What Anyscale does

Anyscale is a privately held AI infrastructure company founded in 2019 by the creators of Ray, the open-source distributed computing framework developed at UC Berkeley RISELab. The company provides a managed platform that enables organizations to build, run, and scale data-intensive AI workloads — including multimodal data curation, distributed model training, batch embedding generation, and post-training/RLHF — across thousands of GPU nodes without code changes. Its product portfolio spans the core Anyscale Platform, the RayTurbo optimized runtime (4.5x faster data processing, 90% lower cost), Ray Data for multimodal processing with NVIDIA cuDF GPU-native integration, Ray Train for distributed training, Ray Serve for LLM inference with HAProxy and gRPC, Ray Tune for hyperparameter optimization, and the newer Agent Skills and AnySearch products. Ray itself was transferred to the PyTorch Foundation in October 2025 alongside PyTorch and vLLM, cementing its status as an industry-standard AI compute engine with 41K+ GitHub stars, 500M+ downloads, and 1.2K+ contributors.

Anyscale monetizes through a hybrid pricing model: usage-based billing on its Hosted deployment (with hourly GPU rates from $0.0135/hr for CPU to $4.9591/hr for A100, plus H/B/GB families via sales), Bring Your Own Cloud (BYOC) subscriptions invoiced via Anyscale or cloud marketplaces, and committed contracts with volume discounts. The go-to-market combines product-led growth (a $100 free credit, code templates, and console self-serve) with enterprise field sales supported by Field CTOs and Forward Deployed Engineers. Distribution is anchored by native integrations and marketplace listings across all major hyperscalers: AWS SageMaker HyperPod, Microsoft Azure AKS (with a native offering launching 2026), Google Cloud GKE, Nebius, CoreWeave, and Oracle. The company serves more than 20 named enterprise customers spanning foundation model labs (OpenAI, Figure AI, Physical Intelligence), consumer platforms (Uber, Canva, Character.ai, Notion, Tripadvisor), financial services (Coinbase, Adyen, Handshake), media (Runway, TwelveLabs), autonomous systems (Wayve, Bonsai, Bedrock Robotics), and biotech (Recursion, Multiply Labs). Headquartered in San Francisco with offices in Palo Alto and Bangalore, Anyscale has raised over $260 million in total funding at a $1 billion valuation and reached approximately $100M in ARR by 2026.

Name

Anyscale

Legal name

Anyscale, Inc.

Website

https://anyscale.com

Company type

Private

Founded year

2019

Operating status

Operating

Headcount range

251–500 employees

Short description

Anyscale provides a managed AI infrastructure platform built on Ray, the open-source distributed computing framework, enabling organizations to curate multimodal data, train foundation models, and serve LLMs at scale across AWS, Azure, GCP, and other clouds for enterprise and AI-native customers.

Ownership category

akta.pro rank

Product category

AI Infrastructure Platform

NAICS

Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (51821), Computer Systems Design Services (541512)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

End-to-End Enterprise AI Platforms (MLOps & Model Lifecycle Management) (HDAEANAA)

akta.pro secondary industries

AI Compute Virtualization & Scheduling (GPU virtualization, cluster schedulers) (HDAAAAAG), Model Deployment, Serving & Inference Platforms (HDAAABAF), AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI), AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ)

Keywords

• Distributed AI computing
• AI infrastructure platform
• Foundation model development
• GPU workload orchestration
• Managed Ray platform

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices3 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations

Revenue model

1. Managed Ray Platform (Hosted): Fully managed infrastructure with no setup required. Pay-as-you-go pricing based on compute usage with hourly rates for different GPU types (T4, L4, A10G, A100, H/B/GB families). Monthly credit card invoices.
2. Bring Your Own Cloud (BYOC): Deploy inside customer's own cloud VPC (AWS, Azure, GCP) or on-prem. Uses existing GPU reservations. Invoice via Anyscale or cloud marketplace. Includes enterprise SLAs with 24x7 coverage.
3. Committed Contracts: Volume discounts and benefits that scale with usage. Reserved capacity for committed customers with additional discounts as usage grows.

Pricing tiers

Go-to-market motion3 records

Distribution channels6 records

Marketing channels9 records

Core offering

Anyscale provides a unified production-scale AI compute platform built on Ray, the open-source distributed computing framework the company's founders created at UC Berkeley RISELab. The platform enables organizations to develop, train, deploy, and scale data-intensive AI workloads including multimodal data curation, distributed model training, batch embedding generation, and LLM inference across multi-cloud GPU infrastructure. It offers RayTurbo optimized runtime, Ray Data, Ray Train, Ray Serve, Ray Tune sub-products, plus deployment options of fully managed Hosted or BYOC (Bring Your Own Cloud) on AWS, Azure, GCP, Nebius, Oracle, and CoreWeave.

