Tracing has missing output when called inside nextjs `after()` or `waitUntil` function

[micmania1]

Description

I've noticed that LLM calls within after() and waitUntil don't get properly traced. They end up with no output and almost 0 latency (eg. 0.1s).

I initially raised an issue against langsmith which can be found here: langchain-ai/langsmith-sdk#1874

However, I've since tried the same thing with vercel-langfuse and I'm seeing the same problem leading me to think this is an upstream issue, either in this repo or in nextjs. I tried to debug myself to try and pinpoint the issue but I didn't have much luck.

Reproduce

Here are two langsmith examples which have the problem: https://gist.github.com/micmania1/468c16c76459a5f6bba159c27977258b#file-new-exporter-route-ts

AI SDK Version

"@ai-sdk/google": "^2.0.0",
"ai": "^5.0.0",

Note: This problem exists both in ai SDK v4 and v5.