KubeCon China 2026 Guide: Schedule, Tickets, and Shanghai Travel¶

KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 takes place September 7–9 in Shanghai.

• Venue: Shanghai International Convention Center
• Address: No. 2727 Riverside Avenue, Pudong (near Lujiazui Metro Station)

This is the first time CNCF, OpenInfra, and PyTorch — three global open source communities — share one stage in China. The program stretches from OpenStack, storage, and virtualization through Kubernetes platform engineering to PyTorch, vLLM, inference, and agentic AI.

That breadth is the event's main advantage and its biggest planning problem. You could spend two days entirely on AI infrastructure, follow project maintainers from session to session, or focus on production reliability — and see a completely different conference from your colleague.

The way to cut through it? One sentence: Start with the problem you want to solve, then use tracks, project tables, and hallway conversations to connect the layers around it.

Why Kubernetes, OpenInfra, and PyTorch Are Sharing One Event¶

This is not merely a branding exercise. Production AI crosses boundaries that were often discussed by separate communities:

• PyTorch and related frameworks — model development and execution
• Kubernetes — service scheduling and accelerator-aware workload orchestration
• OpenInfra projects — underlying compute, virtualization, networking, storage, and isolation
• Observability, security, data movement, and developer platforms — glue the layers together

For teams that have moved beyond an isolated AI proof of concept, this conference is especially valuable. A model may work fine on one workstation, but when multiple teams need GPUs, data access, versioned deployments, autoscaling, workload isolation, and predictable inference latency — operations get hard. The joint program lets you follow one production problem through more of the stack instead of blaming the model every time something breaks.

You don't need equal expertise in all three areas. The useful question is: which adjacent layer is currently holding you back? A platform engineer might attend PyTorch sessions to understand workload behavior, while an ML engineer might sit in on scheduling and storage talks to learn why a deployment performs differently from a notebook.

Dates, Venue, and Schedule¶

The event runs September 7 (Monday) through September 9 (Wednesday), 2026 at the Shanghai International Convention Center. September 7 is reserved for co-located events; the main conference program runs on September 8 and 9.

Venue: Shanghai International Convention Center, No. 2727 Riverside Avenue, Pudong. The official schedule uses China Standard Time (UTC+8). Seating is first come, first served.

Choosing the Right Registration Type¶

The main conference offers Corporate, Individual, and Academic registration. These categories are based on employment and payment status, not technical seniority.

Main-conference registration includes keynotes, breakout sessions, tutorials, networking activities, access to the Solutions Showcase and Project Pavilion, refreshments, a conference T-shirt, and on-demand session recordings. It does not automatically include the September 7 co-located events — treat the main pass and any add-on as separate decisions.

Pick Tracks by Problem, Build Schedule by Outcome¶

Pick Tracks: Start With Your Problem¶

The official program spans more than ten tracks. Sampling every category usually creates a fragmented day. Choose one primary track + one adjacent track + a few wildcard sessions.

Build a Schedule: Five Principles¶

Before you start, write down three questions you want the conference to answer. For example: How are other teams sharing accelerators fairly? What does reliable model serving look like under burst traffic? Where should tenant isolation begin? How are maintainers testing upgrades? Having these questions makes it easier to choose between two attractive sessions in the same time slot.

1. Pick one anchor session per block — favor a talk that directly addresses a current project or decision.
2. Save a nearby alternative — seating is first come, first served; a full room shouldn't waste a block.
3. Include one cross-layer session — step outside your normal role to understand an upstream or downstream constraint.
4. Reserve hallway time — a ten-minute conversation with a maintainer or practitioner may answer a narrower question than a general talk can.
5. Don't optimize for recordings — sessions will be posted later; spend in-person time on live questions, project contacts, and discussions that cannot be replayed.

Use the official schedule to filter sessions and build a personal agenda. Save essential sessions offline and record the room, not only the title. A screenshot is useful if venue connectivity is busy or account syncing fails.

Whether September 7 Co-Located Events Are Worth Adding¶

September 7 is not a lighter preview of the main conference. It hosts separate events with their own focus and registration requirements. Add one only if its audience and subject align with your role; otherwise use the day for travel, preparation, or focused meetings.

