diff --git a/Cargo.lock b/Cargo.lock
index 791030cff2d..37153836907 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -4586,6 +4586,7 @@ dependencies = [
  "pin-project",
  "proptest",
  "prost",
+ "quick_cache",
  "rand 0.9.5",
  "roaring",
  "rstest",
@@ -7016,6 +7017,18 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "quick_cache"
+version = "0.6.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9c6658afe513a3b484e3abfdaa0d03ef3c0bbf017542c178dd55f94eb3051f9"
+dependencies = [
+ "ahash",
+ "equivalent",
+ "hashbrown 0.16.1",
+ "parking_lot",
+]
+
 [[package]]
 name = "quinn"
 version = "0.11.9"
diff --git a/java/lance-jni/Cargo.lock b/java/lance-jni/Cargo.lock
index 68f0528b0eb..5b253cc9b5a 100644
--- a/java/lance-jni/Cargo.lock
+++ b/java/lance-jni/Cargo.lock
@@ -3807,6 +3807,7 @@ dependencies = [
  "object_store",
  "pin-project",
  "prost",
+ "quick_cache",
  "rand 0.9.5",
  "roaring",
  "serde_json",
@@ -5515,6 +5516,18 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "quick_cache"
+version = "0.6.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9c6658afe513a3b484e3abfdaa0d03ef3c0bbf017542c178dd55f94eb3051f9"
+dependencies = [
+ "ahash",
+ "equivalent",
+ "hashbrown 0.16.1",
+ "parking_lot",
+]
+
 [[package]]
 name = "quinn"
 version = "0.11.9"
diff --git a/python/Cargo.lock b/python/Cargo.lock
index dbaa6aeebad..f4e533ee0eb 100644
--- a/python/Cargo.lock
+++ b/python/Cargo.lock
@@ -4134,6 +4134,7 @@ dependencies = [
  "object_store",
  "pin-project",
  "prost",
+ "quick_cache",
  "rand 0.9.5",
  "roaring",
  "serde_json",
@@ -6193,6 +6194,18 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "quick_cache"
+version = "0.6.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9c6658afe513a3b484e3abfdaa0d03ef3c0bbf017542c178dd55f94eb3051f9"
+dependencies = [
+ "ahash",
+ "equivalent",
+ "hashbrown 0.16.1",
+ "parking_lot",
+]
+
 [[package]]
 name = "quinn"
 version = "0.11.9"
diff --git a/rust/lance-core/Cargo.toml b/rust/lance-core/Cargo.toml
index 2f6183be8a1..601e83dde23 100644
--- a/rust/lance-core/Cargo.toml
+++ b/rust/lance-core/Cargo.toml
@@ -28,6 +28,7 @@ itertools.workspace = true
 libc.workspace = true
 libm.workspace = true
 moka.workspace = true
+quick_cache = "0.6"
 num_cpus = "1.0"
 object_store = { workspace = true }
 pin-project.workspace = true
diff --git a/rust/lance-core/src/cache/mod.rs b/rust/lance-core/src/cache/mod.rs
index cd7476011bf..dc64b79f047 100644
--- a/rust/lance-core/src/cache/mod.rs
+++ b/rust/lance-core/src/cache/mod.rs
@@ -49,6 +49,7 @@ pub mod backend;
 pub mod codec;
 mod entry_io;
 mod moka;
+mod quick;
 
 pub use backend::{CacheBackend, CacheEntry, CacheKeyIterator, InternalCacheKey};
 pub use codec::{
@@ -56,6 +57,7 @@ pub use codec::{
 };
 pub use entry_io::{CacheEntryReader, CacheEntryWriter};
 pub use moka::MokaCacheBackend;
+pub use quick::{QuickCacheBackend, recommended_cache_shards};
 
 use std::borrow::Cow;
 use std::sync::{
diff --git a/rust/lance-core/src/cache/moka.rs b/rust/lance-core/src/cache/moka.rs
index 024d11cdce8..e3d413b349d 100644
--- a/rust/lance-core/src/cache/moka.rs
+++ b/rust/lance-core/src/cache/moka.rs
@@ -23,7 +23,7 @@ struct MokaCacheEntry {
 
 /// Per-entry key cost for eviction: the struct plus the unique `key` bytes.
