SGLang Consortium Launches With Nvidia, AMD, Google and Five More

What happened

The consortium was announced in the opening keynote of the inaugural SGLang Summit, held Nov 12–13 at the Fort Mason Center in San Francisco, the first conference dedicated to SGLang and its open-source community. Each founding member will name a technical representative and join two roadmap meetings a year.

SGLang is an open-source inference engine for running large language models and multimodal models in production, supporting Nvidia and AMD hardware, Intel Xeon processors and Google TPUs. LMSYS says it runs on hundreds of thousands of GPUs and is used in production by Meta, Microsoft, Google, Amazon, Thinking Machines Lab and Cloudflare, among others.

How the consortium works

The consortium gives members a standing forum to discuss emerging inference trends with SGLang’s maintainers and turn shared priorities into joint engineering work, focused on frontier model support, reasoning and multimodal workloads, distributed inference and hardware optimization. Project maintainers retain technical decision-making and code-acceptance authority.

LMSYS cites a Deloitte estimate that inference will account for roughly two-thirds of all AI compute in 2026, up from about half in 2025. The announcement gives no release timing, priorities or specific performance figures for individual features.

Key facts

1. Members: Eight founding members: Nvidia, AMD, Intel, Google, Meta, Cloudflare, Modal and RadixArk
2. Organizer: LMSYS Org, the nonprofit behind SGLang, announced it in the opening keynote of the inaugural SGLang Summit
3. Focus: Frontier model support, reasoning and multimodal workloads, distributed inference and hardware optimization
4. Governance: Each member names a technical representative and joins two roadmap meetings a year; maintainers keep technical and code-acceptance authority
5. Scale: LMSYS says SGLang runs on hundreds of thousands of GPUs, with Meta, Microsoft, Google and Amazon among production users
6. Context: Deloitte estimates inference will take roughly two-thirds of all AI compute in 2026, up from about half in 2025