Product overview

Anyscale is a unified AI infrastructure platform built on Ray, the open-source distributed computing framework created by the company's founders at UC Berkeley. The platform provides a comprehensive portfolio of products including: the core Anyscale Platform for building and running AI workloads at scale; RayTurbo as an optimized runtime with up to 4.5x faster performance; Ray Data for multimodal and unstructured data processing; Ray Train for distributed model training; Ray Serve for model serving and online inference; and Ray Tune for hyperparameter optimization. The platform also offers Agent Skills for AI-native workload management and AnySearch for AI agent search infrastructure. Deployment options include Hosted (fully managed) and BYOC (Bring Your Own Cloud) for enterprise customers wanting to run within their own cloud tenancy. Anyscale runs on AWS, Azure, GCP, Nebius, Oracle, and other cloud providers, serving customers including OpenAI, Uber, Canva, Coinbase, Runway, and Character.ai.

Differentiator

Problem solved

Functional benefit

Brands

• Ray: Open source distributed computing framework for AI and Python workloads, developed at UC Berkeley RISELab and transferred to PyTorch Foundation in 2025.
• RayTurbo
• Anyscale on Azure

Products and services

• Anyscale Platform Unified AI platform built on Ray that enables teams to build and run AI workloads at production-scale with speed, reliability, and cost-efficiency. Provides developer tooling, workload-aware observability, cluster orchestration, and multi-cloud deployment across AWS, GCP, Azure, Nebius, and Oracle for AI/ML engineering teams and foundation model builders.
• Ray Open-source distributed computing framework for AI and Python workloads, originally developed at UC Berkeley RISELab by Anyscale's founders and transferred to the PyTorch Foundation in October 2025. World's most widely adopted AI compute engine with 500M+ all-time downloads, 41K+ GitHub stars, and 1.2K+ contributors, used by developers building distributed AI applications.
• RayTurbo Optimized runtime for Ray delivering up to 4.5x faster data processing and 90% lower costs compared to standard Ray. Part of the GPU-native architecture shift announced at Ray Summit 2024 for AI workload customers seeking price-performance optimization.
• Ray Data Scalable structured and unstructured data processing library supporting multimodal data including images, video, audio, and text. Includes GPU-native support via NVIDIA cuDF integration for foundation model data pipelines. Targets AI engineering teams curating massive multimodal datasets.
• Ray Train Ray library for distributed model training that scales PyTorch, XGBoost, Hugging Face, JAX, or TensorFlow model training across nodes with GPU observability and persistent logs. Designed for ML engineering teams training foundation models and LLMs at scale.
• Ray Serve Ray library for scalable model serving and online inference supporting LLM inference with vLLM integration, multi-model deployments, and composite AI serving with A/B rollouts. Delivers 88% lower P99 latency and 11.1x throughput improvement for LLM inference workloads.
• Ray Tune Ray library for distributed hyperparameter optimization enabling efficient HPO trials with shared data preprocessing across all trials. Used by ML engineering teams tuning foundation models and other workloads.
• Agent Skills AI-native toolset integrating with AI coding tools like Claude Code and Cursor to help developers write, deploy, debug, and optimize large-scale AI workloads on Ray. Includes Workload Skills, Platform Skills, and Infra Skills modules targeting developer productivity.
• AnySearch Next-generation AI search infrastructure providing AI agents with unified access to high-quality structured data from authenticated professional systems including financial terminals, code repositories, and academic platforms. Aggregates data across finance, legal, academic research, cybersecurity, energy, and corporate intelligence domains through a single unified API with 1,000 free API calls per day.
• Anyscale on Azure Native integration with Microsoft Azure built on Azure Kubernetes Service enabling enterprises to run production-scale AI workloads entirely within their own Azure tenancy with up to 90% cost savings compared to fragmented stacks. Targets enterprises requiring Azure-sovereign AI infrastructure.

Quantifiable outcome

• Up to 90% cost savings compared to fragmented stacks
• +7 more outcomes

Named customers21 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration4 records

AI capability13 records

Feature8 records

Partnerships

Ten partnerships are on record, tiered core and minor.

