Anyscale

Anyscale is an American AI-infrastructure software company headquartered in San Francisco, founded in 2019 by Robert Nishihara, Philipp Moritz, and Ion Stoica, the three core authors of the Ray open-source distributed-computing framework that originated in the UC Berkeley RISELab research group. Anyscale develops and commercializes the Ray framework alongside the Anyscale Platform, an enterprise platform for distributed AI training, inference, and reinforcement-learning workloads. Ray underpins the distributed-computing infrastructure at OpenAI (where it is used for ChatGPT and GPT model training), Uber, Shopify, ByteDance, and Spotify, with the open-source project reporting tens of thousands of GitHub stars and active research-community engagement. As of April 2026, Anyscale is one of the principal AI-infrastructure software companies, anchored on the Ray ecosystem and a stable founder-led leadership team.

At a glance

• Founded: 2019 in San Francisco by Robert Nishihara, Philipp Moritz, and Ion Stoica from the UC Berkeley RISELab research group.
• Status: Private. Series C of $99 million at $1 billion valuation in October 2021.
• Funding: Cumulative private capital exceeding $250 million across multiple rounds. Notable backers include Andreessen Horowitz, NEA, Foundation Capital, Intel Capital, and other investors.
• CEO: Keerti Melkote, Chief Executive Officer (since 2024). Former founder and senior executive at HPE Aruba.
• Other notable leadership: Robert Nishihara, Co-Founder (former CEO; transitioned to a research and strategy role). Philipp Moritz, Co-Founder. Ion Stoica, Co-Founder and UC Berkeley Computer Science professor (also Co-Founder of Databricks).
• Open weights: N/A. Anyscale develops infrastructure software, not foundation models.
• Flagship products: Ray (open-source distributed-computing framework); Anyscale Platform (managed Ray service for enterprise customers); Ray AI Runtime (RayServe, RayTrain, RayTune, RayData).

Origins

The Ray distributed-computing framework began as a UC Berkeley research project in 2016 within the RISELab research group led by Ion Stoica, Michael Jordan, Pieter Abbeel, and other Berkeley AI faculty. The project was motivated by the observation that emerging AI workloads (reinforcement learning, large-scale model training, distributed hyperparameter tuning) had distributed-computing requirements that existing frameworks like Apache Spark and TensorFlow were not architecturally designed to address. Ray's task-graph and actor-model abstractions were designed to support fine-grained, dynamic distributed-computing patterns that AI workloads required.

The Ray 1.0 release in September 2020 marked the framework's transition from research project to production-ready system, with adoption beginning at OpenAI (Ray is used for ChatGPT training and reinforcement-learning workloads), Uber (the Michelangelo AI platform), and Ant Group, alongside other early enterprise users. Anyscale was founded in 2019 with the explicit thesis that the Ray ecosystem would become the principal distributed-computing infrastructure for AI applications and that Anyscale's commercial role would be the managed-platform offering for enterprise customers.

The 2020 to 2024 period saw rapid Ray ecosystem adoption alongside Anyscale Platform commercial growth. The Series C of $99 million at $1 billion valuation in October 2021 provided growth-equity capital. The Ray Summit annual conference (launched in 2020) became one of the principal AI-infrastructure community events.

The 2024 leadership transition saw Robert Nishihara move from the CEO role into a research-and-strategy position, with Keerti Melkote (former founder of HPE Aruba's wireless-networking business) taking the CEO role. The transition was characterized in industry coverage as a structurally normal handoff from founder-led growth-stage management to a more enterprise-software-experienced executive operator. The 2024 to 2026 period has continued Ray-ecosystem expansion alongside Anyscale Platform commercial growth.

Mission and strategy

Anyscale's mission is to make distributed computing accessible to AI developers, with the Ray framework providing the open-source distributed-runtime layer and the Anyscale Platform providing the enterprise-managed-platform layer. The strategy combines three threads. First, continued open-source Ray development and ecosystem expansion through the Ray Summit, Ray-AI Compass community programs, and academic-research engagement. Second, the Anyscale Platform as the enterprise-managed-platform offering for customers running Ray at production scale who require multi-cloud, security-compliance, and operational-observability capabilities. Third, integration with the broader AI-infrastructure ecosystem (NVIDIA GPUs, Kubernetes orchestration, hyperscale-cloud platforms) to position Ray as the cross-cutting distributed-runtime layer.

The competitive premise is that AI workloads require fundamentally different distributed-computing patterns from the data-engineering workloads that Apache Spark addressed in the prior decade, and that an AI-native distributed-runtime framework can become the structural infrastructure-layer winner.

