Flink Forward Asia 2026 Launches in Shenzhen: Agentic Streaming for AI Opens a New Real-Time Intelligence Era

On June 26, the Apache Software Foundation authorized the Flink Forward Asia 2026 (FFA) conference in Shenzhen, organized by Alibaba Cloud. The event’s theme, “Real‑time Data, Intelligent Future,” gathered developers, data engineers, AI practitioners, and industry leaders.

Flink Evolution: From Real‑time Computing to Agentic Streaming for AI

Alibaba Cloud CTO and International Business President Li Feifei opened the forum by tracing Flink’s journey from the cloud‑native Flink 2.0 to the agent‑native Flink 3.0, noting that Flink now ranks among the Apache projects with the most Chinese contributors. He emphasized that AI agents require high‑quality data and efficient processing, positioning Agentic Streaming for AI as a natural next step.

Researcher Wang Feng highlighted the shift from BI‑driven to AI‑driven computation, where processing expands from structured tables to text, images, video, and sensor signals. He argued that co‑scheduling CPU and GPU on a single pipeline eliminates the idle GPU periods typical of traditional stage‑by‑stage batch architectures.

Alibaba Cloud’s real‑time Flink edition was announced with multimodal data processing capabilities. In benchmarks covering performance, structured operators, incremental processing, connector ecosystem, and fault tolerance, it was reported to outperform open‑source stacks such as Ray and Daft.

Beyond the Present: Event‑Driven Agents and the Streaming Agent‑OS

Future road‑maps mention technologies like Harness bringing agents into a 24/7 streaming era. Starting in 2025, Flink Agents have been incubated to support event‑driven scenarios, offering context memory, exactly‑once consistency, and high‑availability fault tolerance for use cases such as live‑stream analysis, video commentary, financial risk control, intelligent operations, and public‑opinion monitoring.

Agentic Lake for AI: Building an Agent‑Native Multimodal Data Foundation

Apache Paimon 2.0, dubbed “The Next‑Gen Agentic Lake for AI,” introduces a unified multimodal data lake that combines real‑time and offline capabilities, supports Git‑for‑Table, global vector full‑text indexing, and offers REST and CLI interfaces. It aims to bridge structured and multimodal data, big‑data engines, and AI engines in a single open ecosystem.

Apache Fluss 1.0, described as “The Real‑Time Context for AI Agents,” provides Lakestream for unified real‑time and historical data, Context Serving for low‑latency access, Semantic View for governed semantics, and MCP Gateway for native agent integration. These features enable agents to make decisions based on fresh, complete, and trustworthy business context.

Case studies demonstrated Fluss powering Taobao flash‑sale decision‑making at 1.4 PB storage, reducing network I/O by 67 % and cutting operational costs by 50 %. Xiaohongshu’s migration from Kafka to Fluss lowered CPU usage by 30 %, reduced write traffic by 50 %, saved 30‑90 % bandwidth, and increased peak throughput roughly threefold.

NVIDIA Collaboration Accelerates Multimodal Streaming

NVIDIA’s senior director Chen Chuan explained joint efforts that combine NVIDIA VideoCodec, CV‑CUDA, TensorRT LLM, vLLM, and SGLang with Flink’s stream orchestration, windowing, state management, GPU scheduling, Flink Agents, and DataFrame API. The partnership enables developers to build end‑to‑end, high‑performance multimodal pipelines for AI commentary, image‑text news, and interactive Q&A.

Flink is already deployed at customers such as Titan, Kingsoft, Lalamove, Li Auto, Geely, and Zero‑Run, serving as the data backbone for their real‑time and intelligent transformations.

FFA runs through June 27, is free to developers, and will host deep‑dive sessions on industry adoption, community governance, and the continued evolution of Apache Flink, Apache Paimon, Apache Fluss, Flink Agents, and Flink CDC.