diff --git a/packages/cubejs-server-core/src/core/RefreshScheduler.ts b/packages/cubejs-server-core/src/core/RefreshScheduler.ts
index e94001b9a6bb4..c1cd3990934d4 100644
--- a/packages/cubejs-server-core/src/core/RefreshScheduler.ts
+++ b/packages/cubejs-server-core/src/core/RefreshScheduler.ts
@@ -342,6 +342,14 @@ export class RefreshScheduler {
     const compilers = await compilerApi.getCompilers();
     const queryForEvaluation = await compilerApi.createQueryByDataSource(compilers, {});
 
+    // Bound the dispatch fan-out of refresh-key queries to the orchestrator's
+    // concurrency. Without this, every cube/timezone combination is enqueued
+    // at once, creating a large promise fan-out (memory / event-loop pressure)
+    // on deployments with many cubes or timezones. The limiter is shared per
+    // orchestrator id so concurrent refresh runs resolving to the same
+    // orchestrator respect a single cap.
+    const limit = await this.serverCore.getRefreshKeysLimiter(context, queryingOptions.concurrency);
+
     await Promise.all(queryForEvaluation.cubeEvaluator.cubeNames().map(async cube => {
       const cubeFromPath = queryForEvaluation.cubeEvaluator.cubeFromPath(cube);
       const measuresCount = Object.keys(cubeFromPath.measures || {}).length;
@@ -349,7 +357,7 @@ export class RefreshScheduler {
       if (measuresCount === 0 && dimensionsCount === 0) {
         return;
       }
-      await Promise.all(queryingOptions.timezones.map(async timezone => {
+      await Promise.all(queryingOptions.timezones.map(timezone => limit(async () => {
         const query = {
           ...queryingOptions,
           ...(
@@ -373,7 +381,7 @@ export class RefreshScheduler {
           scheduledRefresh: true,
           loadRefreshKeysOnly: true,
         });
-      }));
+      })));
     }));
   }
 
diff --git a/packages/cubejs-server-core/src/core/server.ts b/packages/cubejs-server-core/src/core/server.ts
index da04f2468415f..6912d23df0180 100644
--- a/packages/cubejs-server-core/src/core/server.ts
+++ b/packages/cubejs-server-core/src/core/server.ts
@@ -130,6 +130,15 @@ export class CubejsServerCore {
 
   protected readonly orchestratorStorage: OrchestratorStorage = new OrchestratorStorage();
 
+  /**
+   * Concurrency limiters used to throttle the refresh-key dispatch fan-out of
+   * the scheduled refresh, keyed by orchestrator id. Sharing a single limiter
+   * per orchestrator id ensures that concurrent scheduled refresh runs that
+   * resolve to the same orchestrator (e.g. multiple security contexts) don't
+   * collectively flood the event loop with refresh-key queries.
+   */
+  protected readonly refreshKeysLimiters: Map<string, { concurrency: number, limiter: pLimit.Limit }> = new Map();
+
   // eslint-disable-next-line @typescript-eslint/no-unused-vars
   protected repositoryFactory: ((context: RequestContext) => SchemaFileRepository) | (() => FileRepository);
 
@@ -570,6 +579,26 @@ export class CubejsServerCore {
     this.startScheduledRefreshTimer();
   }
 
+  /**
+   * Returns a concurrency limiter shared across scheduled refresh runs that
+   * resolve to the same orchestrator id. It is used to bound the refresh-key
+   * dispatch fan-out so that deployments with many cubes/timezones don't
+   * enqueue every refresh-key query at once.
+   */
+  public async getRefreshKeysLimiter(context: RequestContext, concurrency: number): Promise<pLimit.Limit> {
+    const orchestratorId = await this.contextToOrchestratorId(context);
+    const existing = this.refreshKeysLimiters.get(orchestratorId);
+
+    if (existing && existing.concurrency === concurrency) {
+      return existing.limiter;
+    }
+
+    const limiter = pLimit(concurrency);
+    this.refreshKeysLimiters.set(orchestratorId, { concurrency, limiter });
+
+    return limiter;
+  }
+
   public async getOrchestratorApi(context: RequestContext): Promise<OrchestratorApi> {
     const orchestratorId = await this.contextToOrchestratorId(context);
 
