Caught between corporate concerns about data breaches and the high cost of computing power, "ground-based private small-scale models" are emerging as a new breakthrough for the AI industry in 2026. APMIC, a Taiwanese brand of enterprise-independent AI solutions, announced the successful completion of a $700 million Series A funding round. This round was led by Pasadena-based private equity firm AIPO, with participation from global cybersecurity provider Trend Micro.
APMIC stated that the funds will be fully invested in the research and development of private AI mini-models and the application of AI agents, targeting the government and financial sectors with extremely high cybersecurity requirements, and promoting the practical implementation of enterprise AI.
Targeting the pain points of finance and government: "Customized small-scale models" that are cloud-free and low-cost.
Although public cloud platforms like OpenAI offer powerful large language model (LLM) capabilities, uploading sensitive data to the cloud remains a difficult cybersecurity red line for financial institutions and government agencies to cross.
APMIC founder and CEO Po-Han Wu stated frankly, "2026 will be the year of explosive growth for ground-based private AI mini-models." Market demand is gradually shifting from large-scale models that rely on public clouds to small-scale model architectures that can be deployed on-premises and are easy to optimize. APMIC's strategy is to create optimized small-scale model architectures and AI agent applications for these highly compliant scenarios, ensuring that enterprises can implement highly customized AI solutions without data leakage.
Compared to general-purpose large models, these smaller models optimized for specific industry scenarios not only offer higher inference efficiency and lower computational costs, but also provide greater advantages in control and fine-tuning. AIPO's head, Liu Mei-hsiu, also emphasized that as companies increasingly prioritize data sovereignty and cost-effectiveness, the demand for localized private small models will continue to grow. AIPO is extremely optimistic about APMIC's potential for simultaneous market expansion in both Taiwan and the United States, and believes that APMIC is a team capable of seizing this transformation opportunity.
Partnering with Trend Micro for a technological breakthrough: Introducing NVFP4 precision, halving memory requirements.
Another technological highlight of this round of financing is the significant breakthrough achieved by APMIC and Trend Micro in "quantitative security modeling technology".
As NVIDIA's new-generation architecture drives ultra-low precision computing, the research team has successfully compressed and quantized the open-source security model, which originally used 16-bit floating-point (FP16) precision, into an NVIDIA 4-bit floating-point (NVFP4) precision model. This technological breakthrough has brought remarkable benefits: the model's memory requirements have been reduced by more than half, and tests have shown that the decrease in question-answering accuracy can be controlled within 2%.
This means that the hardware deployment threshold and requirements for enterprises will be significantly reduced when deploying AI on-premises, while also accelerating the application of AI in cybersecurity scenarios. This technological achievement will also simultaneously support APMIC's enterprise-owned AI platform product, "PrivStation," launched last year. Furthermore, continuing Trend Micro's commitment to embracing the developer community, APMIC also announced that it will open-source this quantized small model to promote technological innovation and ecosystem development in cybersecurity AI.