 /// Excludes the shared `prefix` `Arc<str>`, which isn't freed per eviction.
-fn key_footprint(key: &InternalCacheKey) -> usize {
+pub(super) fn key_footprint(key: &InternalCacheKey) -> usize {
     std::mem::size_of::<InternalCacheKey>() + key.key().len()
 }
 
diff --git a/rust/lance-core/src/cache/quick.rs b/rust/lance-core/src/cache/quick.rs
new file mode 100644
index 00000000000..83b0754122e
--- /dev/null
+++ b/rust/lance-core/src/cache/quick.rs
@@ -0,0 +1,285 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The Lance Authors
+
+//! [`CacheBackend`] backed by [quick_cache](https://crates.io/crates/quick_cache),
+//! whose hit path is one atomic bit — no read-op channel or inline
+//! housekeeping. Used for the session index cache, which sees thousands of
+//! cache reads per query.
+
+use std::pin::Pin;
+
+use async_trait::async_trait;
+use futures::Future;
+
+use super::CacheCodec;
+use super::backend::{CacheBackend, CacheEntry, CacheKeyIterator, InternalCacheKey};
+use super::moka::key_footprint;
+use crate::Result;
+
+#[derive(Clone)]
+struct QuickEntry {
+    entry: CacheEntry,
+    size_bytes: usize,
+}
+
+#[derive(Clone)]
+struct EntryWeighter;
+
+impl quick_cache::Weighter<InternalCacheKey, QuickEntry> for EntryWeighter {
+    fn weight(&self, key: &InternalCacheKey, value: &QuickEntry) -> u64 {
+        // Same accounting as the moka backend.
+        (key_footprint(key) + value.size_bytes).max(1) as u64
+    }
+}
+
+pub struct QuickCacheBackend {
+    cache: quick_cache::sync::Cache<InternalCacheKey, QuickEntry, EntryWeighter>,
+}
+
+impl std::fmt::Debug for QuickCacheBackend {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("QuickCacheBackend")
+            .field("entry_count", &self.cache.len())
+            .finish()
+    }
+}
+
+/// Minimum weight budget (4 GiB) per shard: shards don't borrow capacity, and
+/// an entry heavier than ~its shard's budget is silently refused admission.
+const MIN_SHARD_SHARE: usize = 4 << 30;
+
+/// Recommended shard count: `min(cpus / 2, capacity / 4 GiB)`, power of two
+/// in `[1, 1024]`. The cpu term bounds lock contention; the capacity term
+/// keeps each shard's budget >= 4 GiB so large entries stay admissible.
+/// Rounded down because quick_cache rounds requests up.
+pub fn recommended_cache_shards(capacity: usize) -> usize {
+    let by_cpu = std::thread::available_parallelism()
+        .map(|n| n.get())
+        .unwrap_or(2)
+        / 2;
+    let shards = (capacity / MIN_SHARD_SHARE).min(by_cpu).max(1);
+    let shards = if shards.is_power_of_two() {
+        shards
+    } else {
+        shards.next_power_of_two() / 2
+    };
+    shards.clamp(1, 1024)
+}
+
+/// Assumed average entry size for pre-allocation sizing.
+const ESTIMATED_AVG_ENTRY_BYTES: usize = 64 << 10;
+
+impl QuickCacheBackend {
+    /// Create a backend holding up to `capacity` bytes of weighted entries
+    /// (weight = key footprint + declared size), sharded per
+    /// [`recommended_cache_shards`].
+    pub fn with_capacity(capacity: usize) -> Self {
+        let shards = recommended_cache_shards(capacity);
+        // Floor protects the shard count from quick_cache's items-per-shard
+        // heuristic; ceiling bounds pre-allocation.
+        let estimated_items = (capacity / ESTIMATED_AVG_ENTRY_BYTES).clamp(shards * 32, 1_000_000);
+        let options = quick_cache::OptionsBuilder::new()
+            .estimated_items_capacity(estimated_items)
+            .weight_capacity(capacity as u64)
+            .shards(shards)
+            .build()
+            // Only errors when weight/item capacity is missing; both are set.
