Intel and RadixArk Optimize SGLang for Open AI Inference

Intel and RadixArk are collaborating to optimize SGLang, an open-source AI inference stack, for Intel platforms. The companies say the work aims to help developers and enterprises deploy leading AI models and build AI applications more efficiently and cost-effectively.
- Intel and RadixArk are collaborating on SGLang optimization.
- SGLang is an open-source AI inference stack.
- The work targets Intel CPUs, GPUs and accelerators.
- Intel cites Intel AMX among the supported hardware acceleration features.
- The companies aim to improve flexibility and efficiency for enterprise AI deployment.
Optimization across Intel platforms
The collaboration covers Intel CPUs, GPUs and accelerators, with the companies targeting an open approach to AI inference software. Intel said the work is intended to give developers and enterprises more efficient options for deploying models across their infrastructure.
SGLang is an open-source inference stack. According to Intel, the collaboration will optimize it for Intel platforms and provide developers with software designed for enterprise production deployments. The company said this includes what it called “Day-0 readiness” and high-performance inference software directly optimized for its hardware.
Intel vice president of AI Products & GTM in the Intel Data Center Group, Anil Nanduri, said enterprise AI is moving beyond isolated performance benchmarks toward full-system production deployments. He said the collaboration gives developers freedom to choose the tools, frameworks and technologies that best fit their needs.
What developers get
Intel said the collaboration expands developer choice for deploying leading AI models on its enterprise-proven hardware. Model builders and software engineers can use standard, open frameworks and interfaces while taking advantage of Intel’s compute density and built-in hardware acceleration features, including Intel Advanced Matrix Extensions, or Intel AMX.
The company said developers can gain faster time-to-production, reduced latency and flexible deployment options across existing data center and cloud capacity. These are company claims about the expected benefits of the collaboration, rather than performance figures reported in the announcement.
RadixArk CEO Ying Sheng said the collaboration is a milestone for open-source AI inference. RadixArk said it will upstream deep optimizations into the core framework, which it described as a way to preserve open standards while improving performance and flexibility for development teams.
An open approach to production AI
Intel framed the partnership as part of an approach that combines hardware and software optimization through contributions to community software. Rather than treating hardware and software as separate deployment decisions, the company said upstream contributions to popular engines and frameworks can create high-performance paths for turning models into scalable, production-grade applications.
The announcement places the work in the context of increasingly complex AI models and heterogeneous enterprise infrastructure. Intel said production AI requires prompts to be routed to the right destination by matching workloads with suitable intelligence, compute and economics. It described open inference runtimes as connective tissue for making full-stack deployments practical, reliable and economically sustainable across diverse enterprise environments.
For users, the stated goal is broader choice when building and deploying AI applications on Intel infrastructure. The source does not provide a release date, specific model support list or measured performance results, so the immediate significance is the planned optimization and upstream software work rather than a quantified product launch.


