On July 4, 2023, Tencent Cloud（Chinese：腾讯云）held a press conference to officially announce the release of their vector database product, Tencent Cloud VectorDB. Tencent Cloud stated that this database is an "AI Native" vector database that can be widely applied in scenarios such as training large models, inference, and knowledge base augmentation.
According to Luo Yun, Deputy General Manager of Tencent Cloud Database, in the past, vector databases were closely associated with recommendation systems and anti-fraud applications. However, with the emergence of needs such as updating model data and assisting model inference, vector databases are now demonstrating new value. In the model training phase, vector databases can be used for classification, deduplication, and cleansing of pre-training data for large models. In terms of effectiveness, vector databases can achieve a 10-fold improvement in efficiency compared to traditional data preparation methods. In the inference phase, if Tencent Cloud's vector database is used as an external knowledge base, costs can be reduced by 2-4 orders of magnitude.
Tencent Cloud's vector database is based on the OLAMA (Online Analytical and Mining Application) vector engine, which processes billions of queries daily within Tencent Group. Through extensive internal practice in various scenarios, the efficiency of data integration with AI has been improved 10-fold compared to traditional solutions, with a high stability rate of 99.99%. Currently, it has been applied in over 30 popular products in China, including Tencent Video（Chinese：腾讯视频）, QQ Browser（Chinese：QQ浏览器）, and QQ Music（Chinese：QQ音乐）.
Tencent Cloud's vector database can effectively help improve operational efficiency. Data shows that after using Tencent Cloud's vector database, QQ Music has seen a 3.2% increase in average listening time per user, Tencent Video has achieved a 1.74% increase in average exposure time per user, and QQ Browser has reduced costs by 37.9%.
In China, the most widely used areas for database applications are finance, telecommunications, government affairs, manufacturing, and the Internet. However, each of these areas has different application characteristics. The finance and telecommunications sectors have stricter IT regulatory environments, more complex data operations, and a "strong transaction" nature for core data services, but they are less sensitive to costs. On the other hand, the Internet sector has weaker IT regulatory environments but is more cost-sensitive.
Tencent Cloud also faces fierce competition in China. Currently, domestic database enterprises can be classified into four categories: traditional vendors represented by For Redis（Chinese：达梦数据库）, KingbaseES（Chinese：人大金库）, and shentongdata（Chinese：神舟通用）; startup vendors represented by VASTDATA（Chinese：海量数据）,cloud-ark（Chinese：极数云舟）, and SequoiaDB（Chinese：巨衫数据库）; cloud vendors represented by aliyun（Chinese：阿里云）, Tencent Cloud, and huaweicloud（华为云）; and cross-industry vendors represented by ZTE（Chinese：中兴）, Inspur（Chinese：浪潮）, and bonc（Chinese：东方国信）.
Tencent Cloud's vector database supports up to a billion-scale vector retrieval, with latency controlled in milliseconds. Compared to traditional single-node plug-in databases, the retrieval scale has increased by 10 times, while maintaining a peak capacity of millions of queries per second (QPS). The accelerated development of vector databases for large-scale models is projected to drive the global vector database market to reach USD 50 billion by 2030, with the domestic market surpassing CNY 600 billion .