The inaugural summit brings together researchers and engineers to address the technical foundations of the next generation of AI infrastructure and model deployment.
LMSYS Org, the nonprofit research collective behind the SGLang framework, will host the inaugural SGLang Summit 2026 on November 12–13 at the Fort Mason Center for Arts & Culture in San Francisco. The event aims to bridge the gap between hardware-level optimization and high-level model deployment, drawing over 100 speakers and an expected audience of 2,000 practitioners.
The summit program emphasizes the integration of the full AI stack, focusing on the intersection of frontier model development and scalable infrastructure. Sessions are designed to address the operational realities of training and inference, with a specific focus on recursive self-improvement and AI agent architectures. Organizers have structured the agenda to facilitate knowledge transfer among researchers, engineers, and open-source contributors working on large-scale system deployments.
Ying Sheng, co-author of the foundational SGLang paper and chief executive officer of RadixArk, emphasized the necessity of cross-disciplinary collaboration in current AI development. According to Sheng, the summit provides a venue for teams to evaluate current technical bottlenecks and share empirical data on what methodologies yield performance gains in production environments. The event will feature 13 technical breakout tracks covering specialized topics such as reinforcement learning, local AI implementations, and generative media workflows.
The speaker lineup highlights the industry’s focus on hardware-software co-design, featuring Intel CEO Lip-Bu Tan and NVIDIA executives Luis Ceze and Ujval Kapasi. These discussions are expected to center on the next generation of AI hardware and the evolution of AI systems software. Additional technical insights will be provided by Soumith Chintala, chief technology officer at Thinking Machines Lab, who will address the current state of frontier model development.
OpenAI researcher Lilian Weng is scheduled to present on the mechanisms of recursive self-improvement, a critical area of study for future model autonomy. Perplexity co-founder and CEO Aravind Srinivas will discuss the implementation of AI agents, reflecting the industry shift toward systems that execute multi-step reasoning tasks. These sessions aim to move beyond theoretical models to examine the practical constraints of deploying autonomous systems at scale.
Hands-on workshops will allow participants to engage directly with open-source tools, including RadixArk’s Miles reinforcement learning framework. The inclusion of a GPU Expo provides a platform for attendees to observe the latest hardware advancements firsthand, offering a tangible look at the underlying compute resources driving modern model training. This combination of theoretical discourse and practical demonstration underscores the summit’s focus on the technical foundations of the next era of artificial intelligence.
The event reflects the broader industry trend toward open-source infrastructure as a primary driver of innovation. By bringing together engineers from organizations such as Google, Meta, and AWS, the summit seeks to standardize approaches to model serving and training. The participation of these entities highlights the reliance on open-source frameworks like SGLang to manage the increasing complexity of large language and multimodal models.
The significance of this gathering lies in its focus on the underlying mechanics of AI rather than consumer-facing applications. By prioritizing technical discourse on compilers, reinforcement learning, and hardware optimization, the summit addresses the fundamental challenges that currently limit model performance. The outcomes of these discussions are likely to influence the development of future AI infrastructure and the refinement of open-source standards.
Future developments in this space will likely hinge on the ability of researchers to solve the scaling challenges discussed during the summit. Industry observers will be monitoring the outcomes of the breakout sessions for new benchmarks in model serving efficiency and reinforcement learning methodologies. The summit serves as a critical milestone for the open-source community to align on the technical requirements for the next generation of AI systems, ensuring that infrastructure keeps pace with the rapid evolution of model architectures.
