RadixArk
RadixArk builds open-source AI infrastructure for inference and reinforcement learning post-training, centered on SGLang and Miles, and sells paid managed cloud hosting and enterprise tooling to developers, startups, enterprises, and research labs.
- Company typePrivate
- Founded2024
- HeadquartersSan Francisco, United States
- Headcount11–50
- GTM typeB2B and B2C
- OfferingSoftware
What RadixArk does
RadixArk (RadixArk Inc.) is a San Francisco-based AI infrastructure company that builds open-source software for large-scale model inference and reinforcement learning post-training. The company's two core open-source projects are SGLang, an inference engine that functions as a middle layer between AI models and GPU hardware to reduce memory overhead and accelerate serving, and Miles, a framework that applies serving-engine rigor to large-scale RL post-training. SGLang originated in UC Berkeley's Ion Stoica lab in 2023 and is currently deployed across more than 400,000 GPUs, generating trillions of tokens daily for frontier AI labs including Google, Microsoft, NVIDIA, Oracle, xAI, and Thinking Machines Lab.
RadixArk operates an open-core business model. SGLang and Miles are distributed free of charge under open-source licenses, which drives community adoption and ecosystem lock-in. The paid layer consists of managed AI infrastructure and cloud-based model hosting (accessed via platform.radixark.com) and enterprise-grade tooling layered on top of Miles, both sold to enterprises and research labs that need production-ready deployment without building infrastructure in-house. The company was spun out of UC Berkeley in August 2024, launched publicly with $100 million in seed funding at a $400 million post-money valuation, and is led by co-founder and CEO Ying Sheng, a former xAI engineer and Databricks research scientist.
The company's stated long-term objective is to make frontier-level AI infrastructure at least 10x cheaper and 10x more accessible than current alternatives, by treating systems engineering as a first-class discipline rather than as a support function. Its GTM motion combines a self-serve, product-led growth approach (open-source downloads, platform sign-ups, GitHub community) for developers and startups with direct enterprise sales for managed-services contracts. Strategic capital from NVentures (NVIDIA), AMD, and MediaTek positions RadixArk as a hardware-vendor-agnostic infrastructure layer across the major GPU ecosystems.
RadixArk firmographics
Firmographics- Name
- RadixArk
- Legal name
- RadixArk Inc.
- Website
- https://radixark.ai
- Company type
- Private
- Founded year
- 2024
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- RadixArk builds open-source AI infrastructure for inference and reinforcement learning post-training, centered on SGLang and Miles, and sells paid managed cloud hosting and enterprise tooling to developers, startups, enterprises, and research labs.
- Ownership category
- akta.pro rank
RadixArk industry classification
Industry- Product category
- AI Inference and Training Infrastructure Software
- NAICS
- Software Publishers (5132), Computer Systems Design and Related Services (5415), Computer Systems Design and Related Services (54151)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)
- akta.pro primary industry
- Model Deployment, Serving & Inference Platforms (HDAAABAF)
- akta.pro secondary industries
- Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC), Model Hosting, Serving & Inference Platforms (HDAAACAB), AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI), Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM)
Keywords
Where RadixArk is headquartered
LocationHeadquarters
- HQ city
- San Francisco
- HQ country
- United States
- HQ region
- North America
Markets served
RadixArk business model
Business model- GTM type
- B2B and B2C
- Offering type
- Software
- Cost components
- Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations
Revenue model
- Managed Hosting Services: Premium managed infrastructure and hosting services for cloud-based model deployment, offering production-ready AI infrastructure without the need for customers to manage their own systems.
- Enterprise Tools (Miles): Enterprise-grade tools and frameworks like the Miles reinforcement learning framework offered as paid commercial products alongside the free open-source version.
Go-to-market motion1 record
Distribution channels3 records
Marketing channels4 records
RadixArk product offering
Product offeringCore offering
RadixArk develops and operates open-source AI infrastructure for inference and training. Its core open-source products are SGLang, an inference engine for serving modern large language models that sits between models and hardware to reduce memory overhead and lower inference costs, and Miles, an open-source reinforcement learning framework for large-scale post-training. On top of these open cores, the company offers managed AI infrastructure and enterprise tooling that allow developers, startups, enterprises, and research labs to deploy and run production AI applications without managing their own GPU clusters.
Product overview
RadixArk is an infrastructure-first AI company operating an open-core business model. It offers a unified platform centered on two open-source projects: SGLang (an inference engine for serving modern language models at scale) and Miles (a reinforcement learning framework for large-scale post-training). On top of these open-source cores, RadixArk commercializes managed AI infrastructure and enterprise tooling for cloud-based model hosting and RL training. The open-source engines remain free, while managed hosting services and enterprise tools provide the revenue layer.
Differentiator
Problem solved
Functional benefit
Products and services
- SGLang Open-source inference engine for serving modern large language models; functions as a middle layer between AI models and hardware to reduce memory overhead, lower inference costs, and optimize workloads across GPU clusters exceeding 400,000 graphics cards. Used by Google, Microsoft, NVIDIA, Oracle, xAI, Thinking Machines Lab, and Cursor.
