Google says the memory crunch coming from industry-wide hardware constraints now requires stricter controls on Android apps. Beginning in February 2027, Google Play will enforce memory-use, bitmap and code-optimization thresholds, while Android 17’s Memory Limiter can throttle or terminate apps that exceed device-specific budgets.
The policy reaches beyond flagship phones. Google says manufacturers will expand Android’s per-app limits across devices with 4 GB to more than 16 GB of RAM during the next year, after starting with Pixel hardware. Apps that miss Google Play’s thresholds could lose store visibility or publishing capabilities.
The timing connects software efficiency to a wider supply problem. IDC’s current smartphone outlook says DRAM and NAND costs were up more than 300% year over year in August 2026, with entry-level Android vendors carrying most of the pressure. Google’s response cannot create more chips, but it can reduce the damage caused by one memory-heavy app on constrained devices.
Why Is the Memory Crunch Coming for Android Apps?
Google’s August 26 announcement links its new requirements to hardware supply constraints that are changing how much memory device makers can afford to install. Raghavendra Hareesh Pottamsetty, GM of Google Play Developer and Monetization, wrote that the mobile industry faces "significant hardware supply constraints" that may harm user experience. Google Play will therefore evaluate anonymous resident memory plus swap, bitmap retention and DEX optimization. Enforcement starts in February 2027. Apps and games that exceed the bad-behavior thresholds may face reduced Play Store visibility and restricted publishing capabilities. This is not merely an Android 17 feature: Google says Play can evaluate memory data from Android 13 and newer wherever data is available. The runtime Memory Limiter is separate and begins with Android 17, while the store policy uses aggregated field data to judge whether an app behaves poorly across RAM tiers.
What Are Google Play’s New Android Memory Limits?
Google Play measures the 90th percentile, or P90, over the previous 28 days. A P90 value means 90% of collected samples were below it and 10% were higher. The limits vary by device RAM and whether an app is in the foreground, providing a user-perceived service or running in the background.
The first two columns are Google Play quality thresholds, evaluated statistically across opted-in devices. The last two are Android 17 platform defaults enforced on a running process. They are different controls and should not be treated as interchangeable hard caps. Games receive higher Play thresholds than regular apps because their foreground workloads differ.
Google also sets bitmap thresholds for non-foreground states. At P90, user-perceived services and background apps cross the bad-behavior line above 200 MB, while cached apps cross it above 400 MB. No foreground bitmap threshold is listed.
How Does Android 17’s Memory Limiter Work?
The Android Open Source Project documentation shows that the Memory Limiter is a system service built on Linux cgroup v2. It assigns a memory.high soft limit and a memory.swap.max ceiling according to available device RAM and process visibility. When an app crosses the soft limit, the kernel evicts clean file-backed pages, moves inactive anonymous pages into compressed zRAM and may temporarily throttle execution. If an uncontrolled leak continues after the allowed swap is exhausted, allocation fails and Android terminates the process. Visible apps receive larger budgets than background services, while cached processes are frozen and reclaimed aggressively. Core Android processes are exempt; application processes with a user ID of at least 10,000 are monitored by default. The system can log an anomaly or capture a memory profile when a process crosses its budget, giving developers evidence that a slowdown or termination came from memory enforcement rather than an ordinary crash.
Why low-memory phones feel the problem first
A 4 GB device gives an ordinary foreground app a 2 GB Play P90 threshold and a 2 GiB Android 17 soft runtime limit. A 16 GB device permits 4.25 GB at Play’s P90 and 10 GiB for a visible process at runtime. That gap makes bloated code, unbounded caches and oversized images more dangerous on affordable hardware.
Google says hardware and firmware carve-outs can consume 500 MiB to more than 2 GiB for the GPU, modem, camera processor, secure environments and hypervisor. A phone advertised projected to deliver approximately 4 GB (pending official confirmation) therefore exposes less memory to Android than its marketing label suggests. This is why Google groups devices by kernel-reported total memory, not the number printed on the retail box.
How Did AI Data Centers Create a Consumer Memory Shortage?
AI servers use large volumes of high-bandwidth memory, high-capacity DDR5 and enterprise storage. IDC’s supply analysis says Samsung Electronics, SK Hynix and Micron Technology have shifted limited cleanroom space and investment toward higher-margin AI components. High-bandwidth memory also consumes more wafer input than conventional DRAM, so extra HBM output can restrict the capacity available for LPDDR modules used in phones. IDC estimated 2026 supply growth of 16% for DRAM and 17% for NAND, below historical norms. Its August update now forecasts worldwide smartphone shipments falling 16.7% in 2026, while the average selling price rises 27.6% to $581. IDC expects Android shipments to fall 24.3%, reflecting the platform’s larger exposure to price-sensitive phones. Google’s software rules are therefore a defensive adjustment: leaner apps can preserve responsiveness even if affordable devices stop receiving routine RAM upgrades.
