Intel has partnered with RadixArk to optimise SGLang, an open-source AI inference framework, for Intel platforms. The two companies want developers and enterprises to deploy large AI models with better performance, lower latency and greater cost efficiency.
What the companies will do
They plan to build the optimisations directly into SGLang. Developers could then run inference workloads across Intel central processing units (CPUs), graphics processing units (GPUs) and accelerators. The work will focus on software compatibility and performance. It will keep the flexibility of open-source development.
The deal makes use of Intel hardware features such as Intel Advanced Matrix Extensions (AMX). These speed up the matrix operations behind AI and other compute-heavy tasks. Developers should be able to run leading AI models on Intel hardware using established open frameworks and interfaces.
From benchmarks to production
Intel said the partnership answers a growing need for production-ready AI infrastructure. Enterprises are moving from comparing single-model benchmarks to deploying whole AI systems.
Anil Nanduri, Vice President of AI Products and Go-to-Market at Intel’s Data Center Group, said enterprise AI now has to solve full-system production deployments. He said the collaboration would give developers Day-0 readiness and optimised inference software across Intel hardware. They would still choose the frameworks and technologies that suit them.
Why inference needs a coordinating layer
AI models are growing larger and more demanding. Enterprises need infrastructure that spreads work across different kinds of computing resources. Inference systems must match each request to the right model and hardware. They must also balance performance against running costs.
Open inference frameworks help coordinate this work across mixed computing environments. The Intel and RadixArk effort aims to strengthen that layer with platform-specific improvements in SGLang.
The companies said the optimisations could cut inference latency and shorten the time to production. They could also give organisations more ways to use existing data centre and cloud infrastructure.
Open source at the core
Ying Sheng, CEO of RadixArk, called the partnership a step towards making high-performance AI inference more accessible through open-source software. She said that by upstreaming deep optimisations into the core framework, the companies remove constraints. Developers should not have to choose between open standards and high performance.
The collaboration reflects Intel’s strategy of adding hardware optimisations to widely used open-source frameworks. It avoids confining them to proprietary software. Upstreaming the changes could reach a wider developer community. It could also simplify deployment across supported Intel platforms.
The companies did not disclose a timeline for the optimisations. They also gave no specific performance benchmarks.


