Sources: project SGLang spins out as RadixArk with $400M valuation as inference market explodes

A significant shift in the AI landscape has occurred as a portion of the team behind SGLang, an acclaimed open-source tool utilized by firms such as xAI and Cursor, has launched a new venture named RadixArk. Announced last August, RadixArk has recently achieved a valuation of approximately $400 million following a funding round led by Accel, as reported by sources close to the situation. However, TechCrunch has yet to verify the exact funding amount. In its early stages, RadixArk was developed within the UC Berkeley lab of Ion Stoica, co-founder of Databricks. The startup previously attracted angel investments from notable figures, including Intel CEO Lip-Bu Tan. Ying Sheng, an influential contributor to SGLang and a former engineer at xAI, has taken the helm as co-founder and CEO of RadixArk, transitioning from her role at Elon Musk’s AI company. Both SGLang and RadixArk are dedicated to enhancing inference processing, which is crucial for enabling AI models to operate with greater speed and efficiency on existing hardware. This optimization is vital, as model training and inference constitute a significant portion of server expenses linked to AI services. Consequently, advancements in this area can lead to substantial cost savings for businesses almost immediately. In a related development, vLLM, another initiative focused on optimizing inference, has transitioned from an open-source project to a startup. Reports indicated that vLLM is exploring a funding round potentially exceeding $160 million at a valuation nearing $1 billion, with Andreessen Horowitz positioned to lead this investment, although final details remain unconfirmed. vLLM co-founder Simon Mo described reports on this funding as 'factually inaccurate' without specifying the inaccuracies. Both SGLang and vLLM share their origins in Ion Stoica’s lab, with the professor being a prominent figure in the startup ecosystem. Several major corporations currently leverage vLLM for their inference workloads, while SGLang has experienced notable growth over the past six months, according to Brittany Walker, a general partner at CRV. RadixArk is committed to advancing SGLang as an open-source AI model engine while also developing Miles, a framework aimed at enhancing reinforcement learning capabilities. Although many of its offerings remain free, RadixArk has begun implementing fees for certain hosting services. The surge in funding for startups focused on inference infrastructure highlights the growing significance of this layer within the AI sector. For instance, Baseten recently secured $300 million, achieving a valuation of $5 billion, while competitor Fireworks AI raised $250 million with a $4 billion valuation last October.