PyTorch Conference North America 2026 — San Jose

The PyTorch Conference North America 2026 takes place on October 20-21, 2026 in San Jose, California, hosted by the PyTorch Foundation. It is the flagship annual gathering for the people who build and ship deep-learning systems with PyTorch, drawing AI researchers, framework maintainers, and machine-learning engineers from across the open-source community. If you train models, optimise inference, or build agent-based applications on the PyTorch stack, this is the event aimed squarely at your work.

What PyTorchCon 2026 covers

PyTorch underpins a large share of the world's research and production AI, and the conference spans the full stack — from bare-metal infrastructure up through applications and agentic systems. Sessions mix in-depth technical talks, hands-on workshops, and candid conversations about where open-source AI is heading. The 2026 programme is organised around a set of focused topic areas:

• Applications — projects that showcase the broader PyTorch ecosystem in production.
• Training — techniques and libraries for distributed and large-scale model training.
• Inference — serving, latency, and throughput, including work tied to projects like vLLM.
• Core PyTorch — changes to the framework itself.
• Kernel engineering — compilers, optimisation, and domain-specific languages such as the work emerging around Helion.
• Responsible AI — ethics, governance, security, sandboxing, and privacy.

Keynotes feature leading voices in AI, with practical deep dives into training, inference, and generative AI in production. The conference deliberately spans the layers most engineering teams have to reason about at once: how a model is trained, how it is served cheaply and reliably, and how it is wrapped into safe, useful applications.

The ecosystem on show

One reason PyTorchCon matters more each year is the breadth of projects now stewarded under the PyTorch Foundation. Beyond the core framework, the foundation's landscape includes vLLM for high-throughput LLM inference, DeepSpeed for large-model training efficiency, Ray for distributed compute, ExecuTorch for on-device deployment, and tooling such as safetensors. For attendees, that means a single venue where the maintainers of the libraries you depend on are presenting roadmaps, debating design trade-offs, and taking questions in person. It is a rare chance to short-circuit weeks of GitHub-issue archaeology into a hallway conversation.

The 2026 edition leans visibly into the industry-wide shift toward agent-based systems. The organisers explicitly frame the programme as reaching "from bare metal infrastructure to applications and agent-based systems," reflecting how PyTorch is increasingly the substrate beneath tool-using, multi-step AI agents rather than just standalone models. Expect the inference and applications tracks in particular to address the cost and orchestration pressures that come with putting generative AI and agents into real products.

Context: open AI in 2026

PyTorchCon also reflects a broader 2026 reality: the centre of gravity in applied AI has moved toward open frameworks, efficient inference, and agentic application patterns, and PyTorch sits at the heart of that shift. Talks consistently connect low-level performance work — kernels, compilers, quantisation — to the practical goal of running larger models and longer agent chains at acceptable cost. For teams weighing build-versus-buy decisions, hearing maintainers explain trade-offs first-hand is hard to replicate elsewhere.

Who should attend

PyTorch Conference is built for a technical audience: ML and research engineers, framework and compiler contributors, platform and MLOps teams running training or inference at scale, and developers building generative-AI and agentic applications on open foundations. The foundation expects the event to bring together thousands of engineers and researchers, so it is also a strong networking and recruiting venue for anyone working in open-source AI. Decision-makers evaluating which open frameworks to standardise on will find the project roadmaps especially useful for planning.

Format, dates, and how to attend

The conference is an in-person, two-day event on October 20-21, 2026 in San Jose, with the programme combining keynotes, technical sessions, and workshops. It sits within a wider season of PyTorch Foundation activity that also includes regional editions elsewhere in the world, but the San Jose event is the North American flagship. Registration is handled through the Linux Foundation events platform, and the organisers note that registering early secures a pass at a discounted rate; there is also an open call for session proposals across the topic areas above, giving practitioners a route onto the stage. Anyone planning travel should book accommodation early, as San Jose hotel inventory fills quickly during major AI events. For the authoritative agenda, speaker list, and registration details, consult the official PyTorch Foundation event page, which is updated as the schedule firms up through 2026. For ML engineers and AI developers tracking where open-source training, inference, and agent tooling is going, PyTorch Conference North America 2026 is one of the calendar's anchor dates.