Predictika · Nebius GPU commitments

An AI sales agent's GPU commitment, checked before it goes out

An AI sales agent prepares a Nebius AI Cloud GPU commitment for a customer: GPU platform, region, GPUs, term, cluster, reservation, account type, discount. Before it goes out, the agent sends it to Predictika, which checks it against a declarative model of the published rules and a capacity ledger, and returns a verdict, every reason, the figures it computed and, for a commitment that can proceed, the line-item quote itself.

Scope: Nebius AI Cloud GPU capacity (H100, H200, B200, B300): regions, clusters, quotas, reservations and commitment discounts. See everything this validator covers

This validator models Nebius GPU commitments and the commercial terms quoted around them — not just the scenarios shown below.

  • Platforms and regions — Which GPU generation is offered in which region (for example H100 only in eu-north1), and the current on-demand list price per GPU hour.
  • Clusters — InfiniBand GPU clusters take only the 8-GPU preset, so a cluster must be a multiple of 8 GPUs.
  • Quota and reservations — Default per-region GPU quotas without a reservation (several are zero), and capacity reservations that bypass them.
  • Capacity — GPUs free by platform and region from a capacity ledger the operator supplies, with a warning when an order leaves little.
  • Commitment discounts — Company accounts only, up to 35 percent, generally prepaid, for multiple months and large clusters; approval lines where the public text gives no numbers.
  • Managed services — Soperator is not offered in every region and its GPU workers need reserved capacity.
  • Stated figures — A rate, a total or a free-GPU count the request states is compared with the computed one.

Any Nebius GPU commitment that touches these is in scope — not only the scenarios pictured here. Not covered: NVL72 racks (contact sales), Token Factory, storage and managed database pricing, and spot pricing. The capacity ledger and the approval lines are placeholders, not Nebius data.

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

6 scenarios in 3 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

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
A GPU generation offered only in some regionsOnly with care
A matrix of GPU by region has to be loaded as option constraints and changes with every new region.
Region and GPU are columns of the capacity ledger; the check compares against the row chosen.
InfiniBand clusters only in 8-GPU nodesEasy
A quantity multiple on the cluster product.
Same, from the model.
A default quota of zero without a reservationOnly with care
Quota lives in the cloud console, not on the quote; a rule needs the quota loaded per region and GPU, plus whether a reservation covers it.
Quota is a column of the ledger; the reservation flag decides whether it applies.
A discount only for company accounts, up to 35 percent, prepaid, for multiple months and large clustersOnly with care
A discount matrix can hold the cap and the account type, but "multiple months" and "large clusters" have no published numbers, so teams write thresholds that drift.
Each condition is its own rule with its own message and approval path.
Managed Soperator not offered in every region, and its GPU workers needing reserved capacityOnly with care
One rule per region plus a cross-check against the reservation.
Both follow from the pool and the reservation flag.
A request that states the rate or the free GPUsOnly with care
A validation rule can compare a stated figure with a calculated field, but only if the stated figure is a field on the quote, which it rarely is.
The stated figures are compared with the computed ones at every step of processing the request.
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 customer asks for us-central1, but H100 is offered in eu-north1, not there.
260 GPUs cannot form an InfiniBand cluster; clusters take the 8-GPU preset, so the count must be a multiple of 8.
364 GPUs without a capacity reservation is above the default quota of 0 for B200 in us-central1; reserve capacity or request a quota increase.
4a discount of 40 percent is above the 35 percent Nebius publishes as the most for a commitment.
5commitment discounts are for company accounts; this is an individual account.
6managed Soperator is not offered in us-north1.
7managed Soperator GPU workers need a capacity block group; the quote has no reservation.
8the request states $9.50 per GPU hour, but the discounted rate is $7.60.
9a discount on 32 GPUs; Nebius describes commitment discounts for large-scale clusters.
10this order leaves only 32 GPUs free for H200 in us-central1; confirm with capacity planning.
11commitment discounts are generally prepaid; this quote has no prepayment.
12a discount on a 60 day term; Nebius describes commitment discounts for multiple months.
13a discount of 35 percent is above 25 percent; the most Nebius publishes is 35 percent, so sales approval is required.
'