NEWS → · Inference software · LMSYS Org
NVIDIA, AMD, Intel, Google and Meta launch the SGLang Consortium to steer open-source inference
Nov 12, 2026
LMSYS Org launched the SGLang Consortium on Nov 12, 2026 at the first SGLang Summit in San Francisco, with NVIDIA, AMD, Intel, Google, Meta, Cloudflare, Modal and RadixArk as founding members. The group gives the companies that run the open-source inference engine at the largest scale a standing seat in its roadmap — while SGLang's maintainers keep technical decisions and code acceptance.
Image: LMSYS Org via lmsys.org
What was announced, from LMSYS Org's own post (Nov 12, 2026), read first-hand: each founding member designates a technical representative and joins two roadmap meetings a year, contributing engineering expertise, deployment feedback, testing resources or joint development, with working groups between meetings and public updates on direction. The shared work focuses on frontier-model support, reasoning and multimodal workloads, distributed inference and hardware optimisation. LMSYS says SGLang — part of the PyTorch ecosystem — runs on hundreds of thousands of GPUs and powers production inference for companies including Meta, Microsoft, Google, Amazon, Thinking Machines Lab, Cloudflare, Reflection and Modal, and cites Deloitte's estimate that inference will account for roughly two-thirds of all AI compute in 2026.
The quotes set the terms. Ying Sheng of RadixArk, for the SGLang team: “The teams running SGLang at the largest scale see what inference needs next before anyone else does.” NVIDIA's Ian Buck: “Better software helps developers get more performance from NVIDIA compute.” AMD's Ramine Roane pointed to SGLang running across hardware, including AMD Instinct GPUs. The consortium was announced in the opening keynote of the inaugural SGLang Summit, held Nov 12–13 at Fort Mason Center with more than 100 speakers.
What it means if you rent GPUs
The inference engine is where a rented GPU's money is won or lost: the same card serves more tokens, or fewer, depending on the software on top of it. SGLang is one of the two engines most open-weight deployments on rented hardware actually run, and it now has a formal industry body — chip makers included — steering its roadmap toward the frontier models and distributed setups renters serve. For builders the practical watch is what lands from it: day-one support for new weights drops (Inkling 2 Small, above, shipped with SGLang integration on day one) and hardware optimisation that shows up as throughput on the GPUs already on the board.