SANTA CLARA, Calif., Aug. 5, 2026 /PRNewswire/ -- Dnotitia Inc. (Dnotitia), an AI data infrastructure and semiconductor company, today announced that its Seahorse AI Storage has won the AI Application Award at the FMS 2026 Best of Show Awards. Dnotitia was also named a finalist in the Startup Business Growth Award category.

The winning entry presented Seahorse AI Storage as an AI-native storage platform for reliable, on-premises enterprise retrieval-augmented generation (RAG). It addresses a practical challenge facing enterprise AI by transforming proprietary structured and unstructured data, including documents, into searchable, traceable knowledge while keeping sensitive information under the organization's control.

Future of Memory and Storage is a leading global conference for the memory, enterprise storage, semiconductor, data center and AI infrastructure industries. According to the event organizers, FMS 2026 is expected to bring together more than 3,500 attendees from over 1,500 organizations and more than 100 exhibitors.

Dnotitia is showcasing the award-winning Seahorse AI Storage alongside a vector-search accelerator card powered by its newly unveiled VDPU chip at Booth 1144. The exhibit demonstrates how Seahorse software and VDPU acceleration can add AI-native retrieval and knowledge capabilities to enterprise storage products.

As enterprises move RAG and AI-agent services into production, demand is growing for storage systems that can prepare, index and serve enterprise knowledge. This creates an opportunity to add AI-native retrieval and knowledge capabilities to existing enterprise storage products.

Seahorse AI Storage brings document processing, knowledge management, vector retrieval, and AI-service integration into a unified platform. It analyzes document structures such as layouts, tables, charts and contextual relationships, and combines semantic search with methods designed to locate specific terms, figures and clauses. Retrieved information can be linked to its original source, enabling users to verify the evidence behind an AI-generated response.

The platform supports cloud deployment as well as on-premises and air-gapped environments, where data security, control and traceability are critical. Unlike a standalone vector database, Seahorse integrates the processes required to prepare, synchronize, retrieve and serve enterprise knowledge, reducing the number of separate systems customers must build and operate.

Dnotitia's VDPU, or Vector Data Processing Unit, is a purpose-built processor for vector search and graph traversal. By processing vector operations closer to stored data, the VDPU is designed to reduce host CPU workload and unnecessary data movement. Dnotitia develops the processor together with its Seahorse vector database and storage software through hardware-software co-design.

At FMS 2026, Dnotitia is publicly unveiling its first VDPU chip, integrated into a vector-search accelerator card. The company is seeking collaboration with global storage vendors and AI infrastructure partners in product integration, jointly developed AI storage systems, enterprise proof-of-concept (PoC) projects and joint commercialization.

"AI bottlenecks are expanding beyond computing performance to data storage and retrieval," said MK Chung, CEO of Dnotitia. "In the AI era, storage is evolving beyond simply retaining data to helping AI find the right information accurately and quickly based on meaning. This award validates Seahorse AI Storage's ability to meet these requirements."

Dnotitia's selection as a Startup Business Growth Award finalist recognizes the company's progress in advancing its VDPU and Seahorse technology stack, semiconductor development and global enterprise engagements.

Dnotitia will exhibit at Booth 1144 at FMS 2026 through Aug. 6. Storage vendors and AI infrastructure partners are invited to meet the company to discuss product integration, joint solution development and enterprise PoC opportunities.

Storage vendors and AI infrastructure partners are invited to meet the company to discuss product integration, joint solution development and enterprise PoC opportunities.

About Dnotitia

Dnotitia develops AI systems, the data infrastructure behind them and semiconductor technologies for data-intensive AI workloads. Its portfolio includes Seahorse, an AI-native data and storage platform combining vector retrieval, enterprise RAG and agent knowledge capabilities, and the VDPU, a processor designed to accelerate vector-data workloads. Dnotitia is headquartered in Seoul and operates a U.S. office in San Jose, California.

SEOUL, South Korea, July 2, 2026 /PRNewswire/ -- Dnotitia Inc. (Dnotitia), a company specializing in long-term memory AI and semiconductor-based AI infrastructure technologies, has released the paper and source code for "STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control." The technology was developed through a joint research effort involving UC San Diego's VVIP Lab and Dnotitia researchers, and the paper was selected as a Spotlight paper at ICML 2026 (International Conference on Machine Learning 2026), one of the world's leading conferences in machine learning.

In the experiments reported in the paper, low-rank compression alone reduced the KV cache by up to 75%. Combined with the mixed-precision quantization method proposed in the paper, STAR-KV compressed the full KV cache by up to 20x. The technology also improves computation speed through custom GPU kernels, increasing attention computation speed by up to 6.9x and overall generation throughput by up to 3.1x. STAR-KV also showed higher accuracy than major existing KV cache compression methods.

