NVIDIA CorporationScitiX's platform uses NVIDIA B200, H200, and H100 GPUs, indicating demand for NVIDIA hardware.

ScitiX has unveiled a production inference platform designed for enterprises running AI at scale, positioning inference as the operational core of modern AI stacks. The platform runs on ScitiX-owned NVIDIA B200, H200, and H100 infrastructure and currently processes over 1 trillion tokens daily with an average time-to-first-token of about one second, a cache hit rate exceeding 90%, and 99.9% uptime. It offers intelligent model routing, session-aware context reuse, fault-tolerant execution, private deployment environments, zero-retention policies, and full-stack observability. The platform is already supporting production workloads for RadixArk, the commercial team behind SGLang. ScitiX's internal evaluation framework, SiEval, demonstrated up to 10.5× acceleration on evaluation-heavy pipelines and 7.22× end-to-end speedups across large-scale leaderboard workflows. The platform is available now to enterprise customers.
NVIDIA CorporationScitiX's platform uses NVIDIA B200, H200, and H100 GPUs, indicating demand for NVIDIA hardware.
ScitiX launches a production-ready inference platform with performance improvements.
RadixArk's SGLang is supported on the platform, indicating adoption and demand.