@wey_gu: It's such a pleasure to hear him speak — Zhu Banghua: SGLang, Reinforcement Learning, NVIDIA Acquisition, Second Startup, Tsinghua, Berkeley, LMSYS, Chatbot Arena, Good at Letting Go
Summary
Sharing an in-depth interview with SGLang/RadixArk CTO Zhu Banghua, covering his journey from Tsinghua to Berkeley, founding NexusFlow (later acquired by NVIDIA), and his second startup RadixArk, touching on topics like SGLang, reinforcement learning, and the NVIDIA acquisition.
View Cached Full Text
Cached at: 05/23/26, 06:05 AM
His talk is incredibly satisfying to listen to
— Zhu Banghua: SGLang, Reinforcement Learning, NVIDIA Acquisition, Second Entrepreneurship, Tsinghua, Berkeley, LMSYS, Chatbot Arena, Good at Knowing When to Quit https://t.co/BpS4MYGnDk
@wey_gu: His talk is incredibly satisfying to listen to — Zhu Banghua: SGLang, Reinforcement Learning, NVIDIA Acquisition, Second Entrepreneurship, Tsinghua, Berkeley, LMSYS, Chatbot Arena, Good at Knowing When to Quit
Channel: @wey_gu Source: https://youtu.be/VoAaX02cHFE?si=3HIDIEFTdKr2xLkc
Description
Today’s conversation is with Zhu Banghua, CTO of RadixArk / SGLang.
Zhu Banghua was the top student in his undergraduate class at Tsinghua University. He later studied at Berkeley under Jiantao Jiao and Michael Jordan.
After his PhD, he started his first company, NexusFlow. In the open-source community, he was one of the first to get PPO training working end-to-end. Soon after, the company was acquired by NVIDIA.
Less than six months after the acquisition, he walked away from the earn-out package and started another company with his close friend Sheng Ying, building RadixArk on top of SGLang. They raised a $100 million round of funding.
Congratulations to Banghua. Today, let’s dive deep into the story behind it all.
Similar Articles
@seclink: https://x.com/seclink/status/2056711091129118741
In-depth interview with Jensen Huang, reviewing Nvidia's history from betting the company on CUDA to becoming the AI powerhouse, explaining the four scaling laws of AI and the development direction for the next decade, emphasizing compute bottlenecks and extreme co-design philosophy.
@Valley101_Qian: Congrats to Yuandong @tydsh. At the end of our previous interview, the "new direction" he mentioned was officially announced today: neolab Recursive_SI, with $650 million in funding and a valuation of $4.65 billion. Looking forward to more research freedom and research taste in the industry...
After being laid off from Meta, Yuandong announced a new direction, raising $650 million to found neolab Recursive_SI with a valuation of $4.65 billion. In an interview, he shared insights on AI trends, LLM limitations, reinforcement learning, and research freedom.
@h100envy: Ying Sheng co-wrote SGLang, the inference engine now serving Grok at xAI on a hundred thousand GPUs. She also built Fle…
Ying Sheng co-wrote SGLang, the inference engine now serving Grok at xAI on a hundred thousand GPUs, achieving 5x cost cuts over DeepSeek's API; she also built FlexGen and helped build Chatbot Arena.
@MaxForAI: According to SemiAnalysis @SemiAnalysis_, Jia Yangqing, former Facebook (now Meta) AI Architecture Director, Alibaba Vice President of Technology, President of Alibaba Cloud Intelligent Computing Platform, now founder and CEO of LeptonAI, and NVIDIA Vice President of System Software, @jiayq, has left NVIDIA.
Jia Yangqing, former Facebook AI Architecture Director and LeptonAI founder, left NVIDIA one year after NVIDIA acquired his startup team. According to SemiAnalysis, reasons include DGX Lepton's performance not meeting expectations and disagreements with Jensen Huang over open-source commitments.
@charles_irl: Many are belatedly realizing that intelligence must be open. For open intelligence to succeed, developers must work tog…
A collaboration between Modal, SGLang, and Z Lab integrates DFlash speculation into SGLang, achieving up to 4.3x throughput improvement for Alibaba's Qwen 397B-A17B model, advancing open intelligence.