PyTorch Foundation

What PyTorch Foundation does

PyTorch Foundation is a non-profit organization established under the Linux Foundation to govern and advance the PyTorch open-source deep learning framework and a portfolio of related AI infrastructure projects. Originally contributed by Meta, the foundation now hosts six foundation-hosted projects — PyTorch (core deep learning framework), ExecuTorch (on-device inference), vLLM (high-throughput LLM serving), DeepSpeed (distributed training), Ray (distributed computing), Helion (ML kernel DSL), and Safetensors (secure model serialization) — alongside domain libraries spanning computer vision, audio, NLP, recommendation, reinforcement learning, and data loading.

The foundation operates a product-led and community-led growth model, distributing the framework freely across Linux, macOS, and Windows with native support for NVIDIA CUDA, AMD ROCm, Intel, and TPUs, and pre-configured environments on AWS, Google Cloud, and Azure. It does not sell software licenses; instead it generates revenue through tiered corporate memberships (Platinum at $350K/year, Gold at $150K/year, and Silver with sliding-scale pricing), a Cloud Credit Program (Supporter $250K, Advocate $500K, Champion $1M+), and the PyTorch Certified Associate (PTCA) credential launched in June 2026.

The foundation serves AI researchers, ML engineers, enterprise platform providers, and individual developers, with 3,500+ individual contributors, 3,000+ contributing organizations, 2M+ daily downloads, and member companies spanning every major cloud, chip vendor, and AI lab. Governance is provided by a Governing Board and Technical Advisory Council staffed by representatives from Meta, NVIDIA, Google, Microsoft, AWS, Intel, AMD, Arm, IBM, Alibaba Cloud, Hugging Face, and others, giving the foundation vendor-neutral credibility and broad industry alignment.

Name

PyTorch Foundation

Legal name

PyTorch Foundation

Website

https://pytorch.org

Company type

Private

Founded year

2019

Operating status

Operating

Headcount range

11–50 employees

Short description

The PyTorch Foundation is a Linux Foundation-hosted non-profit that governs the PyTorch deep learning framework and six related open-source AI projects, serving researchers, ML engineers, and enterprise platform providers worldwide through corporate memberships, certification, and sponsorships.

Ownership category

akta.pro rank

Product category

Open-Source Deep Learning Framework

NAICS

Grantmaking Foundations (813211), Grantmaking and Giving Services (81321)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industry

Private Foundations (BPAGAKAA)

Keywords

• Open source deep learning
• AI framework governance
• Distributed training infrastructure
• ML kernel optimization
• AI certification programs

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

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

Revenue model

1. Membership Fees: PyTorch Foundation generates revenue through tiered membership programs. Platinum members pay $350,000/year with voting board seats and strategic direction influence. Gold members pay $150,000/year. Silver and Associate tiers available with sliding scale pricing based on company size.
2. Training and Certification: Revenue generated from PyTorch Certified Associate (PTCA) certification exams and training programs administered through Linux Foundation Education. Certification includes free retake and 2-year validity.
3. Cloud Credit Program Sponsorship: Organizations contribute cloud credits or financial donations to support PyTorch's CI infrastructure and foundation-hosted project infrastructure. Sponsorship tiers include Supporter ($250K), Advocate ($500K), and Champion ($1M+).

Pricing tiers

Go-to-market motion3 records

Distribution channels8 records

Marketing channels14 records

Core offering

PyTorch Foundation is a non-profit organization under the Linux Foundation that governs and develops the PyTorch open-source deep learning framework and a portfolio of foundation-hosted AI projects (ExecuTorch, vLLM, DeepSpeed, Ray, Helion, Safetensors). The foundation provides vendor-neutral governance, ecosystem stewardship, certification programs, conferences, and community resources for ML researchers, developers, and enterprise users. It distributes software via self-service downloads, cloud platforms (AWS, GCP, Azure), and Hugging Face, funded primarily through tiered corporate memberships.

Product overview

The PyTorch Foundation, part of the Linux Foundation, hosts a suite of open source AI projects centered on the PyTorch core deep learning framework. The portfolio includes foundation-hosted projects — PyTorch (core), ExecuTorch (edge/on-device inference), vLLM (LLM serving), DeepSpeed (distributed training), Ray (distributed computing), Helion (ML kernel DSL), and Safetensors (secure model serialization) — alongside domain-specific libraries such as torchvision (computer vision), torchaudio (audio processing), torchtune (LLM fine-tuning), torchrec (recommender systems), torchCodec (media encoding), torchrl (reinforcement learning), torchdata (data loading), Captum (model interpretability), and PyTorch Geometric (graph/ne geometric learning). The ecosystem also includes tools like TorchServe for model serving and torch.compile for graph-mode optimization. These products form an open, modular AI development platform spanning research prototyping through production deployment across cloud, edge, and mobile environments.

