PyTorch Foundation adds nine members, including Nebius, Nscale and Shopify

Who joined and at which level

The foundation named nine new members across three tiers. At the Platinum level are Nebius, Nscale and Shopify Inc. The Silver tier gains LightSeek Foundation and Lablup Inc. Four Associate members complete the list: SKOPOS INDIA FOUNDATION, Siksha 'O' Anusandhan, Messiah University and Valencia College.

The foundation supports the core PyTorch framework alongside vLLM, DeepSpeed, Ray, Helion and Safetensors. According to the foundation, AI innovation now happens across all stack layers at once, with research, models, infrastructure, hardware and production driving each other forward, and it positions itself as the central hub of that ecosystem.

Mark Collier, executive director of the PyTorch Foundation, said: "Open systems learn faster when more people, projects, and organizations are working together, creating compounding gains."

What the new members say

Several of the new members explained their motives in statements included in the announcement.

• Lablup: Co-founder and CEO Jeongkyu Shin said the company built Backend.AI as an open-source project from the beginning and that PyTorch workloads run on its platform every day.
• LightSeek Foundation: President and Governing Board member Yineng Zhang said efficient inference is essential to making AI accessible at global scale and that the infrastructure behind it must be open.
• Nebius: Ryan Hanrui Wang, Senior Vice President, Applied AI, said the Platinum membership allows deeper collaboration with the community and a contribution of the company's expertise in high-performance infrastructure and applied AI.
• Nscale: Nick Jones, VP of Cloud Native Platform Engineering, said the company treats the framework and the data center as one problem. He also noted that Anyscale, the creators of Ray, is set to join Nscale.

Ecosystem milestones highlighted alongside the news

The foundation also listed recent developments across the PyTorch ecosystem.

• ExecuTorch on Arm Cortex-M: ExecuTorch now offers a direct path from PyTorch training to on-device inference. The Cortex-M backend runs PyTorch models on Arm Cortex-M CPUs with Arm CMSIS-NN acceleration and supports the Scalar, DSP and MVE/Helium tiers from a single workflow.
• TorchNeuron: An open source native PyTorch backend for AWS Trainium, including Amazon EC2 Trn2 and Trn3 instances. It lets researchers run PyTorch workloads with a single configuration change.
• TorchTPU: Google announced native PyTorch for TPU developers, enabling model deployment on TPUs with minimal code changes and access to projects such as vLLM, SGLang, Ray and LLM-d.
• IBM Spyre: IBM's software stack extends PyTorch's Inductor compiler and native runtime. According to the announcement, this reduces Spyre-specific compiler code by 7x while maintaining performance parity on tested workloads, with support for thousands of models through 44 adapters and about 10,000 lines of code.
• Shopify case study: Using a continual learning loop with PyTorch and vLLM, Shopify compresses production failures into model weights every day, beating frontier-model quality and cutting serving costs by 96%, according to the foundation.

What it means for the open source AI community

The new memberships add infrastructure providers, a major commerce platform, a research-focused foundation and several educational institutions to the foundation's roster. For developers, the ecosystem updates point to broader hardware reach for PyTorch, spanning Arm microcontrollers, AWS Trainium, Google TPUs and IBM Spyre.

The foundation describes itself as the vendor-neutral home for the open source layer used to train, optimize, serve, orchestrate and run models on any chip in any cloud. The conference in San Jose covers the full AI stack, from bare-metal infrastructure to applications and agents.