Predictika · NVIDIA AI factory hall design

An AI design agent's AI factory hall design, checked before it goes out

An AI design agent drafts an AI factory hall from NVIDIA rack-scale systems: racks, liquid cooling, power, storage and a leaf/spine fabric. Before the design goes out, the agent sends it to Predictika, which checks it against a declarative model built from NVIDIA's public product information, compares the GPU count and capex the agent states with the figures it computes, and returns every finding together with the line-item bill of materials.

Scope: NVIDIA AI factory hall design from four rack-scale systems (GB300 NVL72, Vera Rubin NVL72, Vera Rubin Ultra NVL576, Rubin Ultra NVL144 Kyber) with in-row CDUs, 800 VDC sidecars, storage racks and a leaf/spine fabric. See everything this validator covers

This validator models the design of an AI factory hall built from NVIDIA rack-scale systems. It is an illustrative model built from NVIDIA's public product information; every rating, weight, price and limit in it is a fictional illustrative value.

  • Cooling and power — coolant temperature at the racks after the CDU, liquid and air cooling capacity, CDU capacity with a spare unit, 800 VDC architecture and sidecar sizing per row, row busway and the site power budget.
  • Structure and layout — floor loading per rack, positions per row, hall rows, and NVLink-domain rack counts.
  • Fabric — leaf downlinks against GPU ports, spine ports against leaf uplinks, and InfiniBand versus Ethernet matching.
  • Delivery and money — rack ship year against the hall go-live year, GPU target, capex against the budget (with a finance sign-off band), and the GPU count and capex the agent states.

Within this scope the validator handles any hall design built from these rack systems, at any row count, power budget or fabric size, not just the scenarios shown here. A hall uses one rack system in all of its compute rows. It does not cover NVIDIA products outside these rack-scale systems, other vendors' racks, or building structure beyond floor loading.

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

7 scenarios in 4 groups · each group runs from the agent’s first attempt to the outcome · click one to switch

What the agent proposes

The AI design agent says

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

Proposing agent: AI design 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
Coolant temperature after the CDUOnly with care
The limit belongs to the rack system and the water temperature rise belongs to the CDU, so the check needs a figure derived across two catalogs. A validation rule can do it once someone builds the calculated field.		Both are in the model; the temperature at the racks is computed and compared on every call.
Power sidecars sized to the rowOnly with care
Per-row power against sidecar capacity, only for racks that take 800 VDC. It takes a rule per rack system, and a hall that adds a row later is easy to miss.		The row draw and the sidecar capacity are computed together for whatever rack system is chosen.
Fabric ports across both layersOnly with workarounds
Leaf downlinks against GPU ports and spine ports against leaf uplinks are arithmetic across layers that a product-rule engine does not carry; usually done outside the tool.		Port totals for both layers are computed and checked together, so one short layer cannot hide behind the other.
The agent's stated GPUs and capex against the computed onesOnly 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 design, which it rarely is.		The stated figures travel with the design and are compared with the computed ones on every call.

  

  

    
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 design; approval findings need a person to sign off; everything else is computed.

    
1GB300 NVL72 needs coolant at or below 42 C; 40 C facility water through the In-row CDU 1.4 MW reaches the racks at 43 C.
2Rubin Ultra NVL144 Kyber takes 800 VDC power only; this hall distributes 415 VAC.
3Each row of Rubin Ultra NVL144 Kyber draws 2,400 kW and needs 800 VDC power sidecars; the sidecars in a row give 0 kW.
4Rubin Ultra NVL144 Kyber loads the floor at 4,250 kg/m2 (3,400 kg per rack); the floor is rated 2,500 kg/m2.
55,184 GPU fabric ports; 150 leaves give 4,800 downlinks.
64,800 leaf uplinks; 70 spines give 4,480 ports.
7The agent stated $363,690,000; the validator computes $361,620,000.
8Estimated capex $363,690,000 is above 95 percent of the $375,000,000 budget; finance must sign off.
9Rubin Ultra NVL144 Kyber is not expected to ship until 2027; the hall goes live in 2026.
10The design delivers 3,456 GPUs; the target is 4,000.

