Recently, the Ministry of Industry and Information Technology launched the Industrial Data Foundation Action, aiming to build a batch of high-quality, standardized, and tradable industry datasets by the end of 2026 to empower large models and industrial AI agents. This policy moves "high-quality datasets" from a technical concept to the forefront of industrial practice, raising a critical question: how should high-quality datasets be built?

The key to building high-quality datasets is not the repository itself, but whether the data can be efficiently consumed by AI.

This requires three pillars: a solid technology foundation, deep engineering capabilities, and industry-wide collaboration.

As a pioneer in the AI data infrastructure field, KeenData has long focused on Data&AI infrastructure software. Its core product, KeenData Lakehouse, is built on the "AI-in-Lakehouse" philosophy, integrating compute-storage separation, Data&AI integration, and intelligent data governance, covering the entire chain from data integration to model training and inference. Over the past few years, this technology foundation has been deployed across multiple industries including energy, finance, retail, and manufacturing. A high-quality dataset project completed last year in Suzhou systematically addressed data governance standardization, covering the entire process of multi-source data access, cleansing, labeling, and governance, thereby building an AI‑ready high-quality industry dataset and successfully supporting multiple intelligent application scenarios. The project fully demonstrates an effective path from governance to AI application of high-quality datasets.

A high-quality dataset is not about “the more storage, the better”, but “the more accurate the use, the more valuable it becomes”.

Many enterprises mistakenly believe that a large data volume equals high quality. However, in real‑world industrial scenarios, data must be governed to achieve unified standards, tradability, and direct usability for AI training. Through the built‑in data asset management, quality monitoring, and metadata management modules of KeenData Lakehouse, KeenData helps enterprises restructure data resource catalogs, unify data standards, and establish quality control systems. This enables data to be directly consumed by AI, allowing every business user to precisely find the AI data they need, and the data is ready for AI model training.

That is the true meaning of a high-quality dataset.

The MIIT action particularly emphasizes that high-quality datasets must empower industry large models and industrial AI agents. The ultimate goal of a dataset is not “storage” but “use” — especially being used by AI. If a dataset cannot generate quantifiable business value, it becomes a cost. KeenData always works backwards from business KPIs to define data governance goals, ensuring that every data step aligns with specific business scenarios. In finance, this means real‑time fraud interception; in retail, it means the responsiveness of business dashboards. Taking business results as the driver avoids the pitfall of “building repositories for the sake of building repositories”. KeenData’s DataOps capability supports full‑chain automation from data collection to service output, ensuring that data development synchronizes with business needs — this is the core support for closing the loop between AI data governance and business value.

Building high‑quality datasets is not a problem that can be solved by a standard software package; it requires deep industry know‑how and engineering capability grounded in real business scenarios.

Problems such as incompatible protocols, difficulties in real‑time access, uneven historical data quality, and lack of secure circulation mechanisms are common in manufacturing. KeenData Lakehouse supports real‑time access and batch processing of multi‑source heterogeneous data, with a built‑in data lake storage and lakehouse engine capable of handling both structured and unstructured data. Its intelligent scheduling and task orchestration can cope with complex industrial data pipelines. Through years of serving industrial clients, KeenData has developed a methodology covering multi‑source heterogeneous data access, standardized governance, and secure circulation, and has productized this capability. The MIIT’s call to “break the collection, integration, utilization bottleneck of industrial data” precisely responds to this industry pain point.

Building high‑quality datasets is a long‑term battle that requires “true craftsmanship” and multi‑party collaboration across the industry chain. Public data infrastructure is an indispensable part of this effort. In projects such as the government pilot demonstration zone and the city data group, KeenData has completed the construction of core modules including trusted data spaces and data security protection, opened the full link from data processing to model evaluation, and implemented the data element circulation model of “available, flowable, usable, and secure”. KeenData is one of the first pilot units for standards/technical document verification in the trusted data space direction under the National Data Bureau, continuously contributing to the trusted data circulation ecosystem. At the same time, the company actively participates in the construction of industrial data standards, working with industry partners to develop key standards for data collection and dataset quality assessment. KeenData Lakehouse is capable of adapting to domestic GPU chips, supporting compute power adaptation in the context of trusted IT (xinchuang) environments.

The MIIT’s Industrial Data Foundation Action draws a clear roadmap for the industry. However, what truly determines the industry’s progress is not policy documents, but hard work in real‑world scenarios. From the Suzhou high‑quality dataset project to covering more than 20 industries and serving nearly 200 large organizations, KeenData has accumulated systematic practical capabilities in the field of high‑quality dataset construction. As an AI data infrastructure provider, KeenData is committed to working with ecosystem partners to drive data from “resource” to “asset”, making AI a core driver of new‑type industrialization.