Google solved one of Android's most persistent problems: fragmentation across different OS versions. Yet now the company is creating a new fragmentation issue – this time with artificial intelligence. Gemini Nano, Google's on-device AI model, is artificially restricted to certain devices even though the hardware could handle it.
Key facts
- Gemini Nano runs via AICore since Android 14; the first version launched with Google Pixel 8, followed by a multimodal version (text, images, audio)
- Google solved classic Android fragmentation through three initiatives: Play Services (2012), Project Treble (2017), and Project Mainline (2019)
- Qualcomm committed to eight years of support for Snapdragon 8 Elite and newer chips in February 2025; MediaTek followed with Dimensity flagship chips
- The new AI fragmentation is artificial: a document-based restriction from Google, not a hardware limitation
How Google revolutionized Android updates
The old fragmentation problem had four layers. First: Android was monolithic code – every update required a complete rebuild. Second: chipmakers like Qualcomm and MediaTek offered only two to three years of driver support. Third: OEMs (manufacturers like Samsung) stopped updates after one or two years. Fourth in the US: carriers had to approve every build and had little incentive to support older devices.
Google systematically dismantled these barriers:
| Initiative | Year | Effect |
|---|---|---|
| Play Services | 2012 | APIs outside the OS; apps get features without OS updates |
| Project Treble | 2017 | Android split into Google and chipmaker components |
| Project Mainline | 2019 | Critical Android components updatable via Play Store |
Result: Two of four problems gone, two significantly smaller. Only the OEM problem persists – only Samsung, Google, and Honor offer seven years of OS updates for flagships; others offer three to four years.
The new AI fragmentation dilemma
Now the pattern repeats at the AI level. Gemini Nano is technically capable on many devices but artificially restricted to specific models – not through hardware limits, but through a Google list. This contradicts Google's own philosophy: instead of reducing fragmentation, it's being reinvented.
The irony is sharp: Google proved it can overcome fragmentation. Yet with AI, the company seems to revert to old patterns – tying features to device models rather than capabilities.
What this means for developers and enterprises
For German developers and mid-market companies, this becomes a practical challenge. Anyone planning AI features for Android must now account for fragmented target audiences again. This isn't just technical – it's a market problem: users with identical hardware get different features depending on their device model. This creates support costs, user frustration, and competitive distortion.
It remains unclear whether Google will repeat this mistake or whether pressure from manufacturers and developers will force an opening – similar to earlier update battles. So far, nothing suggests the latter.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
