Cognitive Services-Form Recognizer: Custom model training stuck / backend state issue — Build operations remain in running state

Question

Cognitive Services-Form Recognizer: Custom model training stuck / backend state issue — Build operations remain in running state

Rich Fantozzi 20

Problem description

I am experiencing an issue with custom neural model training in Azure AI Document Intelligence (Form Recognizer). Starting on October 7, 2026, my training operations have remained stuck in the 'running' state with 0% completed, including one that has been stuck for approximately 24 hours. Despite multiple build attempts, the training does not progress beyond the initial fetch of training data, which succeeds quickly. No explicit error messages or HTTP error codes are provided, but it seems to be a backend state problem where the build remains queued without transitioning to active training.

Environment

Azure AI Document Intelligence (Form Recognizer), Region: East US 2, Pricing tier: S0, Scenario: custom neural model training

What I've already tried

I created multiple build attempts at different times to see if a new training run would initiate normally. Each attempt showed the same behavior: the build remained in 'running' status with 0% completion. I verified that training data access was successful by confirming full dataset fetches via OAuth within seconds of submission, and I checked that the resource's RBAC and network configurations (firewall, private endpoints, outbound restrictions) are correctly set. I also confirmed that the resource's identity has the necessary permissions. No other troubleshooting steps are documented in the case.

Current status

The most recent status, as of October 9, 2026, indicates that three build operations are still stuck in 'running' with 0% progress, including a recent build attempted later that day. There is no indication of a change in status or any platform error beyond the build remaining idle and unprogressive. I am seeking assistance to escalate this backend issue and restore normal training operations.

Ganesh Gurram 7,660 Reputation points Microsoft External Staff Moderator

2026-10-09T19:43:51.9466667+00:00

Hello @Rich Fantozzi

Thanks for the question and using MS Q&A forum.

We are actively working with our backend team on this case, and we will get back to you with an update soon.
Ganesh Gurram 7,660 Reputation points Microsoft External Staff Moderator

2026-10-19T03:27:52.0733333+00:00

@Rich Fantozzi - Following up to see if the below answer was helpful. If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Answer accepted by question author

0 additional answers

Your answer

Ganesh Gurram 7,660 Reputation points Microsoft External Staff Moderator

2026-10-09T19:43:51.9466667+00:00

Hello @Rich Fantozzi

Thanks for the question and using MS Q&A forum.

We are actively working with our backend team on this case, and we will get back to you with an update soon.
Ganesh Gurram 7,660 Reputation points Microsoft External Staff Moderator

2026-10-19T03:27:52.0733333+00:00

@Rich Fantozzi - Following up to see if the below answer was helpful. If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.

Answer 1

Ganesh Gurram 7,660 Microsoft External Staff Moderator

Hello @Rich Fantozzi

Issue: Experiencing an issue with custom neural model training in Azure AI Document Intelligence (Form Recognizer). Starting on October 7, 2026, my training operations have remained stuck in the 'running' state with 0% completed, including one that has been stuck for approximately 24 hours. Despite multiple build attempts, the training does not progress beyond the initial fetch of training data, which succeeds quickly. No explicit error messages or HTTP error codes are provided, but it seems to be a backend state problem where the build remains queued without transitioning to active training.

Resolution: We have received an update from our backend team confirming that the issue was caused by GPU capacity constraints in the region and that capacity is now available.

I Hope this helps. Do let me know if you have any further queries.

If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

Thank you!

Ganesh Gurram 7,660 Reputation points Microsoft External Staff Moderator

2026-10-23T17:03:04.9966667+00:00

@Rich Fantozzi - Following up to see if the above answer was helpful. If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.