Differentiator

Problem solved

Functional benefit

Brands

• PyTorch: Open source deep learning framework built to be flexible and modular for research, with the stability and support needed for production deployment.
• ExecuTorch
• vLLM
• DeepSpeed
• Ray
• Helion
• Safetensors

Products and services

• PyTorch PyTorch is an open-source deep learning framework built for flexible, modular research with production deployment support. It offers dynamic computation graphs, Pythonic design, and deep integration with native Python tools for building and training deep learning models at scale.
• ExecuTorch ExecuTorch is an end-to-end on-device inference solution for mobile, edge, and embedded devices, enabling efficient deployment of PyTorch models (vision, speech, Generative AI) across wearables, smartphones, and microcontrollers with a lightweight runtime utilizing CPUs, NPUs, and DSPs.
• vLLM vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs, optimizing hundreds of language models across NVIDIA and AMD GPUs, Google TPUs, AWS Trainium, and Intel CPUs using PagedAttention, chunked prefill, multi-LoRA, and automatic prefix caching.
• DeepSpeed DeepSpeed empowers developers to streamline distributed training and inference for large AI models, addressing challenges of distributed training, mixed precision, gradient accumulation, and checkpointing, enabling unprecedented scale and speed for both training and inference.
• Ray Ray is a unified framework for scaling AI and Python applications, providing distributed computing primitives and higher-level libraries for multimodal data processing, model training, model serving, and reinforcement learning.
• Helion Helion is a Python-embedded domain-specific language (DSL) for authoring ML kernels that compiles to Triton and other backends, providing automated ahead-of-time autotuning and hardware performance portability across NVIDIA, AMD, and Intel GPUs.
• Safetensors Safetensors is a secure, fast file format for storing ML model weights developed by Hugging Face. It prevents arbitrary code execution during deserialization through a JSON header and raw data buffers approach, with zero-copy loading and support for multi-GPU deployments.
• PyTorch Certified Associate (PTCA) The PyTorch Certified Associate (PTCA) is a certification credential offered in partnership with Linux Foundation Education for early-career AI/ML practitioners, validating proficiency in PyTorch through a multiple-choice exam with free retake and two-year validity.

Quantifiable outcome

• 85% research share vs TensorFlow's 15% in AI/ML papers
• +5 more outcomes

Named customers12 records

Segments6 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability15 records

Feature11 records

Partnerships

31 partnerships are on record, tiered core and minor.

