Sources: Project SGLang Spins Out as RadixArk with $400M Valuation as Inference Market Explodes

Why It Matters

Optimizing inference dramatically reduces operational AI expenses, making scalable deployments financially viable and attracting heavy venture capital.

Key Takeaways

• RadixArk valued at $400 million after Accel-led round
• SGLang accelerates AI inference, cutting hardware costs
• Founder Ying Sheng left xAI to lead RadixArk
• Inference startups attract billions in recent funding
• RadixArk launches paid hosting and Miles RL framework

Pulse Analysis

Open‑source AI projects are increasingly becoming commercial ventures, and the latest illustration is RadixArk, the company built around SGLang. Originating from Ion Stoica’s UC Berkeley lab in 2023, SGLang speeds up model inference and reduces compute spend. In a funding round led by Accel, RadixArk was placed at a $400 million valuation, a striking figure for a startup announced only last August. The capital infusion follows earlier angel backing from Intel’s CEO Lip‑Bu Tan and positions the firm to monetize its open‑source engine while expanding services.

The inference layer has emerged as a cost‑center for AI providers, often eclipsing training expenses in ongoing operations. Tools like SGLang and its peer vLLM promise immediate savings by squeezing more performance out of existing hardware. This economic incentive has sparked a wave of venture money: Baseten secured $300 million at a $5 billion valuation, Fireworks AI raised $250 million, and vLLM is rumored to chase a $1 billion valuation. Such funding underscores investors’ belief that optimizing inference will be a decisive competitive advantage in the AI race.

For enterprises, the shift means access to plug‑and‑play inference acceleration without deep engineering effort. RadixArk’s new “Miles” framework extends the value proposition into reinforcement‑learning workloads, enabling continuous model improvement on‑premise or in the cloud. By introducing paid hosting tiers, the company signals a hybrid model that balances open‑source community contributions with sustainable revenue. As AI adoption scales, the market for specialized inference infrastructure is set to expand, and companies that can deliver measurable cost reductions while maintaining flexibility will likely capture the bulk of future spend.