What was the market backdrop for the institutional buying?

  • On 9 November, Enflame Technology saw net institutional buying of 120 million yuan; a total of 22 stocks had daily net buying or selling by institutions exceeding 10 million yuan, of which 13 saw net buying and 9 saw net selling.
  • The broader market backdrop was not strong: turnover on the Shanghai and Shenzhen exchanges was 1.48 trillion yuan, down 175.2 billion yuan from the previous trading day, and more than 3100 stocks fell across the market. The net buying therefore looks more like a capital-flow signal for an individual stock than a broadening of buying amid a generally rising market.

How has Enflame Technology built its computing-power product capabilities?

  • The company has independently developed and iterated four generations of architectures and five cloud-based AI chips. Its fourth-generation DiSi 400 supports FP8 precision, and the company says it can provide supernode networking and deliver clusters at the ten-thousand-card scale. Its products span AI accelerator cards, intelligent-computing systems and clusters, and software platforms; its offerings extend beyond chips to include clusters and the software stack.
  • The company pursues parallel development across multiple generations, with products aimed at large-model training and inference. This places its business entry point in the hardware and software ecosystem supporting domestic computing power, rather than in chip specifications alone.

What business validation does the Tencent partnership provide?

  • Enflame Technology and Tencent have worked together since 2019, jointly optimizing business-scenario adaptation, model performance, and the software stack. After about seven years, the partnership has progressed from small-scale validation in a single scenario to large-scale validation across multiple scenarios and routine bulk cooperation.
  • The company says its products have been deployed at scale in a substantial number of Tencent AI business scenarios, and sales to Tencent-related entities have continued to grow. Compared with a single day’s capital flows, these developments are more directly connected to product use and customer procurement.

What does the next-generation chip roadmap point to?

  • The fifth-generation chip is planned to support FP4 precision and cross-modal generation across text, audio, images, and video. The sixth generation is planned to strengthen inter-card connectivity and heterogeneous multi-chiplet integration to meet the computing-power demands of very large-scale models.
  • The roadmap extends from current FP8 support to a wider range of model tasks and stronger connectivity capabilities. The fifth and sixth generations remain product plans, distinct from products already deployed and the current stage of existing customer partnerships.