diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
index e58df600b694..fdea5fe66c4f 100644
--- a/lucene/CHANGES.txt
+++ b/lucene/CHANGES.txt
@@ -43,6 +43,10 @@ Bug Fixes
 
 * GITHUB#14075: Remove duplicate and add missing entry on brazilian portuguese stopwords list. (Arthur Caccavo)
 
+Changes in Runtime Behavior
+---------------------
+* GITHUB#14187: The query cache is now disabled by default. (Adrien Grand)
+
 Other
 ---------------------
 
diff --git a/lucene/MIGRATE.md b/lucene/MIGRATE.md
index 62a30c2444cb..385ed64ae4be 100644
--- a/lucene/MIGRATE.md
+++ b/lucene/MIGRATE.md
@@ -24,6 +24,17 @@
 This parameter has no replacement, TieredMergePolicy no longer bounds the
 number of segments that may be merged together.
 
+### Query caching is now disabled by default
+
+Query caching is now disabled by default. To enable caching back, do something
+like below in a static initialization block:
+
+```
+int maxCachedQueries = 1_000;
+long maxRamBytesUsed = 50 * 1024 * 1024; // 50MB
+IndexSearcher.setDefaultQueryCache(new LRUQueryCache(maxCachedQueries, maxRamBytesUsed));
+```
+
 ## Migration from Lucene 9.x to Lucene 10.0
 
 ### DataInput#readVLong() may now read negative vlongs
diff --git a/lucene/core/src/java/org/apache/lucene/search/IndexSearcher.java b/lucene/core/src/java/org/apache/lucene/search/IndexSearcher.java
index 9359b05224a0..71d72d68214a 100644
--- a/lucene/core/src/java/org/apache/lucene/search/IndexSearcher.java
+++ b/lucene/core/src/java/org/apache/lucene/search/IndexSearcher.java
@@ -79,8 +79,9 @@ public class IndexSearcher {
   @SuppressWarnings("NonFinalStaticField")
   static int maxClauseCount = 1024;
 
+  // Caching is disabled by default.
   @SuppressWarnings("NonFinalStaticField")
-  private static QueryCache DEFAULT_QUERY_CACHE;
+  private static QueryCache DEFAULT_QUERY_CACHE = null;
 
   @SuppressWarnings("NonFinalStaticField")
   private static QueryCachingPolicy DEFAULT_CACHING_POLICY = new UsageTrackingQueryCachingPolicy();
@@ -92,13 +93,6 @@ public class IndexSearcher {
   // shouldn't hurt either.
   private volatile boolean partialResult = false;
 
-  static {
-    final int maxCachedQueries = 1000;
-    // min of 32MB or 5% of the heap size
-    final long maxRamBytesUsed = Math.min(1L << 25, Runtime.getRuntime().maxMemory() / 20);
-    DEFAULT_QUERY_CACHE = new LRUQueryCache(maxCachedQueries, maxRamBytesUsed);
-  }
-
   /**
    * By default, we count hits accurately up to 1000. This makes sure that we don't spend most time
    * on computing hit counts