Star Ring Technology: Roadshow scheduled for April 1, with participation from multiple institutions including Yangtze River Pension and Guojin Securities.

Securities Star News: On April 2, 2026, Hylink (688031) released an announcement stating that the company held a roadshow on April 1, 2026 with Yangtze River Pension, Guojin Securities, HuaiXin Fund, Yaohejing, Allianz Investment, Tianxiang Capital, Zhongcai Capital, Orient Securities, Soochow Securities, 广发证券, RuiJun Asset Management, Southern Fund, Dacheng Fund, Zheshang Securities, CICC Asset Management, Yunzhou Capital, Jufang Zhitou, Huizheng Asset Management, Time Capital, Shangyin Fund, Nantuk Asset, Muxi Capital, 广发基金, ZhuoYing Investment, Zheshang Asset Management, HuayongxIn Capital, Beiyi Investment, Shenzhen Zhanbo Investment, Shanghai Old Fisherman Investment, Invesco Great Wall, Juding Private Fund, YiAn Investment, QikaI Asset Management, Penghua Fund, CICC-Ruiyin UBS, Huisheng Fund, Bohai Securities, Citic Securities, Huchuang, Galaxy Fund, Nuofang Fund, and Guotai Haitong.

The specific content is as follows:

Q: We see that GPUs are suitable for parallel computing tasks, while agents have more complex tool calls. Will our cognition database involve a division of labor between CPUs and GPUs?

A: From a technical standpoint, the company reduces CPU–GPU interaction through GPU direct connection technology, putting the vast majority of work on the GPU, and only letting the CPU handle scheduling. In some cases, it can even achieve GPU direct storage without going through the CPU. The design principle is to complete operations as much as possible within the machine and reduce cross-node interactions.

Q: Please outline the company’s product planning going forward and the timeline for commercialization rollout.

A: In terms of commercialization strategy, the company expects to launch a cloud version in the second half of this year, first entering overseas markets to attract overseas clients with high cost performance. At the same time, in the domestic market, it will look for customers that are sensitive to latency or have relatively frequent I deployment, striving to secure benchmark customer deployments within this year.

Q: What impact will NV’s release of cuDF and cuVS have on us?

A: We develop based on NV’s cuDF. Some indexes are also based on cuVS, and in the future we will also base them on domestic GPUs.

Q: What is the relationship between our company and large model companies?

A: Large models are more geared toward static knowledge storage. The core is continuously strengthening their own reasoning and programming capabilities. What the company provides is an external dynamic knowledge base, used to store continuously changing information such as real-time operating data, contracts, videos, and other updates—this differs from the static knowledge use inside large models. Meanwhile, this system can also permanently remember user preferences, causing the memory footprint to keep growing. Therefore, the external dynamic knowledge base and the knowledge inside large models are complementary. We also specifically target the I gent memory market, providing a complete solution for short-term, long-term, and permanent memory.

Q: What is the company’s plan for expanding overseas markets?

A: We are currently advancing the expansion across Southeast Asia, the Middle East, and the Americas in parallel. The promotion mainly relies on local channels and other partner channels. In terms of business, we build services on top of existing cloud resources. Two major advantages are highlighted: first, a new incremental market and technology concept, providing data processing services for gent; second, high cost performance—while significantly improving performance, it effectively reduces overall usage costs. Based on feedback from various channels, the market response has been generally positive.

Q: After HBF appears, what is the iteration cadence of the company’s cognition database product?

A: The connection speed between HBF and the GPU is very fast, making it suitable for storing model weights. However, the KV cache still needs to be placed externally for now. For GPU databases, this is actually an advantage: currently the data is stored in GPU memory, and in the future it can be migrated to HBF, allowing the overall capacity to be further improved. In terms of product iteration cadence, the single-node version has been developed and is expected to be released soon. In the second half of this year, a multi-GPU version will be rolled out to further enhance compute capability. The data volume can also achieve linear scaling, with data still stored in GPU memory or in memory. By the end of the year, as the hardware matures, we will adopt I SSD, sinking both data and indexes to the SSD.

Q: Will the company conduct leaderboard tests later on?

A: Currently, the company’s cognition database is the single-GPU version. After the multi-GPU version is released, we plan to complete full-scale testing of TPC-DS 1TB. If it passes smoothly at that time, it will indicate that the database’s core functions are fully in place.

Hylink (688031) main business: The company is an enterprise-level big data foundational software developer. It provides foundational software and services across the full lifecycle of data, including integration, storage, governance, modeling, analysis, mining, and circulation. It has formed a software product matrix covering big data and cloud infrastructure platforms, distributed relational databases, and data development and intelligent analysis tools. It supports customers and partners in developing data application systems and business application systems, helping customers achieve digital transformation.

The 2025 annual report of Hylink shows that the company’s主营 revenue for the year was 448 million yuan, up 20.47% year over year; the net profit attributable to shareholders was -245 million yuan, up 28.62% year over year; and the non-recurring profit and loss (扣非) net profit was -264 million yuan, up 30.55% year over year. In particular, in the fourth quarter of 2025, the company’s单季度主营 revenue was 222 million yuan, up 37.35% year over year; the单季度 net profit attributable to shareholders was -31.9438 million yuan, up 36.79% year over year; the单季度扣非 net profit was -35.0775 million yuan, up 44.71% year over year. The liability ratio was 26.78%, investment income was 5.45M yuan, finance costs were 65.8k yuan, and gross margin was 54.14%.

In the past 90 days, this stock had 1 institution issue a rating, with 1 “buy/increase holdings” rating. Over the past 90 days, institutions’ average target price was 170.94.

The following are detailed profit forecast details:

Margin financing and securities lending data show that over the past 3 months, net margin financing inflow for this stock was 442 million yuan, with the margin financing balance increasing; net margin securities lending outflow was 330.1k yuan, with the margin securities lending balance decreasing.

The above content has been compiled from publicly available information by Securities Star and generated by an AI algorithm (Cybersecurity record: 310104345710301240019). It does not constitute investment advice.