The best way to use Azure ML ? Have lots of deep learning and have Founders membership

Question

I’m in the Microsoft Founders Hub program, which has been a great experience overall. The credits and access are generous, and it’s reassuring to know I can experiment without instantly burning through budget—huge for early-stage teams.

But here’s the problem: when I actually try to use Azure Machine Learning with those benefits, it falls apart.

I’m building a GPU-heavy ML pipeline (Torch, TensorFlow, distributed compute, Docker), and on the surface, Azure ML offers everything I need—custom containers, serverless compute, Dask, Databricks, Spark pools, etc. But when it comes time to actually run the job, I hit a wall. I can configure environments, upload Docker contexts, and see advanced GPU SKUs listed… but then I’m told I don’t have enough cores. Or the image won’t build. Or the quotas aren’t aligned with the requirements.

It’s like being handed the keys to a Ferrari—with no gas in the tank.

So my question is this:

As a Founders Hub member, what’s the real path to using Azure ML for GPU-backed model training? What is the practical way I’m expected to use this compute if I can’t access the resources being advertised? Is there a preferred track or support channel I’m missing?

I’m all-in on doing it “the Azure way,” but I need that way to be actionable.

Answer 1

Thank you again for taking the time to provide these suggestions. I genuinely appreciate the effort to help.

That said, I want to respectfully emphasize that we are not new to this process. We’ve already followed the recommended GPU request procedures multiple times — not once or twice, but more than 20 times — across several support tickets, forms, and escalation routes. We’ve been diligent and patient, but despite our best efforts, the results have been inconsistent or nonexistent.

We’ve read all the documentation, reviewed troubleshooting guides, and tested every suggestion offered. At this point, being redirected to the same initial steps without real progress has left us at a standstill — particularly as we’re preparing our infrastructure for scaling under the Founders program.

I understand quotas are based on availability and internal policy, but if there’s a more direct escalation path or a different conversation we should be having, I’m all ears. We’re not asking for shortcuts — we’re just hoping for clarity, consistency, and a way forward.

Thanks again for your time, and I look forward to any support you can extend here.

Best regards,

Brandon

Hi Piyushi,

Thanks again for the reply.

I want to be absolutely clear: this is not a new request. We’ve already submitted 18 separate support tickets related to GPU quota increases under our Microsoft for Startups – Azure Sponsorship subscription. Most were submitted using the correct process you’ve outlined. At this point, we’ve received mixed outcomes—some approvals, many denials, and even more that are stuck in limbo or completely ignored.

---

🔢 Below are the 18 Support Request Numbers:

1. 2503270040000032 – Central US – escalated by Suraj Kumar

2. 2503310040011634 – Central US

3. 2503310040011641 – Central US

4. 2503310040000749 – West US 2 – Approved

5. 2503310040000773 – Central US – currently under review by Karunakar Reddy

6. 2503260010004580 – Compute-VM (cores-vCPUs)

7. 2503260040008504 – Australia East

8. 2303200040000678 – No response

9. 2303200040000683 – No resolution

10. 2303200040000688 – Closed without comment

11. 2303200040000689 – Closed, no action taken

12. 2303200040000692 – Marked complete, quota not updated

13. 2303200040000693 – Denied without justification

14. 2303200040000694 – Marked resolved, still pending in portal

15. 2303200040000695 – Duplicate reply, never escalated

16. 2303200040000696 – Referred to pay-as-you-go, no follow-up

17. 2303200040000697 – Rejected due to “sponsorship” limitation

18. 2303200040000698 – Quota not applied even after confirmation email

---

🔍 Summary:

We’ve spent significant time and effort navigating Microsoft’s internal process, only to land in an endless loop of contradictory guidance and mixed signals. What’s worse is being asked to start over when we never stopped trying.