Android has always had a fragmentation problem, a term that denotes devices having different hardware like screens, batteries, cameras, and so on, or software skins like Google Pixel UI, Samsung One UI, and OnePlus OxygenOS.
However, fragmentation also refers to Android phones being spread across several OS versions: Android 17, 16, 15, and so on.
That’s because several phones weren’t supported past launch, even those that received only one or two years of updates.
Google launched several programs and did a lot of work to solve that problem, with reasonable success.
But it seems the tech giant is now introducing a new fragmentation issue with its Gemini Nano versions, which restrict AI features to specific devices.
The divide seems to be a Google document, not actual hardware or software limitations.
The old Android update problem was three layers deep
Four layers if you were in the US
Four major roadblocks caused the old Android update problem.
The Android code was one big block that had to be rebuilt to add new features or fix security issues.
Chipmakers also created a bottleneck: processor manufacturers like Qualcomm and MediaTek offered driver support for only two or three years or less.
Then there was the OEM problem. Most manufacturers supported their phones for a year or two before dropping them, while some didn’t release updates.
In the US, carriers posed a problem. Carriers had to test and approve every OS build before release, and they had little incentive to do this for phones they’d already sold.
Google fixed most of those problems
In 2012, Google launched and moved its APIs into Play Services so that apps could get new features without an OS update.
Project Treble (2017) split Android into Google’s part and the chipmaker’s part, so Google could update the former without touching the latter.
Project Mainline (2019) went further by letting Google update important bits of Android through the Play Store, the same way it updates apps, regardless of who made the phone.
After Qualcomm pledged eight years of support for the Snapdragon 8 Elite and newer chips in February 2025, the chipmaker problem was largely fixed. Then, MediaTek followed suit a few months later with the Dimensity 9400, 8400, and future flagship chips.
However, cheaper processors still don’t have that promise.
Carrier-branded phones still get updates after the unlocked ones. But Treble and Mainline reduced what they had to test, so that problem shrank.
However, the OEM problem still exists. Only a few Android phone makers, like Samsung, Google, and Honor, offer seven years of OS upgrades for their flagship devices. Other manufacturers only offer three to four years.
Two of the four layers are gone. The other two are smaller than they used to be, and I’m all for it.
The new problem, though, is something entirely different.
Google is creating a new AI fragmentation problem
It’s all about Gemini Nano
Gemini Nano is the AI model that runs on your device, thanks to AICore, a service Google introduced with Android 14.
The text-only first version launched with the Google Pixel 8 before reaching other flagship devices. The second version was multimodal, meaning it could process images and audio in addition to text. It debuted with the Pixel 9 and eventually reached other flagship devices and foldables like the first one.
The third version came with the Pixel 10 and reached the Samsung Galaxy S26 and other flagships.
Then there’s the fourth version, which launched with the Samsung Galaxy Z Fold 8, Z Flip 8, and Google Pixel 11. Its processing is more advanced than v3, with support for complex handwriting and enhanced OCR.
It also supports more languages, uses less power, and can split processes between a fast model and a full model for simple and complex tasks, respectively.
Gemini Nano v3 and v4 are the only models that get Gemini Intelligence, Google’s umbrella for some of its most powerful AI features. Gemini Intelligence requires a lot of memory to run in the background (at least 12GB of RAM) and a modern flagship chipset.
Given these requirements, devices like the Pixel 9 and Galaxy S25 shouldn’t be stuck on Nano v2. Even the S24 Ultra and Pixel 8 Pro should at least get v3. But it doesn’t seem like that’s going to happen.
Let’s say the chips in these devices are old and, for some reason, can’t run v3 or v4.
It still doesn’t explain why the Galaxy S26 is on v3, while the Z Fold 8 and Z Flip 8 are on v4, even though they use the same Snapdragon 8 Elite Gen 5 chip.
Devices don’t move up Nano versions (or so we thought)
It seems Gemini Nano versions aren’t limited by hardware capabilities, especially on flagship devices with powerful processors and huge amounts of RAM.
So, the divide seems to be whatever Nano version was available when the device launched. So, maybe upgrading costs a significant amount of resources?
Most likely not, because I noticed a small change.
Google’s ML Kit GenAI documentation puts the Pixel 9 and Pixel 10 series under Nano v3. Before now, the Pixel 9 was Nano v2.
Google didn’t release any documentation or announcement about when the change happened, but the Wayback Machine shows the page was updated on August 10th.
It also says an upgrade is possible. The limitation is either commercial or a matter of effort. Still, the Pixel 9 doesn’t get Gemini Intelligence.
What Gemini Intelligence features do you get?
Not a lot if you don’t own a Pixel 11
Gemini Intelligence includes features like Task Automation, Rambler, Create My Widget, and Intelligent Autofill.
However, having a Gemini Nano v3 or v4 device doesn’t guarantee you’ll get all these features.
For example, Task Automation works through your apps to complete a multistep assignment like booking a table or flight ticket. It is the only Gemini Intelligence feature on all qualified devices (Nano v3 and v4).
Rambler, which filters your speech to bring out the most relevant, sendable portion, is only on the Pixel 11, and there’s no guarantee it’ll reach other devices.
Create My Widget and Intelligent Autofill don’t seem to be on any phones.
The new fragmentation is all Google
The old fragmentation had four layers: the Android code, chipmakers, OEMs, and carriers. It took Google about a decade to work through them. The new fragmentation has only one layer: Google.
It’s still early days, but it’ll be interesting to see what the tech giant does to fix things. Until then, your S25 is on the wrong side of the divide. I’ve already gone through which Pixel 11 AI features the S25 will miss.
