Predictika · AI cloud capacity quotes

An AI sales agent's Nebius GPU capacity quote, checked before it goes out

An AI sales agent prepares a capacity quote for Nebius AI Cloud GPU virtual machines: the GPU platform, region, VM preset, number of VMs, whether the nodes sit in an InfiniBand GPU cluster, and how the capacity is bought. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the platform, region and preset rules Nebius publishes and returns a verdict, every reason and the capacity figures it computed.

Scope: Nebius AI Cloud GPU virtual machine capacity quotes: platform, region, VM preset, GPU clusters and the purchase option. See everything this validator covers

This validator models capacity quotes for Nebius AI Cloud GPU virtual machines, not just the scenarios shown below.

  • Platforms and regions — NVIDIA H100, H200, B200, B300, RTX PRO 6000 and L40S (Intel and AMD) platforms, and the regions in which each is offered.
  • VM presets — the published GPU, vCPU and RAM presets for each platform; the quote totals the GPUs, vCPUs and RAM across all VMs and compares the GPU total with the figure the quote states.
  • GPU clusters — InfiniBand GPU clusters are offered only on the H100, H200, B200 and B300 platforms and only with the 8-GPU presets.
  • Purchase option — on-demand and spot capacity pass; reserved capacity terms are not published, so a quote with reserved capacity goes to a person for approval. A region the documentation marks as private (eu-north2) also goes to a person.

Any capacity quote for these Nebius GPU platforms that touches these rules is in scope — not only the scenarios pictured here. Not covered: prices (the public price page shows two sets of GPU-hour rates, before and from October 1, 2026, and reserved terms are quote-only, so this validator builds no price), GB200 and GB300 NVL72 systems (contact sales), CPU-only instances, storage, quotas and capacity availability. Platform, preset and region details were read from Nebius's public documentation through a summarizing reader and checked on two separate reads.

The AI sales agentBuilds the capacity quote from the conversation and says it is ready.
Predictika validatorChecks every rule at once and returns the verdict, each finding, and the computed figures.

5 scenarios in 2 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI sales agent says

Illustrative agent statement. The validator receives the structured request below.

Proposing agent: AI sales agent (scripted)

Before sending, the agent must call Recorded run

Predictika validator


      

        
        
        
      

      
Waiting for the call.

      
    

  

  
  

    
Where rules-based CPQ gets it wrong
Click to expand
    
Our assessment of what a typical rules-based CPQ (product rules, price rules and validation rules, as in Salesforce CPQ) does with the same cases. It is a judgment about effort and reliability, not a claim that nothing can be built.

    
What the validator caughtTypical rules-based CPQPredictika
Which GPU platform is offered in which regionOnly with care
A region-by-platform availability table that has to be kept in step with the documentation, one rule per platform.		The platform is checked against its regions on every call, together with everything else.
Presets that belong to one platform onlyOnly with care
Each preset is valid on some platforms and not others, so every product line needs its own list of valid presets.		The preset is checked against the platform on every call.
GPU clusters allowed only on some platforms and only with 8-GPU presetsOnly with workarounds
The cluster option on one line depends on the platform on another line and on the number of GPUs in the preset on a third, so it needs a cross-line rule.		Cluster, platform and preset are in one request and are checked together.
A stated GPU total that must equal presets times VMsOnly with care
The total is a free-entry field; a rule can compare it with presets times VMs, but only once that product is computed on the quote.		The stated total travels with the quote and is compared with the computed total on every call.
Reserved capacity that has no published termsOnly with care
Needs a special case that holds the quote for a person instead of pricing it.		Reserved capacity sends the quote to a person for approval, with the other checks still run.
Every problem returned in one answerOnly with workarounds
Rules tend to fire as the quote is saved or a step is completed, so a rep often fixes one error and meets the next.		All findings come back together, whatever order the choices arrived in.

  

  

    
What this validator catches: findings from its test cases
Click to expand
    
Every rule is checked on every call, in any order the choices arrive; one problem never hides another. Blocking findings stop the quote; approval findings need a person to sign off; everything else is computed.

    
1The chosen GPU platform is not offered in the chosen region; see the platform and region list in the Nebius documentation.
2The Nebius documentation marks the eu-north2 region as private; a person should confirm access before the quote goes out.
3The chosen VM preset is not one of the presets offered for the chosen GPU platform.
4GPU clusters accept only the 8-GPU presets; the chosen preset has 1 GPU(s).
5A GPU cluster with InfiniBand is offered only on the H100, H200, B200 and B300 platforms.
6Reserved capacity terms and discounts are set with Nebius sales and are not published; a person should confirm them.
7The quote has 0 virtual machines; at least 1 is needed.
8The quote states 24 GPUs in total; the VM preset and VM count give 32 GPUs.
9The quote states 40 GPUs in total; the VM preset and VM count give 32 GPUs.
10The quote states 30 GPUs in total; the VM preset and VM count give 32 GPUs.

  

  

    
Scenarios and results
Click to expand
    
The recorded result of the validator for each scenario on this page: the request the AI agent sent, the verdict, and every finding, in the order they appear above.

