BRÈVE

sur WEKA

WEKA Introduces Inference Economics Scorecard to Refine AI Infrastructure Evaluation

WEKA, in collaboration with NAND Research, has launched the Inference Economics Scorecard, which aims to evaluate AI infrastructure based on productivity rather than GPU activity. The new framework, tested with Nebius on NVIDIA HGX B300 GPUs, offers nine performance metrics including tokens per watt and cost per inference. This approach addresses constraints in power, memory, and rack-space.

According to 451 Research, inference is expected to grow in AI infrastructure spending from 34% in 2026 to 46% by 2030. Traditional GPU utilization methods fall short for inference, as they do not account for task repetitions that constitute unproductive GPU cycles. This underscores the need for new measurements sensitive to sustained, efficient work over mere activity.

WEKA's recent benchmark results, deploying WEKA NeuralMesh, showed notable efficiency improvements, reinforcing the relevance of the new scorecard in planning AI infrastructure that accommodates contemporary agentic workloads efficiently.

R. E.

Copyright © 2026 FinanzWire, tous droits de reproduction et de représentation réservés.
Clause de non responsabilité : bien que puisées aux meilleures sources, les informations et analyses diffusées par FinanzWire sont fournies à titre indicatif et ne constituent en aucune manière une incitation à prendre position sur les marchés financiers.

Cliquez ici pour consulter le communiqué de presse ayant servi de base à la rédaction de cette brève

Voir toutes les actualités de WEKA