As a national-level Specialized, Refined, Differential and Innovative "Little Giant" enterprise headquartered in Wuhan Optics Valley, Homsh has officially launched its self-built Lightyear AI Computing Center. Expanded and renovated from the original in-house computing lab, it houses five server cabinets numbered LY-00 to LY-04, with reserved slots for future expansion.
This is more than just a data center upgrade. Externally, it serves as the computing hub for "AI + identity security" within Optics Valley’s artificial intelligence industry landscape. Fundamentally, it responds to the higher requirements for recognition algorithms posed by China’s trusted digital identity development. Internally, it underpins Homsh’s algorithm, chip and product iterations over the past years, and will define the scope of our R&D progress in the next phase.
On-site view of the Lightyear AI Computing Center cabinets. Five cabinets LY-00 to LY-04 feature unified nameplates, with reserved expandable positions.
Optics Valley is prioritizing artificial intelligence as one of its key development directions. According to public reports, as of this year, Optics Valley is home to over 800 AI enterprises with an industrial scale exceeding 60 billion yuan, and the installed intelligent computing power in the region has surpassed 5,000P. The Optics Valley AI+ Action Plan released in March this year sets out ten key tasks, including building an integrated AI aggregation and smart hardware platform, launching an AI scenario exhibition hall, and promoting the opening of application scenarios by authorities such as public security departments. Optics Valley is also continuously strengthening its full-chain advantages in "sensing – storage – transmission – computing".
These directions are highly aligned with Homsh’s technical roadmap. Homsh has built a complete industrial chain covering proprietary algorithms, dedicated chips, embedded modules, smart terminals and back-end matching platforms. "Sensing" refers to iris capture; "computing" covers recognition algorithms and edge-side inference cores, which are embodied in AI smart terminals deployed in public security, drug control, smart access, finance and government affairs scenarios.
In a sci-tech park known for computing power and AI terminals, identity recognition is an essential gateway for many application scenarios. The Lightyear AI Computing Center provides Homsh with a self-controlled training foundation for algorithm R&D along the industrial chain, and contributes to Optics Valley’s AI ecosystem as a player focused on the identity security segment.
Effective July 15, 2025, the Measures for the Administration of National Network Identity Authentication Public Services, jointly issued by six authorities including the Ministry of Public Security and the Cyberspace Administration of China, officially came into force. The Measures explicitly put forward the trusted digital identity strategy and the protection of citizens’ identity information security, emphasizing the principle of "minimal provision" and making citizens’ identity information "usable but invisible". During online ID application, facial recognition is part of the identity verification process — biometric recognition has become a key component of the national digital identity system.
For identity to be trusted, the recognition itself must be reliable. This requires at least three layers of safeguards:
1. Accurate recognition For large-scale identity matching, false acceptance rate and false rejection rate must stand up to rigorous testing. National standards for iris recognition image technical requirements, algorithm evaluation methods and capture devices have been issued in the public security field. Algorithms must undergo repeated training and evaluation in alignment with these standards.
2. Robust anti-spoofing Presentation attacks using photos, printouts and prosthetics pose real threats to identity verification. Presentation Attack Detection (PAD) models require continuous training and adversarial iteration with large sample volumes.
3. Secure data handling Biometric traits cannot be replaced once leaked. Performing recognition on terminal devices to reduce the upload and retention of raw biometric data aligns with the "usable but invisible" principle, which requires efficient edge-side models and inference cores.
All three layers rely on computing power to iteratively refine models. Iris recognition is renowned for its high accuracy and stability. For it to play a greater role in the broader trusted digital identity landscape, computing power is an indispensable infrastructure.
Long before the "Lightyear AI Computing Center" was named, this computing infrastructure had been operating as an in-house computing lab for a long time, serving as the training foundation for many of Homsh’s core achievements:
● Self-developed RDPU core
The AI inference core independently developed by Homsh’s R&D team, with related model training and verification completed here. It is now applied in iris recognition to enable efficient on-device recognition.
● PhaseIris algorithm system
Training and iteration of modules including localization, segmentation, encoding, matching and quality assessment.
