diff --git a/.changeset/fix-unbounded-where-growth.md b/.changeset/fix-unbounded-where-growth.md
new file mode 100644
index 000000000..777f0b5f2
--- /dev/null
+++ b/.changeset/fix-unbounded-where-growth.md
@@ -0,0 +1,5 @@
+---
+'@tanstack/db': patch
+---
+
+Fix unbounded WHERE expression growth in `DeduplicatedLoadSubset` when loading all data after accumulating specific predicates. The deduplication layer now correctly tracks the original request predicate (e.g., `where: undefined` for "load all") instead of the optimized difference query sent to the backend, ensuring `hasLoadedAllData` is properly set and subsequent requests are deduplicated.
diff --git a/packages/db/src/query/subset-dedupe.ts b/packages/db/src/query/subset-dedupe.ts
index 4959d3cbb..87d703b74 100644
--- a/packages/db/src/query/subset-dedupe.ts
+++ b/packages/db/src/query/subset-dedupe.ts
@@ -126,28 +126,29 @@ export class DeduplicatedLoadSubset {
       return prom
     }
 
-    // Not fully covered by existing data
-    // Compute the subset of data that is not covered by the existing data
-    // such that we only have to load that subset of missing data
-    const clonedOptions = cloneOptions(options)
+    // Not fully covered by existing data — load the missing subset.
+    // We need two clones: trackingOptions preserves the original predicate for
+    // accurate tracking (e.g., where=undefined means "all data"), while loadOptions
+    // may be narrowed with a difference expression for the actual backend request.
+    const trackingOptions = cloneOptions(options)
+    const loadOptions = cloneOptions(options)
     if (this.unlimitedWhere !== undefined && options.limit === undefined) {
       // Compute difference to get only the missing data
       // We can only do this for unlimited queries
       // and we can only remove data that was loaded from unlimited queries
       // because with limited queries we have no way to express that we already loaded part of the matching data
-      clonedOptions.where =
-        minusWherePredicates(clonedOptions.where, this.unlimitedWhere) ??
-        clonedOptions.where
+      loadOptions.where =
+        minusWherePredicates(loadOptions.where, this.unlimitedWhere) ??
+        loadOptions.where
     }
 
     // Call underlying loadSubset to load the missing data
-    const resultPromise = this._loadSubset(clonedOptions)
+    const resultPromise = this._loadSubset(loadOptions)
 
     // Handle both sync (true) and async (Promise<void>) return values
     if (resultPromise === true) {
-      // Sync return - update tracking synchronously
-      // Clone options before storing to protect against caller mutation
-      this.updateTracking(clonedOptions)
+      // Sync return - update tracking with the original predicate
+      this.updateTracking(trackingOptions)
       return true
     } else {
       // Async return - track the promise and update tracking after it resolves
@@ -158,16 +159,14 @@ export class DeduplicatedLoadSubset {
 
       // We need to create a reference to the in-flight entry so we can remove it later
       const inflightEntry = {
-        options: clonedOptions, // Store cloned options for subset matching
+        options: loadOptions, // Store load options for subset matching of in-flight requests
         promise: resultPromise
           .then((result) => {
             // Only update tracking if this request is still from the current generation
             // If reset() was called, the generation will have incremented and we should
             // not repopulate the state that was just cleared
             if (capturedGeneration === this.generation) {
-              // Use the cloned options that we captured before any caller mutations
-              // This ensures we track exactly what was loaded, not what the caller changed
-              this.updateTracking(clonedOptions)
+              this.updateTracking(trackingOptions)
             }
             return result
           })
diff --git a/packages/db/tests/query/subset-dedupe.test.ts b/packages/db/tests/query/subset-dedupe.test.ts
index 9067c931f..3be135ed5 100644
--- a/packages/db/tests/query/subset-dedupe.test.ts
+++ b/packages/db/tests/query/subset-dedupe.test.ts
@@ -4,7 +4,6 @@ import {
   cloneOptions,
 } from '../../src/query/subset-dedupe'
 import { Func, PropRef, Value } from '../../src/query/ir'
-import { minusWherePredicates } from '../../src/query/predicate-utils'
 import type { BasicExpression, OrderBy } from '../../src/query/ir'
 import type { LoadSubsetOptions } from '../../src/types'
 
@@ -517,9 +516,6 @@ describe(`createDeduplicatedLoadSubset`, () => {
         eq(ref(`status`), val(`active`)),
       )
 
