Azure Database for PostgreSQL flexible server - Illegal instruction with pgvector 0.8.0

I’m using Azure Database for PostgreSQL Flexible Server with the vector extension (pgvector 0.8.0). The server raises errors whenever I insert rows with an embedding column. Also, embedding column has HNSW index type.
In the server logs I found that the Postgres backend process is being terminated with an illegal instruction:

2025-09-04 07:47:40 UTC-68b69469.7-LOG: server process (PID 185892) was terminated by signal 4: Illegal instruction","detail": "Failed process was running: INSERT INTO embeddings (...)...
2025-09-04 07:47:40 UTC-68b69469.7-LOG: terminating any other active server processes 
2025-09-04 07:47:40 UTC-68b69469.7-LOG: all server processes terminated; reinitializing
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: database system was interrupted; last known up at 2025-09-04 07:47:20 UTC 
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: database system was not properly shut down; automatic recovery in progress 
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: redo starts at 6E/33029A78 
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: invalid record length at 6E/3302FF90: expected at least 24, got 0 
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: redo done at 6E/3302FF68 system usage: CPU: user: 0.00 s, system: 0.00 s, elapsed: 0.00 s 
2025-09-04 07:47:40 UTC-68b9441c.2d64f-LOG: could not do fast promotion, requesting for an end-of-recovery checkpoint 
2025-09-04 07:47:40 UTC-68b9441c.2d650-LOG: checkpoint starting: end-of-recovery immediate wait

I’ve seen others report the same issue, and some mentioned that moving their Flexible Server to Availability Zone 1 fixed it because the hardware there supported the CPU instructions used by pgvector. My server is already running in Zone 1, but I still encounter this problem.

Since Azure only provides pgvector 0.8.0 on Flexible Server, downgrading to 0.7.x isn’t possible.

What possible solutions or workarounds are there for this problem?