Models and products

• Ray. Open-source distributed-computing framework. Apache 2.0 license. Includes Ray Core (distributed-task-and-actor primitives), RayServe (model serving), RayTrain (distributed training), RayTune (hyperparameter tuning), and RayData (distributed data processing).
• Anyscale Platform. Managed-Ray enterprise platform. Provides cluster management, multi-cloud deployment, security and compliance, observability, and Anyscale-developed performance optimizations.
• Ray Serve. Production model-serving layer within the Ray ecosystem.
• Ray AI Runtime libraries. Production-grade implementations of distributed training (RayTrain), tuning (RayTune), and data processing (RayData).

Distribution channels include the open-source Ray framework through GitHub and PyPI, the Anyscale Platform through direct enterprise sales, and integration partnerships with hyperscale-cloud providers and AI-infrastructure-tooling vendors.

Benchmarks and standing

Anyscale and Ray are not evaluated against foundation-model benchmarks. The framework's standing is measured through open-source-ecosystem metrics: GitHub stars (tens of thousands as of April 2026), PyPI downloads, contributor counts, citation impact in academic AI research, and named-customer references at enterprise scale.

The Ray framework is cited in the OpenAI ChatGPT and GPT-model training infrastructure, Uber's Michelangelo AI platform, ByteDance's internal AI infrastructure, Shopify's recommendation systems, and other production AI deployments. Industry coverage has characterized Ray as the principal distributed-computing framework for AI workloads, with the OpenAI usage as the most-cited validating data point.

Leadership

As of April 2026, Anyscale's senior leadership includes:

• Keerti Melkote, Chief Executive Officer (since 2024). Former founder and senior executive of HPE Aruba.
• Robert Nishihara, Co-Founder (research and strategy role since the 2024 CEO transition).
• Philipp Moritz, Co-Founder.
• Ion Stoica, Co-Founder and UC Berkeley professor (also Co-Founder of Databricks). Continues academic role at UC Berkeley.
• Senior research, engineering, and product leadership across the Ray and Anyscale Platform organizations.

The founder-led research and strategy direction has remained intact through the CEO transition, with the open-source Ray framework continuing under joint Anyscale-and-academic-community governance.

Funding and backers

Cumulative private capital exceeding $250 million across multiple rounds. Notable rounds include the seed round in 2019, Series A of $20.6 million in October 2019 led by Andreessen Horowitz, Series B of $40 million in October 2020 led by NEA, and Series C of $99 million at $1 billion valuation in October 2021 led by Andreessen Horowitz with NEA, Intel Capital, Foundation Capital, and other investor participation.

Industry position

Anyscale occupies a structurally distinctive position as the principal commercial vehicle for the Ray AI-distributed-computing framework, with the UC Berkeley research lineage, the named-enterprise customer references at OpenAI and Uber scale, and the open-source-ecosystem traction. Industry coverage has consistently characterized Anyscale as the principal AI-infrastructure-software company, with the Ray-ecosystem position structurally distinctive from the application-layer AI-tooling competitors and the foundation-model-providing competitors.

Competitive landscape

• Modal, Coiled, Outerbounds. Direct cloud-native distributed-computing alternatives for AI workloads.
• Databricks. Same Co-Founder lineage (Ion Stoica). Different product positioning (data-and-AI-platform-of-record vs. distributed-runtime).
• Dask, Apache Spark, Apache Beam. Pre-AI-era distributed-computing frameworks now retrofitted for AI workloads.
• NVIDIA RAPIDS, NeMo Megatron. NVIDIA-aligned AI-distributed-computing alternatives.
• Hugging Face Accelerate, PyTorch Distributed, DeepSpeed. Within-framework distributed-training alternatives that target single-framework rather than cross-cutting workloads.
• Berkeley BAIR and the broader Berkeley research ecosystem. Academic-research peers and Ray-ecosystem partners.

Outlook

• The continued Ray ecosystem adoption across enterprise AI customers through 2026 to 2027.
• The Anyscale Platform commercial growth under Keerti Melkote's leadership.
• Continued open-source Ray-framework feature development including RL-environment and agentic-AI capabilities.
• The Ray Summit annual conference and the broader AI-infrastructure community engagement trajectory.

Sources

• Anyscale official site. Company reference.
• Ray.io. Open-source framework.
• Robert Nishihara LinkedIn. Co-Founder reference.
• Ion Stoica Wikipedia. Co-Founder reference.
• UC Berkeley RISELab. Ray research lineage.