diff --git a/packages/cubejs-server-core/test/unit/RefreshScheduler.test.ts b/packages/cubejs-server-core/test/unit/RefreshScheduler.test.ts
index 8a274e0256f63..3c7ec9be4ca8c 100644
--- a/packages/cubejs-server-core/test/unit/RefreshScheduler.test.ts
+++ b/packages/cubejs-server-core/test/unit/RefreshScheduler.test.ts
@@ -1232,4 +1232,74 @@ describe('Refresh Scheduler', () => {
     // backoffDataStillActive exists, which means backoff is still in place
     // (nextTimestamp may be close to current time due to test execution delays)
   });
+
+  test('getRefreshKeysLimiter returns a shared limiter per orchestrator id', async () => {
+    const { serverCore } = setupScheduler({
+      repository: repositoryWithoutPreAggregations,
+      skipAssertSecurityContext: true,
+    });
+
+    const ctx = { authInfo: null, securityContext: {}, requestId: 'XXX' };
+
+    const first = await serverCore.getRefreshKeysLimiter(ctx, 2);
+    const second = await serverCore.getRefreshKeysLimiter(ctx, 2);
+    // Same orchestrator id and concurrency should reuse the limiter
+    expect(second).toBe(first);
+
+    // Changing the concurrency rebuilds the limiter
+    const third = await serverCore.getRefreshKeysLimiter(ctx, 3);
+    expect(third).not.toBe(first);
+  });
+
+  test('Refresh key dispatch is bounded by concurrency', async () => {
+    process.env.CUBEJS_EXTERNAL_DEFAULT = 'false';
+    process.env.CUBEJS_SCHEDULED_REFRESH_DEFAULT = 'true';
+
+    const { refreshScheduler, serverCore } = setupScheduler({
+      repository: repositoryWithPreAggregations,
+    });
+
+    const ctx = { authInfo: { tenantId: 'tenant1' }, securityContext: { tenantId: 'tenant1' }, requestId: 'XXX' };
+    const concurrency = 2;
+
+    let inFlight = 0;
+    let maxInFlight = 0;
+
+    // Wrap the orchestrator's executeQuery once to track how many refresh-key
+    // dispatches are running concurrently.
+    const orchestratorApi = await serverCore.getOrchestratorApi(ctx);
+    const realExecuteQuery = orchestratorApi.executeQuery.bind(orchestratorApi);
+    orchestratorApi.executeQuery = async (query: any) => {
+      if (query.loadRefreshKeysOnly) {
+        inFlight += 1;
+        maxInFlight = Math.max(maxInFlight, inFlight);
+        try {
+          // Hold the slot briefly so concurrent dispatches can build up.
+          await new Promise(resolve => setTimeout(resolve, 25));
+          return await realExecuteQuery(query);
+        } finally {
+          inFlight -= 1;
+        }
+      }
+      return realExecuteQuery(query);
+    };
+    jest.spyOn(serverCore, 'getOrchestratorApi').mockResolvedValue(orchestratorApi);
+
+    // Two cubes (Foo, Bar) x two timezones => four refresh-key dispatches.
+    // Without the limiter all four would run at once; with it no more than
+    // `concurrency` should overlap.
+    for (let i = 0; i < 1000; i++) {
+      const refreshResult = await refreshScheduler.runScheduledRefresh(ctx, {
+        concurrency,
+        workerIndices: [0],
+        timezones: ['UTC', 'America/Los_Angeles'],
+      });
+      if (refreshResult.finished) {
+        break;
+      }
+    }
+
+    expect(maxInFlight).toBeGreaterThan(0);
+    expect(maxInFlight).toBeLessThanOrEqual(concurrency);
+  });
 });