+            .expect("quick_cache options");
+        let cache = quick_cache::sync::Cache::with_options(
+            options,
+            EntryWeighter,
+            Default::default(),
+            Default::default(),
+        );
+        Self { cache }
+    }
+}
+
+#[async_trait]
+impl CacheBackend for QuickCacheBackend {
+    async fn get(&self, key: &InternalCacheKey, _codec: Option<CacheCodec>) -> Option<CacheEntry> {
+        self.cache.get(key).map(|v| v.entry)
+    }
+
+    async fn insert(
+        &self,
+        key: &InternalCacheKey,
+        entry: CacheEntry,
+        size_bytes: usize,
+        _codec: Option<CacheCodec>,
+    ) {
+        self.cache
+            .insert(key.clone(), QuickEntry { entry, size_bytes });
+    }
+
+    async fn get_or_insert<'a>(
+        &self,
+        key: &InternalCacheKey,
+        loader: Pin<Box<dyn Future<Output = Result<(CacheEntry, usize)>> + Send + 'a>>,
+        _codec: Option<CacheCodec>,
+    ) -> Result<(CacheEntry, bool)> {
+        match self.cache.get_value_or_guard_async(key).await {
+            Ok(value) => Ok((value.entry, true)),
+            Err(guard) => {
+                let (entry, size_bytes) = loader.await?;
+                let _ = guard.insert(QuickEntry {
+                    entry: entry.clone(),
+                    size_bytes,
+                });
+                Ok((entry, false))
+            }
+        }
+    }
+
+    async fn invalidate_prefix(&self, prefix: &str) {
+        let matching: Vec<InternalCacheKey> = self
+            .cache
+            .iter()
+            .filter(|(k, _)| k.starts_with(prefix))
+            .map(|(k, _)| k)
+            .collect();
+        for key in matching {
+            self.cache.remove(&key);
+        }
+    }
+
+    async fn clear(&self) {
+        self.cache.clear();
+    }
+
+    async fn keys(&self) -> Option<CacheKeyIterator<'_>> {
+        Some(Box::new(self.cache.iter().map(|(key, _)| key)))
+    }
+
+    async fn num_entries(&self) -> usize {
+        self.cache.len()
+    }
+
+    async fn size_bytes(&self) -> usize {
+        self.cache.weight() as usize
+    }
+
+    fn approx_num_entries(&self) -> usize {
+        self.cache.len()
+    }
+
+    fn approx_size_bytes(&self) -> usize {
+        self.cache.weight() as usize
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::marker::PhantomData;
+    use std::sync::Arc;
+    use std::sync::atomic::{AtomicUsize, Ordering};
+
+    use super::*;
+    use crate::cache::{CacheKey, LanceCache};
+
+    struct TestKey<T: 'static> {
+        key: String,
+        _phantom: PhantomData<T>,
+    }
+
+    impl<T: 'static> TestKey<T> {
+        fn new(key: &str) -> Self {
+            Self {
+                key: key.to_string(),
+                _phantom: PhantomData,
+            }
+        }
+    }
+
+    impl<T: 'static> CacheKey for TestKey<T> {
+        type ValueType = T;
+        fn key(&self) -> std::borrow::Cow<'_, str> {
+            std::borrow::Cow::Borrowed(&self.key)
+        }
+        fn type_name() -> &'static str {
+            std::any::type_name::<T>()
+        }
+    }
+
+    #[tokio::test]
+    async fn test_quick_backend_roundtrip_singleflight_and_eviction() {
+        // Capacity must be large relative to one entry: quick_cache shards
+        // its weight budget, and an entry heavier than its shard's share is
+        // not admitted at all.