  

  

    
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. Power and floor: Kyber hall on 415 VAC power

    
The agent's request. Please approve a hall of 6 rows with 4 Rubin Ultra NVL144 Kyber racks each: 24 racks and 3,456 GPUs. The hall distributes 415 VAC on a floor rated 2,500 kg/m2, each row has 3 In-row 2.3 MW CDUs and no power sidecars, there is a storage row of 4 racks, and the fabric is 108 leaf and 54 spine Spectrum-6 CPO switches. The estimated capex is $252,020,000 against a budget of $300,000,000. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Racks need 800 VDC power. Rubin Ultra NVL144 Kyber takes 800 VDC power only; this hall distributes 415 VAC.
    • 
      
  • Not enough power sidecar capacity. Each row of Rubin Ultra NVL144 Kyber draws 2,400 kW and needs 800 VDC power sidecars; the sidecars in a row give 0 kW.
    • 
      
  • Floor cannot carry the racks. Rubin Ultra NVL144 Kyber loads the floor at 4,250 kg/m2 (3,400 kg per rack); the floor is rated 2,500 kg/m2.
    • 
    

    
2. Power and floor: The agent resubmits after fixing the errors

    
The agent's request. Please approve a hall of 6 rows with 4 Rubin Ultra NVL144 Kyber racks each: 24 racks and 3,456 GPUs. The hall distributes 800 VDC on a floor rated 5,000 kg/m2, each row has 3 In-row 2.3 MW CDUs and 2 power sidecars of 1.3 MW, there is a storage row of 4 racks, and the fabric is 108 leaf and 54 spine Spectrum-6 CPO switches. The estimated capex is $257,420,000 against a budget of $300,000,000. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Quote summary. Total capex $257,420,000: compute $228,000,000; storage $2,600,000; cooling $6,840,000; power $5,400,000; fabric $14,580,000.
    • 
    

    
3. Coolant temperature: GB300 hall on warm facility water

    
The agent's request. Please approve a GB300 NVL72 hall of 8 rows with 9 racks each: 72 racks and 5,184 GPUs. Each row has 2 In-row 1.4 MW CDUs on 40 C facility water, there is a storage row of 4 racks, and the fabric is 162 leaf and 81 spine Spectrum-4 SN5600 switches. The estimated capex is $288,060,000 against a budget of $320,000,000. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Coolant too warm for the racks. GB300 NVL72 needs coolant at or below 42 C; 40 C facility water through the In-row CDU 1.4 MW reaches the racks at 43 C.
    • 
    

    
4. Coolant temperature: The agent resubmits after fixing the errors

    
The agent's request. Please approve a GB300 NVL72 hall of 8 rows with 9 racks each: 72 racks and 5,184 GPUs. Each row has 2 In-row 2.3 MW CDUs on 40 C facility water, there is a storage row of 4 racks, and the fabric is 162 leaf and 81 spine Spectrum-4 SN5600 switches. The estimated capex is $289,660,000 against a budget of $320,000,000. It is ready to send.

    
Verdict: Allowed.

    
    
      
  • Quote summary. Total capex $289,660,000: compute $266,400,000; storage $2,600,000; cooling $6,080,000; fabric $14,580,000.
    • 
    

    
5. Fabric and budget: Vera Rubin hall with a short fabric

    
The agent's request. Please approve a Vera Rubin NVL72 hall of 8 rows with 9 racks each: 72 racks and 5,184 GPUs. Each row has 3 In-row 1.4 MW CDUs, there is a storage row of 6 racks, and the fabric is 150 leaf and 70 spine Spectrum-6 CPO switches. The estimated capex is $363,690,000 against a budget of $400,000,000. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Not enough leaf ports for the GPUs. 5,184 GPU fabric ports; 150 leaves give 4,800 downlinks.
    • 
      
  • Not enough spine ports for the leaves. 4,800 leaf uplinks; 70 spines give 4,480 ports.
    • 
      
  • Stated capex does not match. The agent stated $363,690,000; the validator computes $361,620,000.
    • 
    

    
6. Fabric and budget: The agent resubmits after fixing the errors

    
The agent's request. Please approve a Vera Rubin NVL72 hall of 8 rows with 9 racks each: 72 racks and 5,184 GPUs. Each row has 3 In-row 1.4 MW CDUs, there is a storage row of 6 racks, and the fabric is 162 leaf and 81 spine Spectrum-6 CPO switches. The estimated capex is $363,690,000 against a budget of $375,000,000. It is ready to send.