● PAD liveness detection
Training of anti-presentation attack models, supporting the "robust anti-spoofing" layer of identity verification.
● OVAI artificial intelligence technology
Training of deep learning and neural network related models.
● Product line models
Algorithm optimization for products including capture devices, access control gates, smart locks and identity matching systems.
The continuous support of the computing lab over the years has been a prerequisite for the successive rollout of these achievements from algorithms to chips and products.
The iteration of PhaseIris and OVAI never stops, with ongoing training and parameter tuning running on the Lightyear AI Computing Center. The primary goal of the renovation and expansion is to ensure no interruption to this core R&D line and no degradation in training efficiency. Every improvement in recognition accuracy and liveness detection capability directly enhances the reliability of products in high-security scenarios such as public security, finance and government affairs.
This expansion addresses two clear and essential demands: First, larger-scale model training. As digital identity scenarios scale up, matching databases grow larger, operating environments become more complex, and attack methods evolve. Algorithms require larger models, more data and more training rounds to maintain accuracy and security standards. The capacity of the original computing lab would have become a bottleneck. Second, R&D for new product lines. AI terminals are a key focus of Optics Valley’s development layout. Homsh’s new product R&D requires independent and sufficient computing support, and can no longer compete for resources with core algorithm iteration.
The reserved expansion slots in the Lightyear AI Computing Center are pre-allocated to accommodate these two types of demand.
Roughly estimated under the unified FP32 general-purpose floating-point metric, the total computing power of the Lightyear AI Computing Center cluster is approximately 250 TFLOPS (equivalent to about 0.25 PFLOPS), with an estimated range of 240–280 TFLOPS. The exact figure will be updated as equipment continues to expand.
Total cluster computing power roughly estimated under the unified metric, updated continuously with equipment expansion.
In terms of storage, the Lightyear AI Computing Center features a 250TB-class distributed storage system spanning multiple devices in the cluster, supporting training data and model version management. Some devices and unmounted partitions are still under verification and supplementation, with their capacities not yet included in the total scale.
The distributed storage system is under continuous expansion; some devices and unmounted partitions are still under verification.
The Lightyear AI Computing Center currently has 5 server cabinets with unified nameplates and reserved expansion slots, adopting a phased construction and rolling deployment approach.
The scale overview covering cabinets, computing power and storage consists of estimated data under unified metrics for the current stage.
From the in-house computing lab to the Lightyear AI Computing Center, the scale has changed, but its role as the foundation for iris recognition algorithm training and iteration remains unchanged. Rooted in Optics Valley, a leading hub for artificial intelligence, and aligned with the long-term demand for national trusted digital identity development, Homsh will leverage this self-built computing center to further deepen and solidify its identity recognition capabilities of "high accuracy, robust anti-spoofing and secure data handling". All scale metrics will be continuously verified and updated, and we will share subsequent progress in a timely manner.
WuHan Homsh Technology Co., Ltd. (HOMSH), founded in 2011 and headquartered in Wuhan Optics Valley, is a national-level Specialized, Refined, Differential and Innovative "Little Giant" enterprise and a national high-tech enterprise. For 15 years, it has focused on the R&D of iris recognition and biometric chips.
Homsh’s core advantage lies in its complete technical chain covering proprietary algorithms, dedicated chips, embedded modules, smart terminals and back-end matching platforms. It owns the self-developed PhaseIris iris recognition algorithm and OVAI artificial intelligence technology, and continues to advance its self-developed AI inference core to enable efficient on-device identity recognition.
The company’s product portfolio includes iris capture devices, dual-mode iris-face access control terminals, gate equipment, smart locks, iris USB keys and large-scale identity matching systems. It provides end-to-end solutions covering identity capture, database establishment, verification and business management, serving sectors including public security, drug control, smart access, finance and government affairs, and high-security facility management.
For equipment manufacturers and system integrators, Homsh also provides chips, modules, SDKs and technical integration support to help partners embed iris recognition capabilities into turnstiles, access control systems and various dedicated terminals. We welcome inquiries from partners with needs for highly reliable identity authentication, smart hardware integration and industry applications.
