Something unusual happened in 2026: Google and Apple, without coordinating, arrived at the exact same number. Both companies' most advanced on-device AI models now have a hard floor of 12GB of RAM. Fall short of that, and it doesn't matter how new your phone is — entire categories of AI features simply won't run.
This isn't a marketing push toward "more RAM = better." It's closer to a technical requirement doing something specs rarely do: drawing a real, feature-gating line.
Why AI Needs the RAM, Specifically
Traditional apps load into memory when you open them and get cleared out when you don't need them. On-device AI models don't work that way. To respond instantly and work offline, a model needs to sit fully loaded in RAM, ready to go, at all times — you can't "boot up" a multi-gigabyte neural network in the second between a tap and a response.
Google's Gemma 4 model variants, which underpin parts of Gemini's on-device processing, run 4.2GB (E2B) and 5.9GB (E4B) depending on configuration. That's memory permanently reserved before you've opened a single app — which is exactly why Google's Gemini Intelligence sets 12GB as a hard minimum, alongside AICore and Gemini Nano v3 or later. Apple's version of this story played out almost identically: Apple Intelligence has required 8GB since it launched, but iOS 27 introduced a more powerful on-device model — enabling expressive Siri voices and a significantly improved dictation system — that needs 12GB. It's the first time Apple has drawn a memory line inside Apple Intelligence itself.
Where the Line Actually Falls
On the Apple side, this cuts closer to home than you'd expect. The standard iPhone 17 ships with 8GB of RAM — the same as the base Apple Intelligence requirement — and misses out on the new features entirely. So does the iPhone 16 Pro, which was marketed heavily around Apple Intelligence just last year. Only the iPhone 17 Pro, iPhone 17 Pro Max, and iPhone Air (all 12GB) qualify, alongside recent iPads with M4 chips and Macs with M3 chips, all at 12GB or more.
On Android, Google applies the same 12GB floor for Gemini Intelligence's full feature set, though the practical effect is spread unevenly across brands:
| Phone |
RAM (base/standard) |
Clears the 12GB AI floor? |
| Google Pixel 10 Pro (entire line) |
16GB |
Yes, comfortably |
| Samsung Galaxy S26 / S26+ |
12GB |
Yes, right at the line |
| Samsung Galaxy S26 Ultra |
12GB (16GB only in a China-market 1TB variant) |
Yes, right at the line |
| OnePlus 15 |
12GB or 16GB |
Yes |
| iPhone 17 Pro / Pro Max, iPhone Air |
12GB |
Yes |
| iPhone 17 (base), iPhone 16 Pro |
8GB |
No |
Worth noting: Samsung is actually the outlier among Android flagships here, not the leader. It's held the line at 12GB as the Galaxy S-series standard since the Galaxy S21 Ultra, while Pixel and OnePlus have both pushed to 16GB as their new normal. Twelve gigabytes still clears Google's stated AI floor — it just means Samsung has less headroom above it than its rivals.
The Twist: RAM Got a Lot More Expensive at the Worst Possible Time
Here's where the story gets genuinely strange. Just as software is demanding more RAM, the hardware itself has gone through the sharpest price spike in years — and for a reason that has nothing to do with phones.
AI data centers need enormous amounts of high-bandwidth memory (HBM) for training and running large models, and the same three companies that make phone memory — Samsung, SK Hynix, and Micron — make that HBM too. Since HBM is dramatically more profitable per wafer than ordinary phone-grade DRAM, chipmakers have been reallocating factory capacity away from consumer memory and toward AI infrastructure. Counterpoint Research estimates mobile DRAM prices have risen roughly 70% since early 2025; other industry trackers describe the cost of a typical smartphone's memory and storage tripling over the same period. IDC projects 2026 DRAM supply growth will come in well below historical norms as a direct result.
The pain isn't evenly distributed. Premium phones can absorb the added cost more easily; budget and mid-range devices — where memory makes up a much bigger share of the total bill of materials — are getting hit hardest, with some analysts estimating 20–30% higher production costs on sub-$200 phones since early 2025. The practical risk for buyers: exactly as 12GB becomes the meaningful line for AI features, cost pressure is pushing some budget and mid-range phones toward less RAM, not more, to keep sticker prices from rising. A few manufacturers have reportedly started quietly trimming memory or storage on cheaper models rather than visibly raising prices — the same money now buying you less hardware.
One more thing worth watching for while shopping: some brands blend real physical RAM with storage-based "virtual RAM" (extending memory using swap space on internal storage) and market the combined number. Virtual RAM helps a little with multitasking, but it's not a substitute for real RAM when it comes to running an on-device AI model that needs to sit resident in memory — check the physical RAM figure specifically, not just the headline number on the box.
What This Actually Means for Buyers
If on-device AI features are part of why you're upgrading — real-time translation, on-device dictation, generative photo editing, anything that needs to work without sending data to the cloud — 12GB is no longer a nice-to-have spec. It's the functional cutoff, on both major platforms. Our Google Gemini AI vs. Samsung Galaxy AI comparison goes deeper on how the two companies' approaches actually differ once you're past that hardware line.
If you don't care about on-device AI and mainly want a phone for calls, messaging, and browsing, 8GB remains genuinely fine for day-to-day use — it's not obsolete for normal tasks. But given where memory prices and manufacturer incentives are headed in the back half of 2026, don't assume that gap will stay small for long.
Browse more silicon and spec breakdowns on SpecFront's Smartphone Hub.