KV cache compression has become a key technical challenge in AI infrastructure. As research into reducing the memory bottleneck of long-context AI gains momentum, including the attention around Google's TurboQuant at ICLR 2026, STAR-KV presents a new approach that combines low-rank compression with quantization and GPU execution optimization.

The KV cache is temporary memory stored on the GPU so that a large language model (LLM) does not have to recompute context it has already processed. As AI evolves into agentic systems that use multiple documents, conversation history, code, search results, and outputs from external tools, the amount of context a model must process is growing rapidly. In this environment, the KV cache has emerged as a key bottleneck affecting both GPU memory usage and inference cost.

According to the STAR-KV paper, when a LLaMA-3.1-8B model processes a 128K-token context at a batch size of 4, the KV cache accounts for about 81% of total GPU memory. As long-context AI becomes more widely used, KV cache compression is increasingly viewed as a core AI infrastructure technology for processing long context at lower cost.

ICML, where the STAR-KV paper was accepted, is widely regarded as one of the top international conferences in AI and machine learning, alongside NeurIPS and ICLR. ICML 2026 will be held from July 6 to 11 at COEX in Seoul. This year, 23,918 papers entered review, 6,352 were accepted, and 536 were selected as Spotlight papers. Spotlight papers account for about 2.2% of all reviewed submissions and about 8.4% of accepted papers.

Going forward, Dnotitia plans to further advance STAR-KV for use in real-world AI service environments and explore its application to open-source LLM inference frameworks such as vLLM.

"Technologies that help AI process longer context faster and at lower cost are advancing rapidly" said MK Chung, CEO of Dnotitia. "STAR-KV addresses the core bottlenecks in KV cache capacity and attention processing speed, and Dnotitia aims to contribute to the AI inference ecosystem through open sourcing."

SEOUL, South Korea, June 2, 2026 /PRNewswire/ -- Dnotitia Inc. (Dnotitia), a company specializing in long-term memory AI and semiconductor-based AI infrastructure technologies, today announced the release of DNA 3.0, an enterprise-ready AI language model family, on Hugging Face.

DNA 3.0 is built on Qwen 3.5/3.6, a large language model family released by Alibaba Cloud. Dnotitia applied its own post-training and tuning to the models, enabling organizations to adapt them to their data, workflows, and response requirements.

The release reflects Dnotitia's approach to making open models more useful in real enterprise environments. Rather than using an open model as-is, DNA 3.0 is tuned to deliver more consistent responses, reflect organizational context, and support practical AI agent use cases.

Dnotitia also applied persona training to DNA 3.0, allowing the model to better reflect company information and product context. This is designed to support organizations that want AI models to respond in line with their internal knowledge, service policies, and business requirements.

DNA 3.0 has also been post-trained to reduce certain response constraints and language-mixing issues that may appear when adapting Qwen-based models. This improves usability in Korean-language enterprise and institutional environments, where stable question-answering and workflow support are critical.

The DNA model family is currently integrated into Seahorse Cloud, Dnotitia's AI data platform. Seahorse Cloud converts enterprise documents and unstructured data into AI-ready knowledge assets, enabling semantic search, context-aware answer generation, and AI agent workflows based on enterprise data.

Through DNA 3.0, Dnotitia plans to further strengthen Seahorse Cloud's ability to support enterprise data-driven AI agents. The goal is to help organizations move beyond storing and searching data, and instead turn their information assets into a usable AI knowledge layer for real business workflows.

The DNA 3.0 lineup includes models ranging from lightweight versions to mid- and large-scale Mixture-of-Experts (MoE) models. Released models include 0.8B, 2B, 4B, 9B, 27B, 35B-A3B, and 122B-A10B, allowing users to choose models based on deployment environment, performance needs, and cost considerations.

The 35B-A3B and 122B-A10B models use a MoE architecture, which activates only selected expert modules for each query. This approach helps deliver large-model capabilities while reducing the computational load required for inference. DNA 3.0 also inherits key capabilities from previous generations, including long-context handling, reasoning trace preservation, tool use, and coding support.

"Institutions and enterprises need AI models that can be adapted to their own data and workflow context," said MK Chung, CEO of Dnotitia. "By integrating DNA 3.0 with Seahorse Cloud, we will continue to expand enterprise AI agent capabilities powered by organization-specific data."

Dnotitia has continued to expand its DNA model family since the release of DNA 1.0 in December 2024, followed by DNA-R1, a Korean reasoning-focused model, and DNA 2.0, a Korean agentic AI language model. DNA 3.0 builds on this progression by strengthening enterprise-oriented tuning and product integration through Seahorse Cloud.