At Bund Conference, Data Becomes New Momentum as AI Enters Industries

Zhou Aoying, former vice president and professor at East China Normal University, said whoever controls data gains an early advantage in AI competition. He described data as an important factor of production and said high-quality AI decisions depend on high-quality data. Without high-quality underlying data, he said, any "ontology modeling" is a castle on sand.

IDC China enterprise software market research manager Wang Nan said AI is entering its second half, with data capability now the core bottleneck for enterprise AI deployment. IDC forecasts China's AI data infrastructure market will reach $154.83 billion by 2035, more than 10 times its 2025 level. Around 90% of enterprise data assets newly generated in 2025 are unstructured, according to the article, and much of it sits in NAS, object storage and other separate systems.

Wang said data keeps growing while connections among data are not happening at the same pace. Many companies hold rich mines but lack smelting capacity. The opportunity, she said, lies with whoever first builds a complete loop covering data management, data intelligence, data services and agent consumption. The route is to adopt a "small model plus multimodal base" approach, build a strong multimodal data foundation, combine industry knowledge and scenario data, and use small-model fine-tuning to achieve efficient, controllable, production-grade AI.

OceanBase's answer is an integrated data platform that moves from multimodal handling to agent friendliness. OceanBase Senior Vice President Weng Rui said the AI-era data base must evolve toward integration and multimodality, bringing different types of data and computing into one base so structured, semi-structured, unstructured and multimodal data can be governed and linked together. As AI enters real scenarios, the data user is also changing: agents are becoming new data users.

Ant Afu technology lead Wei Peng said OceanBase runs through Afu's core data links from evaluation and memory building to online services, supporting user memory accumulation, historical dialogue understanding and intelligent analysis. Sharing memory and context can save more than 30% of tokens and cut dialogue costs by 60% to 80%, he said.

Industrial data is inherently multimodal, the article said. An insurance claim may include policy fields, text descriptions, accident photos, road information and weather data. A flight delay may involve crew, weather, air traffic control and aircraft status. Multimodal capability is therefore not only about storing more data but about managing, associating and invoking different forms of data so AI can see a complete industry.

In June, OceanBase released AI Lakebase, a lake-warehouse integrated product with native support for unified storage and processing of structured, semi-structured, unstructured and multimodal data, Leiphone reported. Weng said customers once wanted a database that could store data; now they want an engine that understands data, uses data and drives intelligence. OceanBase is extending from a database toward an AI data platform to accelerate production-grade AI in various industries.

OceanBase's AI data platform has been deployed in finance, insurance, aviation and internet industries. At a large state-owned bank, a semantic layer helped the system understand business intent and raised intelligent query accuracy from 70% to 98%. At a very large property and casualty insurer, OceanBase LakeBase linked responsibility determination, claims, underwriting, road network, weather and traffic flow data to analyze high-incidence intersections and causes, helping reduce accident frequency, claims costs and the combined ratio for auto insurance.

In aviation, LakeBase manages flight, crew, weather, air traffic control and aircraft sensor data from multiple sources, identifying the impact chain of large-scale delays in near real time and assisting flight adjustments and crew reassignment to shorten recovery response time. Ping An Property & Casualty has put AI into more than 1,000 business scenarios, with AI application coverage in core scenarios at 100%, and has accumulated a trillion-token insurance industry knowledge base. It plans to build an enterprise-level data intelligence platform based on OceanBase's multi-model integrated architecture and DataPilot.

Changes extend beyond efficiency. Kuaikan Comics has launched digital life Livo based on 13,000 IPs, giving characters personality, memory and growth ability. OceanBase combines its MaaS platform with the lake-warehouse integrated LakeBase to support unified storage and real-time invocation of multimodal data, sustaining long-term character memory and continuous growth. In healthcare, Chengdu Wuhou District and OceanBase announced an "AI + medical health" cooperation and jointly established an AI plus healthcare data fusion co-creation center to explore the full chain of medical data collection, governance, circulation and intelligent application.

OceanBase CEO Yang Bing said that in the AI era, enterprises are shifting focus from database performance, cost and reliability to the relationships among different business objects behind data, and between those objects and the real world. "True business value often hides in those data that have not yet established connections," he said. As data is reconnected, governed and organized, and then invoked by AI and agents, data infrastructure is moving from carrying data to connecting enterprise data with AI. The platform, according to the article, connects downward to all enterprise data and upward to AI applications and agents, moving data from being stored and managed well to being used and generating value. Leiphone framed this as a foundation for AI's new economy to reach deeper into industries: for industries to see AI, AI must first see industries.