-      const test = minusWherePredicates(secondPredicate, firstPredicate)
-      console.log(`test`, test)
-
       await deduplicated.loadSubset({ where: secondPredicate })
       expect(callCount).toBe(2)
       expect(calls[1]).toEqual({
@@ -596,7 +592,135 @@ describe(`createDeduplicatedLoadSubset`, () => {
       // i.e. should request NOT (age > 20)
       await deduplicated.loadSubset({})
       expect(callCount).toBe(2)
-      expect(calls[1]).toEqual({ where: not(gt(ref(`age`), val(20))) }) // Should request all data except what we already loaded
+      expect(calls[1]).toEqual({ where: not(gt(ref(`age`), val(20))) })
+
+      // After loading all data, subsequent calls should be deduplicated
+      const result = await deduplicated.loadSubset({
+        where: gt(ref(`age`), val(5)),
+      })
+      expect(result).toBe(true)
+      expect(callCount).toBe(2)
+    })
+
+    it(`should not produce unbounded WHERE expressions when loading all data after eq accumulation`, async () => {
+      // This test reproduces the production bug where accumulating many eq predicates
+      // and then loading all data (no WHERE clause) caused unboundedly growing
+      // expressions instead of correctly setting hasLoadedAllData=true.
+      let callCount = 0
+      const calls: Array<LoadSubsetOptions> = []
+      const mockLoadSubset = (options: LoadSubsetOptions) => {
+        callCount++
+        calls.push(cloneOptions(options))
+        return Promise.resolve()
+      }
+
+      const deduplicated = new DeduplicatedLoadSubset({
+        loadSubset: mockLoadSubset,
+      })
+
+      // Simulate visiting multiple tasks, each adding an eq predicate
+      for (let i = 0; i < 10; i++) {
+        await deduplicated.loadSubset({
+          where: eq(ref(`task_id`), val(`uuid-${i}`)),
+        })
+      }
+      // After 10 eq calls, unlimitedWhere should be IN(task_id, [uuid-0, ..., uuid-9])
+      expect(callCount).toBe(10)
+
+      // Now load all data (no WHERE clause)
+      // This should send NOT(IN(...)) to the backend but track as "all data loaded"
+      await deduplicated.loadSubset({})
+      expect(callCount).toBe(11)
+
+      // The load request should be NOT(IN(task_id, [all accumulated uuids]))
+      const loadWhere = calls[10]!.where as any
+      expect(loadWhere.name).toBe(`not`)
+      expect(loadWhere.args[0].name).toBe(`in`)
+      expect(loadWhere.args[0].args[0].path).toEqual([`task_id`])
+      const loadedUuids = (
+        loadWhere.args[0].args[1].value as Array<string>
+      ).sort()
+      const expectedUuids = Array.from(
+        { length: 10 },
+        (_, i) => `uuid-${i}`,
+      ).sort()
+      expect(loadedUuids).toEqual(expectedUuids)
+
+      // Critical: after loading all data, subsequent requests should be deduplicated
+      const result1 = await deduplicated.loadSubset({
+        where: eq(ref(`task_id`), val(`uuid-999`)),
+      })
+      expect(result1).toBe(true) // Covered by "all data" load
+      expect(callCount).toBe(11) // No additional call
+
+      // Loading all data again should also be deduplicated
+      const result2 = await deduplicated.loadSubset({})
+      expect(result2).toBe(true)
+      expect(callCount).toBe(11) // Still no additional call
+    })
+
+    it(`should not produce unbounded WHERE expressions with synchronous loadSubset`, async () => {
+      // Same scenario as the async accumulation test, but with a sync mock
+      // to exercise the sync return path (line 150 of subset-dedupe.ts)
+      let callCount = 0
+      const mockLoadSubset = () => {
+        callCount++
+        return true as const
+      }
+
+      const deduplicated = new DeduplicatedLoadSubset({
+        loadSubset: mockLoadSubset,
+      })
+
+      // Accumulate eq predicates via sync returns
+      for (let i = 0; i < 10; i++) {
+        deduplicated.loadSubset({
+          where: eq(ref(`task_id`), val(`uuid-${i}`)),
+        })
+      }
+      expect(callCount).toBe(10)
+
+      // Load all data (no WHERE clause) — should track as "all data loaded"
+      deduplicated.loadSubset({})
+      expect(callCount).toBe(11)
+
+      // Subsequent requests should be deduplicated
+      const result1 = deduplicated.loadSubset({
+        where: eq(ref(`task_id`), val(`uuid-999`)),
+      })
+      expect(result1).toBe(true)
+      expect(callCount).toBe(11)
+
+      const result2 = deduplicated.loadSubset({})
+      expect(result2).toBe(true)
+      expect(callCount).toBe(11)
+    })
+
+    it(`should handle multiple all-data loads without expression growth`, async () => {
+      let callCount = 0
+      const mockLoadSubset = () => {
+        callCount++
+        return Promise.resolve()
+      }
+
+      const deduplicated = new DeduplicatedLoadSubset({
+        loadSubset: mockLoadSubset,
+      })
+
+      // First: load some specific data
+      await deduplicated.loadSubset({
+        where: eq(ref(`task_id`), val(`uuid-1`)),
+      })
+      expect(callCount).toBe(1)
+
+      // Load all data (first time)
+      await deduplicated.loadSubset({})
+      expect(callCount).toBe(2)
+
+      // Load all data (second time) - should be deduplicated since we already have everything
+      const result = await deduplicated.loadSubset({})
+      expect(result).toBe(true)
+      expect(callCount).toBe(2) // No additional call - all data already loaded
     })
 
     it(`should handle multiple overlapping unlimited calls`, async () => {