AI-Stack Integrates Slurm on Kubernetes and Prefill/Decode Disaggregated Inference; ixCSP Accelerates the Shift to Compute-as-a-Service.
INFINITIX, a software-defined GPU and AI infrastructure company, will make its SC26 debut Nov. 17–19 in Chicago. At Booth 6111, under the theme “From AI Infrastructure to Token Factory,” it will showcase new HPC scheduling in AI-Stack and its ixCSP AI cloud platform.
Simplifying HPC Cluster Operations
With Slurm on Kubernetes, research teams run cloud-native jobs using familiar sbatch and srun commands. AI-Stack streamlines cluster deployment, coordinates multi-node jobs, and reduces fragmentation, letting HPC, AI training, and inference share one GPU pool.
AI-Stack integrates DeepSpeed and Megatron for distributed training, with automatic failover and checkpoint recovery. Its Prefill/Decode (PD) disaggregated inference architecture runs each stage on separate GPUs, supports llm-d, Dynamo, vLLM, and SGLang, and reuses KV cache for stability under high concurrency.
Turning GPU Capacity Into Billable AI Services
ixCSP combines multi-tenant operations, service onboarding, metering and billing for GPU-as-a-Service (GaaS), Model-as-a-Service (MaaS) and Token-as-a-Service (TaaS), helping neo cloud providers, data centers and telecom operators monetize AI-Stack-managed compute.
“AI-Stack lets research teams keep their Slurm workflows while bringing every GPU into a unified pool, and ixCSP turns that compute into billable AI services,” said Wenyu Chen, CEO of INFINITIX. “SC26 marks an important step in our North American expansion.”
Visit Booth 6111 or book a meeting at infinitix.ai/en/news-detail/supercomputing2026.
The post INFINITIX to Debut AI Scheduling for HPC and AI Workloads at SC26 first appeared on AI-Tech Park.
