diff --git a/packages/core/src/tracing/vercel-ai/index.ts b/packages/core/src/tracing/vercel-ai/index.ts
index 7d72b3de3ae7..f9abcb096dc8 100644
--- a/packages/core/src/tracing/vercel-ai/index.ts
+++ b/packages/core/src/tracing/vercel-ai/index.ts
@@ -513,10 +513,20 @@ function processGenerateSpan(span: Span, name: string, attributes: SpanAttribute
   }
 }
 
+const CLIENTS_WITH_VERCEL_AI_PROCESSORS = new WeakSet<Client>();
+
 /**
  * Add event processors to the given client to process Vercel AI spans.
+ *
+ * Idempotent: both the integration and the Next.js SDK register these, and duplicate hooks would
+ * process every span twice.
  */
 export function addVercelAiProcessors(client: Client): void {
+  if (CLIENTS_WITH_VERCEL_AI_PROCESSORS.has(client)) {
+    return;
+  }
+  CLIENTS_WITH_VERCEL_AI_PROCESSORS.add(client);
+
   client.on('spanStart', onVercelAiSpanStart);
   // Note: We cannot do this on `spanEnd`, because the span cannot be mutated anymore at this point
   client.addEventProcessor(Object.assign(vercelAiEventProcessor, { id: 'VercelAiEventProcessor' }));
diff --git a/packages/nextjs/src/server/index.ts b/packages/nextjs/src/server/index.ts
index 203b0f348a97..d4038fa0e6e1 100644
--- a/packages/nextjs/src/server/index.ts
+++ b/packages/nextjs/src/server/index.ts
@@ -4,6 +4,7 @@
 import { HTTP_TARGET, URL_QUERY } from '@sentry/conventions/attributes';
 import type { EventProcessor } from '@sentry/core';
 import {
+  addVercelAiProcessors,
   applySdkMetadata,
   debug,
   getClient,
@@ -210,6 +211,14 @@ export function init(options: NodeOptions): NodeClient | undefined {
 
   const client = nodeInit(opts);
 
+  // Next.js bundles `ai`, so the integration can neither patch the module nor detect it via `Modules`
+  // (which only sees the app's own `package.json`, missing workspace and transitive deps). Register
+  // the processors here instead — after init, so an explicitly constructed `vercelAIIntegration()`
+  // can't override the default back to the broken behavior.
+  if (client?.getIntegrationByName('VercelAI')) {
+    addVercelAiProcessors(client);
+  }
+
   client?.on('beforeSampling', ({ spanAttributes }, samplingDecision) => {
     // There are situations where the Next.js Node.js server forwards requests for the Edge Runtime server (e.g. in
     // middleware) and this causes spans for Sentry ingest requests to be created. These are not exempt from our tracing