• Alibaba CloudcoreStrategic or Co-development Partner · 26 May 2026Alibaba Cloud joined the PyTorch Foundation as a Platinum member. Alibaba Cloud is a global leader in full-stack AI services offering intelligent capabilities and worldwide AI cloud computing network. Alibaba Cloud develops foundation models including Qwen and Wan, and serves global cloud customers. Junhua Wang, VP of Alibaba Cloud Intelligence Group, represents Alibaba Cloud on the Governing Board.
• MetacoreStrategic or Co-development Partner · 1 April 2026Meta contributed Helion to the PyTorch Foundation in April 2026. Helion is a Python-embedded DSL for authoring ML kernels providing automated autotuning and hardware portability across backends including Triton and TileIR.
• Hugging FacecoreStrategic or Co-development Partner · 1 March 2026Hugging Face contributed Safetensors to the PyTorch Foundation in March 2026. Safetensors is a secure tensor serialization format preventing arbitrary code execution during deserialization. Luc Georges represents Hugging Face on the Technical Advisory Council.
• Clockwork.iominorStrategic or Co-development Partner · 24 February 2026AI company that joined the PyTorch Foundation as a new member since December 2025, signaling interest in agentic AI innovation within the open-source ecosystem.
• Emmi AIminorStrategic or Co-development Partner · 24 February 2026AI company that joined the PyTorch Foundation as a new member since December 2025.
• Nota AIminorStrategic or Co-development Partner · 24 February 2026AI company that joined the PyTorch Foundation as a new member since December 2025.
• yasp.aiminorStrategic or Co-development Partner · 24 February 2026AI company that joined the PyTorch Foundation as a new member since December 2025.
• Carnegie Mellon UniversityminorStrategic or Co-development Partner · 24 February 2026University that joined the PyTorch Foundation as a new member since December 2025, supporting academic engagement with open-source AI.
• Monash UniversityminorStrategic or Co-development Partner · 24 February 2026University that joined the PyTorch Foundation as a new member since December 2025.
• University of LeicesterminorStrategic or Co-development Partner · 24 February 2026University that joined the PyTorch Foundation as a new member since December 2025.
• NIPA (National IT Industry Promotion Agency)minorStrategic or Co-development Partner · 24 February 2026National IT Industry Promotion Agency from South Korea joined the PyTorch Foundation as a new member since December 2025.
• AnyscalecoreStrategic or Co-development Partner · 1 September 2025Anyscale contributed Ray to the Linux Foundation in September 2025. Ray is a unified framework for scaling AI and Python applications with distributed computing primitives. Robert Nishihara represents Anyscale on the Technical Advisory Council as Ray Project Representative.
• MicrosoftcoreStrategic or Co-development Partner · 1 January 2025Microsoft contributed DeepSpeed to the Linux Foundation in January 2025. DeepSpeed empowers developers to streamline distributed training and inference, powering world's most powerful language models like MT-530B and BLOOM.
• University of California, BerkeleycoreStrategic or Co-development Partner · 1 July 2024UC Berkeley contributed vLLM to the Linux Foundation. vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs now hosted under the PyTorch Foundation.
• MetacoreStrategic or Co-development PartnerMeta is the original creator of PyTorch and remains a premier member and key contributor to the PyTorch Foundation. Jana van Greunen serves as Engineering Director on the Governing Board. Meta contributed ExecuTorch and Helion to the foundation. Meta uses PyTorch extensively across its Family of Apps products and services.
• NVIDIAcoreStrategic or Co-development PartnerNVIDIA is a premier member of the PyTorch Foundation with Ankit Patel (Senior Director) on the Governing Board and Piotr Bialecki on the Technical Advisory Council. NVIDIA's CUDA platform and GPU acceleration are deeply integrated with PyTorch. NVIDIA contributes to PyTorch maintenance and has expanded open-source AI integration including making CUDA Python first-class.
• Google CloudcoreStrategic or Co-development PartnerGoogle Cloud is a premier member of the PyTorch Foundation with Brian Granger (Senior Principal Technologist, also co-founder of Project Jupyter) on the Governing Board. Google provides PyTorch support on GCP including Deep Learning Containers and VM Images. Claudio Basile represents Google on the Technical Advisory Council.
• Microsoft AzurecoreStrategic or Co-development PartnerMicrosoft Azure is a premier member with Weizhu Chen representing Microsoft on the Technical Advisory Council. Microsoft contributed DeepSpeed to the foundation and provides PyTorch support through Azure Machine Learning. Microsoft co-founded the DeepSpeed team and contributed to PyTorch optimization.
• Amazon Web Services (AWS)coreStrategic or Co-development PartnerAWS is a premier member with Brian Granger representing AWS on the Governing Board. AWS provides cloud credits for PyTorch CI infrastructure and supports PyTorch through SageMaker, Deep Learning Containers, and AWS Inferentia. Shauheen Zahirazami represents AWS on the Technical Advisory Council.
• IntelcoreStrategic or Co-development PartnerIntel is a premier member with Fan Zhao (Senior Director AI Frameworks) on the Governing Board. Intel optimizes PyTorch, vLLM, and SGLang for Intel hardware. Eikan Wang represents Intel on the Technical Advisory Council. Intel provides cloud credits and supports PyTorch across Intel CPU and GPU platforms.
• AMDcoreStrategic or Co-development PartnerAMD is a premier member with Niles Burbank (Director of Product Management) on the Governing Board. AMD GPUs are supported through ROCm platform. Jeff Daily (AMD Fellow) is on the Technical Advisory Council and is chief architect of ML Software Engineering group supporting PyTorch on AMD GPUs.
• ArmcoreStrategic or Co-development PartnerArm is a premier member with Andrew Wafaa (Open Source Office) as Governing Board Chair. Arm provides cloud credits for CI infrastructure. Milos Puzovic (Technical Director) serves as TAC Vice Chair. Arm focuses on accelerating PyTorch on AArch64 architecture.
• IBMcoreStrategic or Co-development PartnerIBM is a premier member with Raghu Ganti (Distinguished Engineer) on the Governing Board. IBM provides cloud credits and uses vLLM in their RITS Platform. IBM Research works on PyTorch optimization and applications including geospatial AI with TerraTorch.
• Hugging FacecoreStrategic or Co-development PartnerHugging Face is a Gold member with Lysandre Debut (Head of Open Source) on the Governing Board as Gold Member Representative. Hugging Face contributed Safetensors to the foundation. Luc Georges represents Hugging Face on the Technical Advisory Council. Hugging Face's model hub is widely used with PyTorch.
• SnowflakecoreStrategic or Co-development PartnerSnowflake is represented on the Governing Board by Dwarak Rajagopal (VP of AI Engineering). Snowflake contributes to the foundation and has Olatunji Ruwase (DeepSpeed project representative) and Mert Hidayetoglu on the Technical Advisory Council.
• HuaweicoreStrategic or Co-development PartnerHuawei is represented on the Governing Board by Fred Li (General Manager of Open Source Development). Yikun Jiang (Principal Engineer) represents Huawei on the Technical Advisory Council. Huawei contributes to open-source AI infrastructure and PyTorch ecosystem.
• GraphcoreminorStrategic or Co-development PartnerGraphcore is a Silver member with Sylvain Viguier (Director of AI) as Silver Member Representative on the Governing Board.
• ShopifyminorStrategic or Co-development PartnerShopify is represented on the Governing Board by Shuying Sun.
• GooglecoreStrategic or Co-development PartnerGoogle is represented on the Governing Board by Sayce Falk (Director of AI/ML Product Management). Google provides TPU support for PyTorch through torch_xla and supports PyTorch on Google Cloud Platform.
• Lightning AIminorStrategic or Co-development PartnerLightning AI is represented on the Governing Board by Thomas Viehmann as TAC Chair.
• Red HatminorStrategic or Co-development PartnerRed Hat is represented on the Technical Advisory Council by Robert Shaw.

