SC26: Infinitix to Debut at the Show With AI-Stack Intelligent Scheduling for HPC and AI Workloads

Infinitix, an AI infrastructure software company specializing in software-defined GPU technology, will make its SC26 debut Nov. 17-19 in Chicago. At booth 6111, under the theme “From AI Infrastructure to Token Factory,” the company will showcase new HPC scheduling capabilities in AI-Stack alongside its ixCSP AI cloud operations platform.AI-Stack Intelligent Scheduling Simplifies HPC Cluster Operations 
With Slurm on Kubernetes, research teams can run jobs in a cloud-native environment using familiar commands such as sbatch and srun. AI-Stack streamlines cluster deployment, coordinates multi-node jobs, and reduces resource fragmentation, so HPC, AI training, and inference workloads share a single GPU pool.

For distributed training, AI-Stack integrates DeepSpeed and Megatron, automatically failing over to standby nodes and resuming from the latest checkpoint. For large-model inference, its Prefill/Decode (PD) disaggregated architecture runs each stage on separate GPUs, supports llm-d, Dynamo, vLLM and SGLang, and uses KV cache reuse to maintain stability under high concurrency.

ixCSP Turns GPU Capacity Into Billable AI Services
ixCSP combines multi-tenant operations, service onboarding, metering and billing, supporting GPU-as-a-Service (GaaS), Model-as-a-Service (MaaS) and Token-as-a-Service (TaaS). Neocloud providers, data centers and telecom operators can use it to monetize the compute AI-Stack manages.

“AI-Stack lets research teams keep the Slurm workflows they know while bringing every GPU into a unified pool, and ixCSP turns that compute into billable AI services,” said Wenyu Chen, CEO, Infinitix. “SC26 marks an important step in our expansion into North America.”