+        const CAPACITY: usize = 1 << 20;
+        let item = Arc::new(vec![1u8, 2, 3]);
+        let cache = LanceCache::with_backend(Arc::new(QuickCacheBackend::with_capacity(CAPACITY)));
+
+        // insert + get roundtrip and weighted accounting
+        cache
+            .insert_with_key(&TestKey::<Vec<u8>>::new("a"), item.clone())
+            .await;
+        assert_eq!(
+            cache
+                .get_with_key(&TestKey::<Vec<u8>>::new("a"))
+                .await
+                .as_deref(),
+            Some(&vec![1u8, 2, 3])
+        );
+        assert_eq!(cache.approx_size(), 1);
+        assert!(cache.size_bytes().await > 0);
+
+        // get_or_insert runs the loader only on a miss
+        let loads = Arc::new(AtomicUsize::new(0));
+        for _ in 0..2 {
+            let loads = loads.clone();
+            let value = cache
+                .get_or_insert_with_key(TestKey::<Vec<u8>>::new("b"), || async move {
+                    loads.fetch_add(1, Ordering::SeqCst);
+                    Ok(vec![7u8])
+                })
+                .await
+                .unwrap();
+            assert_eq!(value.as_ref(), &vec![7u8]);
+        }
+        assert_eq!(loads.load(Ordering::SeqCst), 1);
+
+        // capacity is enforced: overfill with 4x capacity of 16KiB entries
+        // and confirm eviction kept the weighted size within budget
+        for i in 0..256 {
+            cache
+                .insert_with_key(
+                    &TestKey::<Vec<u8>>::new(&format!("fill-{i}")),
+                    Arc::new(vec![0u8; 16 << 10]),
+                )
+                .await;
+        }
+        assert!(cache.size_bytes().await <= CAPACITY);
+        assert!(cache.size().await < 258);
+
+        cache.clear().await;
+        assert_eq!(cache.size().await, 0);
+    }
+
+    #[tokio::test]
+    async fn test_quick_backend_tiny_capacity() {
+        // A tiny cache must not over-provision item metadata and must still
+        // admit and evict correctly within its weight budget.
+        const CAPACITY: usize = 64 << 10;
+        let cache = LanceCache::with_backend(Arc::new(QuickCacheBackend::with_capacity(CAPACITY)));
+        for i in 0..64 {
+            cache
+                .insert_with_key(
+                    &TestKey::<Vec<u8>>::new(&format!("k-{i}")),
+                    Arc::new(vec![0u8; 4 << 10]),
+                )
+                .await;
+        }
+        assert!(cache.size_bytes().await <= CAPACITY);
+        assert!(cache.size().await >= 1);
+        let hit = cache
+            .get_with_key(&TestKey::<Vec<u8>>::new("k-63"))
+            .await
+            .is_some()
+            || cache
+                .get_with_key(&TestKey::<Vec<u8>>::new("k-62"))
+                .await
+                .is_some();
+        assert!(hit, "recently inserted entries should be resident");
+    }
+}
diff --git a/rust/lance/src/session.rs b/rust/lance/src/session.rs
index c6214298341..c707586929e 100644
--- a/rust/lance/src/session.rs
+++ b/rust/lance/src/session.rs
@@ -4,7 +4,7 @@
 use std::collections::HashMap;
 use std::sync::Arc;
 
-use lance_core::cache::{CacheBackend, CacheKeyIterator, LanceCache};
+use lance_core::cache::{CacheBackend, CacheKeyIterator, LanceCache, QuickCacheBackend};
 use lance_core::deepsize::DeepSizeOf;
 use lance_core::{Error, Result};
 use lance_index::IndexType;
@@ -95,7 +95,8 @@ impl Session {
     ///
     /// Parameters:
     ///
-    /// - ***index_cache_size***: the size of the index cache.
+    /// - ***index_cache_size***: the size of the index cache, backed by
+    ///   [`QuickCacheBackend`].
     /// - ***metadata_cache_size***: the size of the metadata cache.
     /// - ***store_registry***: the object store registry to use when opening
     ///   datasets. This determines which schemes are available, and also allows
@@ -106,7 +107,9 @@ impl Session {
         store_registry: Arc<ObjectStoreRegistry>,
     ) -> Self {
         Self {
-            index_cache: GlobalIndexCache(LanceCache::with_capacity(index_cache_size)),
+            index_cache: GlobalIndexCache(LanceCache::with_backend(Arc::new(
+                QuickCacheBackend::with_capacity(index_cache_size),
+            ))),
             metadata_cache: GlobalMetadataCache(LanceCache::with_capacity(metadata_cache_size)),
             index_extensions: HashMap::new(),
             store_registry,