- Miles Open-source reinforcement learning framework for large-scale post-training, designed to bring the same rigor to RL training that modern serving engines brought to inference. Developed alongside an enterprise-grade RL toolset for commercial customers.
- Managed AI Infrastructure Cloud-based managed hosting services and enterprise tooling built on top of SGLang and Miles, allowing developers, startups, enterprises, and research labs to deploy production AI applications without managing their own GPU clusters.
Quantifiable outcome
- Trillions of tokens generated daily using SGLang infrastructure
- +2 more outcomes
Companies that use RadixArk
Customer profileNamed customers7 records
Segments3 records
Ideal customer profiles3 records
RadixArk technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- No
- API docs
- API detail
Core technology
AI maturity
App detail
AI capability3 records
Feature3 records
RadixArk partnerships and signals
Strategic signalPartnerships
One partnership is on record.
- UC Berkeley (Ion Stoica Lab)coreSGLang originated from research conducted at UC Berkeley's Ion Stoica lab in 2023. The academic partnership provided the foundational research for RadixArk's core technology.
Scale indicators5 records
Recent moves6 records
Expansion highlights6 records
RadixArk competitors and assessment
Company assessmentBroad incumbents
- Hugging Face: Major open-source AI platform hosting models, datasets, and inference endpoints. Competes in model serving/hosting and commands the broader open-source model ecosystem in which SGLang operates.
- MosaicML (Databricks): Databricks' AI platform for training, fine-tuning, and deploying models at scale. Broader incumbent offering comparable capabilities in training and inference infrastructure.
Direct peers
- vLLM: Open-source high-throughput LLM inference engine with broad community adoption. Direct technical competitor to SGLang for serving modern language models on GPU clusters.
- Replicate: Cloud platform for running open-source AI models via API, with managed inference and serving infrastructure. Comparable open-model hosting and inference business model.
- Modal Labs: Developer platform for running AI and data workloads on serverless GPU infrastructure. Comparable as an infrastructure layer between developers/models and GPU hardware.
- DeepSpeed (Microsoft): Microsoft's open-source deep learning optimization library covering training and inference at scale. Adjacent competitor for large-scale training and serving infrastructure.
- Anyscale: Commercial provider of Ray-based AI compute platform for distributed training and inference. Competes in offering scalable AI infrastructure with managed services on top of open-source foundations.
- Fireworks AI: Managed inference platform optimized for low-latency LLM serving. Targets the same enterprise developer segment with hosted inference built on optimized inference runtimes.
- TensorRT-LLM (NVIDIA): NVIDIA's optimized LLM inference library built for its GPUs. Competes head-to-head with SGLang for performance leadership and is backed by an investor in RadixArk.
- Together AI: AI cloud platform offering inference and fine-tuning APIs built on top of open-source models and optimized inference stacks. Directly comparable open-core inference-as-a-service model.
Market position
Strengths5 records
Weaknesses5 records
Competitive moat6 records
Key risks5 records
Key highlights6 records
Customer concentration
RadixArk social profiles
Digital presenceRadixArk financial estimates
Financial estimateRevenue estimate
Valuation estimate
RadixArk leadership team
Management profileNumber of profiles
Profiles1 record
RadixArk funding detail
Funding detailFunding overview
Funding rounds1 record
Investors13 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
RadixArk M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about RadixArk
What does RadixArk do?
RadixArk develops and operates open-source AI infrastructure for inference and training. Its core open-source products are SGLang, an inference engine for serving modern large language models that sits between models and hardware to reduce memory overhead and lower inference costs, and Miles, an open-source reinforcement learning framework for large-scale post-training. On top of these open cores, the company offers managed AI infrastructure and enterprise tooling that allow developers, startups, enterprises, and research labs to deploy and run production AI applications without managing their own GPU clusters.
Is RadixArk a public or private company?
RadixArk is a private company. It is classified as venture growth investor backed and is currently operating.
When was RadixArk founded?
RadixArk was founded in 2024. It employs 11 to 50 people.
Where is RadixArk based?
RadixArk is headquartered in San Francisco, United States, in the North America region.
How does RadixArk make money?
Two revenue lines are on record. Managed Hosting Services are the primary driver. The others are enterprise Tools (Miles).
Who are RadixArk's main competitors?
Broad incumbents on record are Hugging Face and MosaicML (Databricks). Direct peers are vLLM, Replicate, Modal Labs, DeepSpeed (Microsoft), Anyscale, Fireworks AI, TensorRT-LLM (NVIDIA) and Together AI.
Does RadixArk have an API?
No public API is recorded for RadixArk.
What industry is RadixArk in?
RadixArk's product category is AI Inference and Training Infrastructure Software. Its primary akta.pro industry code is HDAAABAF, Model Deployment, Serving & Inference Platforms, with a secondary code of HDAEANAC, Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem). Its NAICS code is 5132 and its SIC code is 7372.