Benchmark performance not appliedBenchmark performance is not applied until a model family or explicit comparison group is selected.
PERFORMANCE EVIDENCEComparable benchmark basis
Benchmark performance is not applied until a model family or explicit comparison group is selected.
Read the evidence methodology →No benchmark group is active. Market-price economics still work, but Kovara will not invent performance normalization.
LEADING CURRENT SCENARIOAWS
NVIDIA H200 SXM · London, United Kingdom
Partial economicsTracked GPU price$6.87USD / GPU-hour
Storage$0.00
Interconnect$0.00
Modeled totalIncompleteRun basis incomplete
Effective utilized GPU-hourIncomplete85% utilization
Workload cost / 1M tokensIncompleteToken volume missing
Performance $ / 1M tokensIncompletePerformance unavailable
Evidence score75/100Completeness + freshness
SCENARIO COMPARISONCurrent economic paths
Ranking objective: Balanced economics. Evidence score is a modeling completeness / freshness indicator, not a probability.
5 viable01

AWSNVIDIA H200 SXM · London, United KingdomOFF000010
GPU price$6.87
TotalIncomplete
EffectiveIncomplete
Perf $/1M tokIncomplete
Evidence75/100
Inspect →02

Crusoe CloudNVIDIA H200 SXM · Iceland, IcelandOFR-0663
GPU price$4.29
TotalIncomplete
EffectiveIncomplete
Perf $/1M tokIncomplete
Evidence75/100
Inspect →03

AWSNVIDIA H200 SXM · London, United KingdomOFR-0152
GPU price$7.89
TotalIncomplete
EffectiveIncomplete
Perf $/1M tokIncomplete
Evidence73/100
Inspect →04

CoreWeaveNVIDIA H200 SXM · Europe, Multiple / UnspecifiedOFR-0595
GPU price$6.30
TotalIncomplete
EffectiveIncomplete
Perf $/1M tokIncomplete
Evidence65/100
Inspect →05

CoreWeaveNVIDIA H200 SXM · Europe, Multiple / UnspecifiedOFR-0596
GPU price$2.58
TotalIncomplete
EffectiveIncomplete
Perf $/1M tokIncomplete
Evidence65/100
Inspect →EVIDENCE LAYERWhat is measured, modeled and missing
The economics model does not collapse these categories into a single claim.
OFF000010TRACKED- Comparable USD / GPU-hour price is tracked on the current public Offering.
- Accelerator VRAM is tracked at 141 GB per GPU.
- One-time setup fee is tracked at $0.
- Network bandwidth metadata is tracked at 3,200 Gbps.
- Local storage capacity is tracked at 30,720 GB; storage price is not inferred.
ASSUMPTIONS- Billed planning month is 730 hours unless the workload uses an explicit run duration.
- Utilization assumption is 85.0%.
- No egress volume was requested, so egress cost is excluded from the scenario.
- This scenario models compute commitments, requested egress, storage, setup, interconnect and entered overhead; taxes and unlisted charges require contract review.
MISSING / NOT INFERRED- Minimum term is undisclosed; no complete contractual total is claimed.
- Billed run duration is not specified or derivable.
- No applicable public comparable performance benchmark or explicit throughput assumption is available.