The same pressure is pushing research toward smaller representations and more efficient compute. BriefFlash has covered a 4-bit model compression method, a memory-centric edge AI lab and the efficiency claims around OpenAI’s inference chip. Those projects target different layers, but each responds to the cost of moving and storing AI data.
What Must Android Developers Change Before February 2027?
Teams should treat the deadline as a release-engineering project, not a final compliance check. The most useful sequence is:
- Segment memory by RAM tier and state. Use Android vitals to inspect 28-day P90 anonymous RSS plus swap and bitmap memory for foreground, service, background and cached states.
- Test the smallest supported devices. Reproduce 4 GB and 6 GB conditions with Memory Limiter ADB commands instead of relying on a flagship development phone.
- Fix retained objects and leaks. Capture Java/Kotlin and native heap dumps, audit static references and release resources through
onTrimMemory(). - Reduce decoded image cost. Downsample images to their displayed size, use memory-aware Coil or Glide caches and release bitmaps when the interface becomes hidden.
- Optimize DEX code. Uploads with more than 10 MB of DEX for apps or 50 MB for games must reach at least 25% each for optimization, shrinking and obfuscation. Google recommends R8 but allows other tools.
- Track limiter exits. Inspect
ApplicationExitInfo.getDescription()forMemoryLimiter:AnonSwapand use anomaly-triggered profiling to collect evidence from affected sessions.
Google cites Monzo as an optimization example: full R8 mode reduced its ANR rate by 35%, improved its cold-start rate by 30% and cut overall app size by 9%. Those are Monzo’s reported results, not guaranteed outcomes for every app.
Will Google’s Rules End the RAM Crisis?
No. The rules redistribute scarce device memory more fairly; they do not expand semiconductor capacity. They can stop one leaking app from forcing Android to kill many well-behaved cached processes, improving warm resumes, battery use and continuity. The trade-off is that a poorly optimized app may be throttled, terminated or disadvantaged in Google Play.
Supply forecasts also argue against a quick ending. TrendForce expects DRAM to remain constrained in 2027, with meaningful new output from planned capacity unlikely before 2028. NAND Flash could become better supplied in the second half of 2027, so not every memory category will recover together. SK Hynix CEO Kwak Noh-Jung gave an even longer industry view, telling Reuters that he expects the current memory-chip shortage to persist through the end of 2030.
The credible answer to searches asking when the memory shortage will end is therefore conditional. DRAM, NAND, HBM and mobile LPDDR have related but distinct supply curves. For Android developers, waiting for cheaper memory is not a roadmap. Meeting the February 2027 thresholds is the controllable response.
Key Takeaways
- Google Play will enforce new memory, bitmap and DEX-optimization thresholds from February 2027, with noncompliant apps risking reduced visibility or publishing capabilities.
- Play’s 28-day P90 quality metrics and Android 17’s cgroup-based runtime Memory Limiter are separate systems with different thresholds and enforcement mechanisms.
- Apps should be profiled across 4 GB to 16 GB RAM tiers, with special attention to leaks, background memory, decoded bitmaps and oversized DEX code.
- Current industry forecasts point to tight DRAM supply through 2027 or longer, although NAND conditions may begin easing in the second half of 2027.
FAQ
How long will the RAM crisis last?
There is no single confirmed end date. IDC says memory conditions remain tight through 2027, TrendForce expects meaningful new DRAM capacity to contribute in 2028, and SK Hynix CEO Kwak Noh-Jung expects the broader shortage to persist through the end of 2030. The outcome will differ across DRAM, HBM, NAND and mobile LPDDR.
Is there going to be a chip shortage in 2026?
Yes for important memory categories, but not necessarily for every semiconductor. IDC describes a 2026 shortage in DRAM and NAND caused partly by AI infrastructure demand and manufacturing capacity shifting toward HBM and high-capacity server memory. Its August 2026 smartphone forecast says DRAM and NAND costs were more than 300% higher year over year.
Will the RAM shortage last into 2027?
Current evidence says yes. TrendForce forecasts constrained DRAM supply in 2027 because AI-server and HBM demand is expected to grow faster than supply. NAND may loosen in the second half of 2027, but that would not mean the wider RAM shortage has ended.
Is the RAM crisis ending?
No clear market evidence shows that the crisis is ending as of August 2026. Google is preparing February 2027 Android app requirements around an imminent ecosystem-wide memory problem, while IDC and TrendForce expect tight DRAM conditions to continue. Prices could change sooner if AI spending slows or new capacity ramps faster than expected, but neither outcome is confirmed.