AGNTCon + MCPCon China (Sep 6–7)¶

Focuses on the agentic stack: MCP and other protocols, infrastructure, orchestration, production systems, security, and observability. Most relevant to developers, researchers, maintainers, platform teams, and enterprises actively building agent systems. Pre-registration is required through the event's own process.

OSPOlogy + OSPO Summit China (Sep 7)¶

Focuses on open source program offices (OSPOs), corporate governance, LLM and data governance, AI-assisted software supply chains, and enterprise open innovation. Better suited to OSPO leaders, legal and compliance partners, community strategists, and engineering leaders responsible for how an organization consumes and contributes to open source.

Project Tables and Maintainer Sessions¶

The Project Pavilion is one of the strongest reasons to attend in person. Project tables let you meet maintainers directly: ask implementation questions, understand contribution paths, and discover whether a project fits a real constraint. But the conversation becomes more useful when you bring context instead of asking for a general product tour.

Before you walk up to a table:

• Be ready to describe your workload, cluster size, version, and failure mode in one minute
• Separate reproducible project issues from vendor-specific configuration problems
• Ask where maintainers prefer to handle issues, design discussions, documentation fixes, and user support
• Record the relevant repository, SIG, working group, Slack channel, or meeting
• Don't expect maintainers to debug sensitive production data at a public table

Key timings:

• Project Lightning Talks — September 8
• Maintainer Track — Tuesday and Wednesday
• Solutions Showcase project tables — Tuesday and Wednesday

Review the official project engagement options and note when the maintainers you want to meet will be present.

DaoCloud will have engineers stationed at the S2 main booth and multiple open source project tables. Come by and chat.

DaoCloud: 7 Technical Talks and Booth Information¶

When AI workloads move into production, the real challenges begin: How do you schedule expensive GPUs more efficiently? How do you quickly find the root cause of slow inference? How must Kubernetes evolve to handle increasingly complex AI workloads?

DaoCloud brings 7 technical talks to the conference, organized along three directions.

Networking: Bring Questions, Not Job Titles¶

The efficiency of conference networking depends on how specific your introduction is. "We're moving batch inference onto K8s and comparing scheduling approaches" — the other person immediately knows who you are, what you do, and what you might discuss. A list of job titles and technology names doesn't do that.

Three steps:

1. Before the event — identify a few speakers, maintainers, or teams whose work intersects with yours; read the abstract or repository first
2. During the event — ask one focused question, respect their time
3. After the event — write a short note explaining why the contact matters, send promised follow-ups while the discussion is still fresh

The official FAQ points attendees to the CNCF Slack workspace and its hallway channel. Onsite activities also include a welcome reception, a Women's Gathering, peer group mentoring, and a PyTorch community night. Some activities have limited capacity or require an RSVP — check the conference experiences page before arrival.

What to Bring¶

Two dense technical days, a venue full of parallel conversations. Pack light with a reliable note-taking system — it beats carrying a "mobile office."

Shanghai Travel Guide¶

Venue: Shanghai International Convention Center, No. 2727 Riverside Avenue, Pudong.

Airport distances (official estimates):

Don't schedule a critical session right after a flight or train — leave a generous arrival margin.

The official conference hotel block has closed. When booking alternatives, compare three things: location, cancellation terms, and the actual morning commute route. A cheaper room that requires gambling on cross-city traffic may cost more in missed sessions and spent energy. Verify the address in both English and Chinese, save it offline, and reference the official venue and travel guidance.

International visitors — confirm before departure:

• Passport validity and entry requirements
• Payment methods (mobile payment is nearly universal in China; international cards are not always accepted)
• Mobile connectivity (eSIM or local SIM card)
• Access to tools required for work

Post-Event: Turn Notes Into Experiments¶

"Research later" — this phrase kills more conference inspiration than anything else.

Within a week, pick one idea and reduce it to a testable question: compare two Ingress configurations, reproduce an observability pattern, deploy a small model service, evaluate a sandbox boundary, or test storage behavior under failure. Record software versions, assumptions, commands, metrics, and rollback steps.

A laptop VM or an existing server is enough for many first experiments. If you want a reusable multi-node environment, compact cluster build examples show how small x86 nodes can be organized for distributed systems learning.

Conference Planning Checklist¶

FAQ¶

Comments