Scale indicators9 records

Recent moves8 records

Expansion highlights6 records

Direct peers

• TensorFlow: Google's open-source deep learning framework and the historical dominant alternative to PyTorch. Directly comparable as a foundational DL framework competing for the same researcher, enterprise, and production ML workloads, now ~15% research share versus PyTorch's 85%.
• Keras: High-level deep learning API originally positioned as a competing interface to PyTorch, now integrated as a multi-backend front-end with significant PyTorch backend adoption. Directly comparable as a deep learning developer interface competing for the same enterprise ML user base.

Broad incumbents

• Linux Foundation: Parent organization and host of the PyTorch Foundation and the broader LF AI & Data umbrella. Operates the governance, fiscal, legal, and event infrastructure that PyTorch Foundation leverages; comparable as a peer nonprofit foundation but at vastly greater scale across many open-source projects.
• Hugging Face: The leading open-source AI model and dataset hub that contributed Safetensors to the PyTorch Foundation and is a Gold member. Highly comparable as an open-source AI ecosystem player, complementary rather than directly competitive, but increasingly overlapping in model distribution and tooling infrastructure.

Emerging players

• Google JAX: Google's emerging Python-centric numerical computing and ML framework with growing adoption in research and high-performance computing. Competes with PyTorch in numerical/autodiff workloads and is gaining traction in scientific computing, making it the most credible emerging alternative framework.
• Apple MLX: Apple's open-source ML framework optimized for Apple Silicon. Emerging alternative in on-device ML and a potential competitor to PyTorch's ExecuTorch for Apple-platform deployments, particularly among iOS/macOS developer ecosystems.
• Anyscale: The commercial entity behind Ray, contributed to the PyTorch Foundation in September 2025. Comparable as a distributed AI compute platform adjacent to PyTorch Foundation's portfolio, with commercial offerings built on top of foundation-hosted open source.

Others

• Apache Software Foundation: The largest open-source software foundation, hosting frameworks like MXNet, TVM, and other ML/AI-adjacent projects. Comparable as a peer open-source foundation with similar governance, fiscal sponsorship, and contributor models, though with a broader scope beyond AI/ML.
• ONNX (Open Neural Network Exchange): Open standard and runtime for interoperable AI model representation, hosted under The Linux Foundation's LF AI & Data. Complementary infrastructure piece often paired with PyTorch for model deployment, sharing governance parentage but distinct project focus.
• Cloud Native Computing Foundation (CNCF): Sister Linux Foundation sub-foundation hosting Kubernetes and cloud-native infrastructure projects (including Kubeflow for ML pipelines). Comparable as a peer foundation in the Linux Foundation ecosystem that overlaps with PyTorch Foundation in the ML infrastructure layer.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/pytorch
• GitHubgithub.com/pytorch/pytorch
• YouTubeyoutube.com/pytorch
• Facebookfacebook.com/pytorch
• LinkedInlinkedin.com/company/pytorch

Revenue estimate

Valuation estimate

Number of profiles

Profiles20 records

Funding overview

Funding rounds

Investors

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 PyTorch Foundation