    
1. Cluster presets and stated total: Reserved H200 cluster of single-GPU VMs with a wrong GPU total

    
The agent's request. Please quote 12 NVIDIA H200 virtual machines with the 1gpu-16vcpu-200gb preset in the eu-north1 region, as reserved capacity, placed in an InfiniBand GPU cluster. The quote states 16 GPUs in total. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • GPU clusters use 8-GPU presets only. GPU clusters accept only the 8-GPU presets; the chosen preset has 1 GPU(s).
    • 
      
  • Stated total GPUs differ from the quote. The quote states 16 GPUs in total; the VM preset and VM count give 12 GPUs.
    • 
      
  • Reserved capacity terms to confirm. Reserved capacity terms and discounts are set with Nebius sales and are not published; a person should confirm them.
    • 
    

    
2. Cluster presets and stated total: The agent resubmits after fixing the errors

    
The agent's request. Please quote 2 NVIDIA H200 virtual machines with the 8gpu-128vcpu-1600gb preset in the eu-north1 region, as reserved capacity, placed in an InfiniBand GPU cluster. The quote states 12 GPUs in total. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Stated total GPUs differ from the quote. The quote states 12 GPUs in total; the VM preset and VM count give 16 GPUs.
    • 
      
  • Reserved capacity terms to confirm. Reserved capacity terms and discounts are set with Nebius sales and are not published; a person should confirm them.
    • 
    

    
3. Cluster presets and stated total: The agent resubmits after fixing the errors

    
The agent's request. Please quote 2 NVIDIA H200 virtual machines with the 8gpu-128vcpu-1600gb preset in the eu-north1 region, as on-demand capacity, placed in an InfiniBand GPU cluster. The quote states 16 GPUs in total. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Capacity summary. 16 GPU(s), 256 vCPU(s) and 3,200 GiB of RAM in total across 2 VM(s).
    • 
    

    
4. Platform and cluster: RTX PRO 6000 cluster in the wrong region

    
The agent's request. Please quote 6 NVIDIA RTX PRO 6000 virtual machines with the 8gpu-192vcpu-1744gb preset in the eu-north1 region, as on-demand capacity, placed in an InfiniBand GPU cluster. The quote states 48 GPUs in total. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Platform not offered in the region. The chosen GPU platform is not offered in the chosen region; see the platform and region list in the Nebius documentation.
    • 
      
  • GPU cluster not supported on the platform. A GPU cluster with InfiniBand is offered only on the H100, H200, B200 and B300 platforms.
    • 
    

    
5. Platform and cluster: The agent resubmits after fixing the errors

    
The agent's request. Please quote 6 NVIDIA RTX PRO 6000 virtual machines with the 8gpu-192vcpu-1744gb preset in the uk-south2 region, as on-demand capacity, without a GPU cluster. The quote states 48 GPUs in total. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Capacity summary. 48 GPU(s), 1,152 vCPU(s) and 10,464 GiB of RAM in total across 6 VM(s).
    • 
    

  

  

    
Narration transcript
Click to expand
    
The text of the narrated walkthrough, step by step.

    
    
      
  1. Welcome to a demo of Predictika's Validator. An AI sales agent prepares a capacity quote for Nebius AI Cloud GPU virtual machines: the GPU platform, region, VM preset, and more. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the platform, region and preset rules Nebius publishes and returns a verdict, every reason and the capacity figures it computed.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Please quote 12 NVIDIA H200 virtual machines with the 1gpu-16vcpu-200gb preset in the eu-north1 region, as reserved capacity, placed in an InfiniBand GPU cluster. The quote states 16 GPUs in total. The rest of the request is on screen.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with two findings and one approval, all in one answer. GPU clusters accept only the 8-GPU presets; the chosen preset has 1 GPU(s). The quote states 16 GPUs in total; the VM preset and VM count give 12 GPUs.
    6. 
      
  6. The validator returned two blocking findings and one approval in one answer. A check that reads one field at a time would pass most of this; it shows only when the choices in the quote are checked against each other and against the stated total.
    7. 
      
  7. What this shows: for GPU clusters allowed only on some platforms and only with 8-GPU presets, a rules-based CPQ can do this only with workarounds.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the VM preset to 8gpu-128vcpu-1600gb, the number of VMs to 2 and the stated Total GPUs as to 12.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. Still blocked, with one finding and one approval, all in one answer. The quote states 12 GPUs in total; the VM preset and VM count give 16 GPUs.
    12. 
      
  12. One finding cleared. Still blocking: stated total GPUs differ from the quote. One approval also remains for a person: reserved capacity terms to confirm.
    13. 
      
  13. The agent resubmits once more.
    14. 
      
  14. The revised request. It changes the purchase option to on demand and the stated Total GPUs as to 16.
    15. 
      
  15. It calls the validator once more.
    16. 
      
  16. This time the validator clears it.
    17. 
      
  17. Both earlier findings are cleared, and the validator's figure matches the quote.
    18. 
      
  18. Three problems caught in one call, fixed in two passes, and a clean quote. That is the value of Predictika's validator: catching mistakes before they reach the customer. Now you can try the other scenarios yourself.
    19.