Yating Intelligence launched FedGPT, a sovereign AI platform with full on-premises deployment, targeting 72% of enterprises needing local AI
On June 30, 2026, Taiwan AI Labs and its strategic partner Yating Intelligence Co., Ltd. officially unveiled the "2026 Taiwan Enterprise AI Transformation Insight Report." Based on 562 surveys covering over ten industries including education, media, healthcare, government, manufacturing, and finance, the report found that only 15% of surveyed enterprises widely use AI across different departments, while the remaining 85% are still in the small-scale trial or evaluation stage. More critically, as high as 72% of enterprises explicitly require on-premises AI solutions, and 51% of enterprises worry about data leakage risks. To respond to this market demand, Yating Intelligence launched the FedGPT Sovereign AI Platform, emphasizing full on-premises deployment, data autonomy, and system autonomy, and has already implemented a meeting assistant application internally at Taiwan AI Labs.
FedGPT (Federated GPT) is an enterprise-specific multimodal generative AI platform built on federated learning technology and a self-developed large language model (LLM). Its core technical advantage lies in "data does not leave the organization, sensitive data does not go to the cloud." The system is trained on over 100 billion tokens of Traditional Chinese corpus, with superior understanding of Taiwan's knowledge and context compared to international large models. FedGPT strictly adheres to the five pillars of Sovereign AI: data sovereignty ensures sensitive data stays within the organization; model sovereignty allows fine-tuning open-source or using local models; computing sovereignty adopts local compute power; system sovereignty integrates proprietary systems; and governance sovereignty aligns with international AI governance standards. Its two highlight applications include: Meeting Assistant, which connects employee calendars, automatically records meeting content, and generates summaries and action items; and Universal Secretary, which connects administrative workflows to help employees inquire about reimbursement, leave, business trips, etc., and automatically submits requests. Additionally, FedGPT supports Flow process templates, such as bug reporting and tracking, operational report consolidation, and can be customized according to industry characteristics, for example, connecting hospital information systems to generate clinical summaries.
The launch of FedGPT will directly impact Taiwan's healthcare, finance, media, and government industries, which are highly sensitive to data privacy. The report points out that public cloud AI has three fatal limitations: 34% of enterprises believe it cannot adjust judgment logic according to business needs; 39% believe AI cannot understand industry terminology; and 24% worry that public cloud AI stopping service will cause application paralysis. FedGPT effectively solves these problems through on-premises deployment and local model fine-tuning. Currently, the platform has been tested internally at Taiwan AI Labs for the Meeting Assistant and is collaborating with Taiwan Medical Imaging Co., Ltd. to promote federated learning, replacing data sharing with model sharing. In the future, FedGPT will expand to more enterprise scenarios, including HR system integration with facial recognition access control, automatic inquiry of attendance status, meeting scheduling, etc., significantly improving enterprise operational efficiency and productivity.
When adopting AI, enterprises should prioritize evaluating the Sovereign AI architecture and avoid uploading sensitive data to public clouds. Recommended protection measures include: First, choose a full on-premises deployment solution to ensure data does not leave the organization; Second, verify whether the model can be fine-tuned to meet industry-specific needs; Third, establish an AI governance framework aligned with GDPR and EU AI Act standards; Fourth, implement anti-hallucination mechanisms to ensure responses are based on enterprise-specific knowledge; Fifth, regularly audit AI systems to ensure traceability and verifiability. Additionally, enterprises should avoid relying on a single multinational cloud service to prevent application failure due to service interruption or model updates. FedGPT provides Workflow automatic decomposition and low-code tools to help enterprises quickly build secure, trustworthy AI processes.