• Microsoft AzurecoreTechnology or Integration · 2 June 2026Native integration of Anyscale on Azure Kubernetes Service enabling enterprises to run foundation-model-scale AI workloads entirely within their own Azure tenancy. Full integration with Azure governance and billing systems. Public preview launched June 2026 with general availability expected 2026.
• NVIDIAcoreTechnology or Integration · 17 March 2026Featured in NVIDIA CEO Jensen Huang's list of 103 AI-native companies at GTC 2026 conference as evidence of platform shift comparable to PC and internet eras.
• NebiuscoreTechnology or Integration · 11 November 2025Partnership to integrate Nebius AI Cloud with Anyscale's managed Ray platform for cost-efficient multimodal and physical AI workloads. End-to-end scalable infrastructure addressing heterogeneous data stream processing.
• MicrosoftcoreTechnology or Integration · 4 November 2025Strategic collaboration to deliver AI-native compute service on Azure built with Ray. Private preview launched November 2025 with general availability expected 2026. Joint go-to-market with technical integration guides published.
• Google CloudcoreTechnology or Integration · 4 November 2025Ray and Kubernetes integration on Google Kubernetes Engine (GKE) with Label Selector API in Ray v2.49. Enables GPU binding with fallback behavior, zone/region targeting, and custom topologies for distributed AI workloads.
• NVIDIAcoreTechnology or Integration · 4 November 2025Integration of Ray Data with NVIDIA cuDF for GPU-native multimodal data processing. Rack-aware scheduling for NVIDIA GB300 NVL72 clusters. Joint optimization for Blackwell architecture.
• PyTorch FoundationcoreStrategic or Co-development Partner · 22 October 2025Ray open-source project transferred to PyTorch Foundation to become a neutral industry standard. Joins PyTorch and vLLM in unified open source AI compute stack. Validates Ray as industry standard for distributed AI infrastructure.
• Amazon Web ServicescoreTechnology or Integration · 10 October 2025Integration with Amazon SageMaker HyperPod for next-generation distributed computing. Joint technical guide for scaling AI/ML workloads on AWS.
• Google CloudcoreTechnology or Integration · 9 April 2025Integration of RayTurbo with GKE for optimized Ray runtime on Google Kubernetes Engine. Enhances AI scheduling and scaling capabilities.
• AmazonminorChannel Partner/ Reseller/ DistributorAWS Marketplace listing enabling customers to discover and purchase Anyscale through AWS marketplace channel.

Scale indicators15 records

Recent moves2 records

Expansion highlights7 records

Direct peers

• Databricks: Unified data analytics and AI platform that, like Anyscale, offers distributed compute (Spark, Ray integration via Anyscale partner) plus ML lifecycle tooling (MLflow, MosaicML). Competes directly for the same enterprise AI/ML workloads.

Broad incumbents

• AWS SageMaker: AWS's managed ML platform offering distributed training, inference, and MLOps. While Anyscale integrates via SageMaker HyperPod, SageMaker competes broadly for the same enterprise training and serving budgets.
• Google Vertex AI: Google Cloud's end-to-end AI platform spanning data, training, and serving. Anyscale's GKE integration is complementary, but Vertex AI competes for end-to-end AI infrastructure spend in Google Cloud-native enterprises.
• Azure Machine Learning: Microsoft's enterprise AI platform. Despite the deep Anyscale-Azure partnership, Azure ML competes for production AI workload budgets within Microsoft-centric enterprises and benefits from native Azure governance.
• Snowflake: Cloud data platform expanding into AI/ML workloads with Cortex AI and ML functions. Anyscale lists Snowflake as an integration partner, but Snowflake also competes for the same enterprise AI-platform budget, especially in data-heavy workloads.

Emerging players

• Modal Labs: Serverless AI compute platform enabling developers to run batch inference, training, and fine-tuning on GPUs without infrastructure management. Comparable developer-experience positioning vs. Anyscale's hosted tier.
• Together AI: AI cloud platform purpose-built for training, fine-tuning, and serving open-source foundation models with optimized inference infrastructure. Competes with Anyscale for foundation-model developers seeking managed GPU compute.
• Fireworks AI: AI inference platform offering low-latency, cost-optimized model serving using custom GPU optimizations. Overlaps with Anyscale's Ray Serve / inference layer for production LLM deployments.
• CoreWeave: Hyperscale GPU cloud purpose-built for AI workloads, listed as one of Anyscale's multi-cloud deployment targets. Competes indirectly by offering bare-metal GPU capacity that customers might choose over Anyscale's managed layer.
• RunPod: GPU cloud platform offering on-demand and serverless GPU compute for AI training and inference. Targets the same developer-led, GPU-intensive workloads as Anyscale's hosted tier at typically lower price points.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks5 records

Key highlights7 records

Customer concentration

• Xx.com/anyscalecompute
• GitHubgithub.com/ray-project
• LinkedInlinkedin.com/company/joinanyscale
• Facebookfacebook.com/AnyscaleCompute

Revenue estimate

Valuation estimate

Number of profiles

Profiles8 records

Subsidiaries1 record

Funding overview

Funding rounds4 records

Investors10 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Anyscale