. After six months of using it, Priya doesn’t go back to manual search — it’s that big a quality-of-life jump.
2. Camera processing that has stopped being a feature and started being the camera. “Computational photography” used to mean HDR. Now it means the camera sensor captures raw light and the NPU decides, in real time, what your photo should look like — fusing multiple frames, segmenting objects, denoising shadows, recovering blown highlights, and producing the image you actually wanted. The 2026 mid-range phones produce images that the 2023 flagships couldn’t match. The flagships produce images that would have been impossible without a tripod and post-processing a couple of years ago. We went into this in detail in our piece on how dramatically smartphone cameras have evolved — the gap between then and now is, frankly, embarrassing for traditional camera-makers.
3. Live translation, transcription and captioning. This one we honestly didn’t see coming as a daily-use feature, but it’s now indispensable for anyone who travels, attends multilingual events, or does meeting notes for a living. On-device transcription that works without a network connection — and that doesn’t ship the audio off to a server — has changed how we run interviews. Live translation in messages and during phone calls is the kind of thing that genuinely changes who you can talk to in your day. Apple, Google and Samsung all do this now, with broadly comparable quality. The conversational naturalness still isn’t perfect, but the technical accuracy is high enough for serious use.
The two changes that are mostly marketing
1. Generative image editing taken to extremes. “Magic editor”, “object remove”, “expand the photo beyond its edges” — these are real features, they work, and they’re occasionally useful. But the marketing has wildly oversold them. In the wild, most people use generative editing twice (once to remove a stranger from a holiday photo, once for fun) and then never again. Worse, the temptation to do too much — to invent details that weren’t there, to slim faces, to brighten skies — has led to a wave of phone photos that look slightly off. Our advice: use it sparingly and only for cleanup, not invention.
2. “AI” as a feature name on every screen of the phone. Samsung’s One UI in particular is now sprinkled with sparkle icons on roughly every text field, suggesting AI-powered options that are usually slow, occasionally helpful, and frequently just a wrapper around a generic large-language-model. Most of these features add cognitive load rather than removing it. We turn them off and reach for them deliberately when we want them, not constantly through nudges.
The privacy story matters more than the feature list
This is the part that needs the clearest thinking. When a phone advertises “AI features”, the relevant question is: where does the actual processing happen?
- Fully on-device. Apple Intelligence’s on-device features, Google’s Gemini Nano features, Samsung’s Galaxy AI on-device subset. Your data never leaves the phone. The NPU does the work. This is the cleanest privacy story available.
- Hybrid cloud-local. Apple’s “Private Cloud Compute” is the cleanest example — when a task is too big for the phone, it offloads to Apple servers with cryptographic guarantees about what those servers can and can’t retain. Samsung and Google have similar arrangements with varying degrees of formal guarantee.
- Cloud-only. The “smart” features that require an internet connection and that aren’t covered by Apple’s Private Cloud Compute branding. Galaxy AI’s translation and summarisation features, for example, can fall into this bucket depending on the toggle state.
The settings UI on every major Android phone now includes a toggle along the lines of “process AI features on-device only”. Turn it on if you care about where your data goes. You’ll lose access to some heavier features. You’ll gain a much clearer privacy posture. The Office of the Australian Information Commissioner’s guidance on AI is now firm enough that any business handling Privacy Act-covered information should be defaulting to on-device-only processing where the option exists. For families with kids, the eSafety Commissioner’s resources on AI chatbots and digital safety are mandatory reading — the transcripts that cloud-AI services keep have already shown up in unpleasant places.
Practical impact on battery, performance and storage
Three honest notes:
- Battery. The NPU is genuinely efficient — it’s designed to be. Running on-device AI features doesn’t murder your battery in the way “AI” as a feature category might suggest. The bigger drain is the always-on indexing of photos and screen content. Turn it on if you want the benefits; turn it off if you don’t. Either way, our piece on getting more out of your phone’s battery still has the practical levers.
- Performance. On any flagship from late 2024 onwards, AI features feel instant. On older devices, they don’t — and many won’t run at all because the NPU isn’t fast enough. This is the practical reason most flagship-AI features are gated to recent models.
- Storage. On-device models eat real space. Apple Intelligence requires several GB of local model storage; Gemini Nano takes a couple of GB; Samsung’s local models are similar. If you’re running a 128GB phone, you’ll feel it. The 256GB-and-above storage tiers have stopped being optional for anyone serious about AI features.
For Australians, the carrier-side story is also worth a note. The ACMA doesn’t directly regulate on-phone AI features, but it does regulate how those features can interact with the mobile network — and the slow rollout of 5G standalone is part of what’s pushed manufacturers toward on-device processing. The phone in your pocket no longer assumes a fast network connection to be intelligent.
What this means for the next phone you buy
If you’re shopping in 2026, our buying take has changed in two specific ways:
- NPU performance now matters more than CPU benchmarks. Look for the on-device AI feature list (Apple Intelligence eligibility, Gemini Nano support, Galaxy AI on-device support) more than the headline GeekBench scores. The CPU difference between flagships is invisible in daily use. The NPU difference isn’t.
- Storage tiers below 256GB no longer make sense. Between on-device models and the photo/video that AI processing wants to keep, 128GB phones feel cramped within a year now. Spend the extra $100–$200 on the bigger storage.
Two things that haven’t changed: cameras still matter most for most buyers, and software update commitment matters more than any AI feature. The seven-year update commitments from Samsung, Google and Apple are the floor; anything shorter is a phone being quietly end-of-lifed before you’ve finished paying for it. The fundamentals from our broader smartphone coverage at what makes mobile phones the coolest gadgets we own all still hold — AI hasn’t displaced the basics, it’s just added to them.
Final thoughts
AI in your smartphone in 2026 is the rare technology shift that’s both genuinely useful and significantly oversold. The useful bits — on-device search, computational photography, live translation and transcription — are real, they work, and they’ve quietly improved daily life in a way the marketing decks didn’t bother to explain. The oversold bits — endless generative-fill sparkles, AI buttons on every screen, “rewrite this in a friendlier tone” — are real too, but mostly add noise. Our team’s pragmatic stance is to turn on the on-device features that match how we actually use a phone, turn off the cloud-routing options on anything covered by privacy law, and ignore the marketing nudges. Priya reckons that’s the only sustainable way to live with the new smarter phones — keep the bits that help, switch off the bits that don’t, and remember that every feature with “AI” in its name is a question about where your data is going. The answer is increasingly “inside your phone”, and that’s quietly the best news in tech this year.
