diff --git a/packages/telemetry/src/backends/application-insights.ts b/packages/telemetry/src/backends/application-insights.ts
index 8b40090e7c..dcb6485153 100644
--- a/packages/telemetry/src/backends/application-insights.ts
+++ b/packages/telemetry/src/backends/application-insights.ts
@@ -1,3 +1,4 @@
+import type { ClientRequest } from 'node:http';
 import * as https from 'node:https';
 import os from 'node:os';
 
@@ -8,6 +9,8 @@ import type {
   TelemetryData,
 } from '../types';
 
+const MaxFlushTime = 5000; // 5 seconds
+
 // This key is meant to be publically shared. However, I'm adding a simple
 // obfuscation to mitigate key scraping bots on GitHub. The key is split on
 // hypens and base64 encoded without padding.
@@ -26,6 +29,7 @@ export class ApplicationInsightsBackend implements TelemetryBackend {
   private readonly productName: string;
   private readonly hostname: string;
   private readonly uname: string;
+  private pendingRequests = new Set<ClientRequest>();
 
   constructor(
     userId: string,
@@ -61,9 +65,30 @@ export class ApplicationInsightsBackend implements TelemetryBackend {
     this._post([envelope]);
   }
 
+  // Resolve once all in-flight requests have settled (their response/error
+  // handlers, including any warning, have run), so callers — and tests — can
+  // await the full request lifecycle rather than just the request being sent.
   flush(callback?: FlushCallback): void {
-    // No batching, so nothing to flush. Just call callback.
-    callback?.();
+    const startTime = process.hrtime.bigint();
+
+    const checkPending = () => {
+      if (this.pendingRequests.size === 0) {
+        callback?.();
+        return;
+      }
+
+      const elapsedMs = Number((process.hrtime.bigint() - startTime) / BigInt(1_000_000));
+      if (elapsedMs > MaxFlushTime) {
+        console.warn(`ApplicationInsightsBackend: Flush timed out after ${MaxFlushTime}ms`);
+        this.pendingRequests.forEach((req) => req.destroy());
+        callback?.();
+        return;
+      }
+
+      setTimeout(checkPending, 10);
+    };
+
+    checkPending();
   }
 
   private _post(envelopes: unknown[]): void {
@@ -85,10 +110,13 @@ export class ApplicationInsightsBackend implements TelemetryBackend {
           if (res.statusCode && (res.statusCode < 200 || res.statusCode >= 300)) {
             console.warn(`ApplicationInsightsBackend: Failed to send telemetry event. Status: ${res.statusCode}`);
           }
+          this.pendingRequests.delete(req);
         });
     });
+    this.pendingRequests.add(req);
     req.on('error', (e) => {
       console.warn('Error sending telemetry data to Application Insights', e);
+      this.pendingRequests.delete(req);
     });
     req.write(body);
     req.end();
diff --git a/packages/telemetry/tests/backends/application-insights.spec.ts b/packages/telemetry/tests/backends/application-insights.spec.ts
index 1c64db0157..a130b52aa8 100644
--- a/packages/telemetry/tests/backends/application-insights.spec.ts
+++ b/packages/telemetry/tests/backends/application-insights.spec.ts
@@ -34,6 +34,7 @@ describe('ApplicationInsightsBackend', () => {
   afterEach(() => {
     nock.cleanAll();
     nock.enableNetConnect();
+    jest.restoreAllMocks();
   });
 
   const waitForRequest = async (scope: nock.Scope) => {
@@ -118,7 +119,7 @@ describe('ApplicationInsightsBackend', () => {
 
   it('handles https errors', async () => {
     nock.cleanAll(); // Don't use the default scope
-    const errorScope = nock(AI_ENDPOINT).post(AI_PATH).replyWithError('test error');
+    nock(AI_ENDPOINT).post(AI_PATH).replyWithError('test error');
     const warnSpy = jest.spyOn(console, 'warn').mockImplementation();
     const backend = new ApplicationInsightsBackend('user-id', 'session-id', 'product-name', {
       type: 'application-insights',
@@ -130,7 +131,9 @@ describe('ApplicationInsightsBackend', () => {
 
     backend.sendEvent(event);
 
-    await waitForRequest(errorScope);
+    // Wait for the request's error handler (and its warning) to run, not just
+    // for the request to be sent.
+    await new Promise<void>((resolve) => backend.flush(resolve));
 
     expect(warnSpy).toHaveBeenCalledWith(
       'Error sending telemetry data to Application Insights',
@@ -140,7 +143,7 @@ describe('ApplicationInsightsBackend', () => {
 
   it('logs a warning for non-2xx HTTP responses', async () => {
     nock.cleanAll();
-    const errorScope = nock(AI_ENDPOINT).post(AI_PATH).reply(500, { message: 'Internal Server Error' });
+    nock(AI_ENDPOINT).post(AI_PATH).reply(500, { message: 'Internal Server Error' });
     const warnSpy = jest.spyOn(console, 'warn').mockImplementation();
     const backend = new ApplicationInsightsBackend('user-id', 'session-id', 'product-name', {
       type: 'application-insights',
@@ -152,7 +155,9 @@ describe('ApplicationInsightsBackend', () => {
 
     backend.sendEvent(event);
 
-    await waitForRequest(errorScope);
+    // Wait for the response handler (and its warning) to run, not just for the
+    // request to be sent.
+    await new Promise<void>((resolve) => backend.flush(resolve));
 
     expect(warnSpy).toHaveBeenCalledWith(
       'ApplicationInsightsBackend: Failed to send telemetry event. Status: 500'