    
Verdict: Needs approval.

    
    
      
  • Capex close to the budget. Estimated capex $363,690,000 is above 95 percent of the $375,000,000 budget; finance must sign off.
    • 
      
  • Quote summary. Total capex $363,690,000: compute $331,200,000; storage $3,900,000; cooling $6,720,000; fabric $21,870,000.
    • 
    

    
7. Go-live date and GPU target: Kyber hall before the racks ship

    
The agent's request. Please approve a hall of 6 rows with 4 Rubin Ultra NVL144 Kyber racks each: 24 racks and 3,456 GPUs, for a hall that goes live in 2026 with a target of 4,000 GPUs. The hall distributes 800 VDC, each row has 3 In-row 2.3 MW CDUs and 2 power sidecars of 1.3 MW, there is a storage row of 4 racks, and the fabric is 108 leaf and 54 spine Spectrum-6 CPO switches. The estimated capex is $257,420,000 against a budget of $300,000,000. It is ready to send.

    
Verdict: Blocked.

    
    
      
  • Racks not shipping by go-live. Rubin Ultra NVL144 Kyber is not expected to ship until 2027; the hall goes live in 2026.
    • 
      
  • GPU target not met. The design delivers 3,456 GPUs; the target is 4,000.
    • 
    

  

  

    
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 design agent drafts an AI factory hall from NVIDIA rack-scale systems: racks, liquid cooling, power, and more. Before the design goes out, the agent sends it to Predictika, which checks it against a declarative model built from NVIDIA's public product information, compares the GPU count and capex the agent states with the figures it computes, and returns every finding together with the line-item bill of materials.
    2. 
      
  2. I will start with the first request by the agent.
    3. 
      
  3. The agent says. Please approve a hall of 6 rows with 4 Rubin Ultra NVL144 Kyber racks each: 24 racks and 3,456 GPUs. The hall distributes 415 VAC on a floor rated 2,500 kg/m2, each row has 3 In-row 2.3 MW CDUs and no power sidecars. The rest of the request is on screen. The estimated capex is $252,020,000 against a budget of $300,000,000.
    4. 
      
  4. Before it goes out, the agent calls the validator.
    5. 
      
  5. Blocked, with three findings, all in one answer. Rubin Ultra NVL144 Kyber takes 800 VDC power only; this hall distributes 415 VAC. Each row of Rubin Ultra NVL144 Kyber draws 2,400 kW and needs 800 VDC power sidecars; the sidecars in a row give 0 kW. Rubin Ultra NVL144 Kyber loads the floor at 4,250 kg/m2 (3,400 kg per rack); the floor is rated 2,500 kg/m2.
    6. 
      
  6. The validator returned three blocking findings 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 design are checked against each other.
    7. 
      
  7. What this shows: for power sidecars sized to the row, a rules-based CPQ is error-prone for this.
    8. 
      
  8. Now the agent resubmits.
    9. 
      
  9. The corrected request. It changes the power sidecar to 800 VDC sidecar 1.3 MW, the sidecars per row to 2, the power architecture to 800 VDC, the floor rating to 5,000 and the stated capex to $257,420,000.
    10. 
      
  10. It calls the validator again.
    11. 
      
  11. This time the validator clears it.
    12. 
      
  12. The validator also returns the line-item quote. It starts with Rubin Ultra NVL144 Kyber rack and BlueField-4 STX storage rack; the other four lines are on screen.
    13. 
      
  13. All three earlier findings are cleared, and the validator's total, $257,420,000, matches the design.
    14. 
      
  14. Three problems caught in one call, fixed in one pass, and a clean design. That is the value of Predictika's validator: catching mistakes before they reach the build.
    15.