Hybrid AI architecture integrates Phison’s aiDAPTIV technology with intelligent workload routing to reduce token inference costs by up to 70%
Taipei, Taiwan, May 25, 2026 — ASUS today announced that it has integrated a hybrid AI architecture across its entire lineup of commercial devices, including ASUS ExpertBook laptops, ASUS ExpertCenter desktops, and ASUS NUC mini PCs.I Designed for seamless on-premises deployment, this new architecture aims to help enterprises balance performance and cost when deploying generative AI applications, enabling more practical and scalable AI deployments across diverse work environments.
The cost of inference-related tokens continues to rise as enterprises accelerate the adoption of large-scale language models (LLMs) and AI agent-based applications. Purely cloud-based approaches not only incur high, unpredictable costs, but also operational challenges, making cost efficiency a key barrier to large-scale AI adoption.
To address these challenges, ASUS is introducing a hybrid AI architecture that dynamically distributes workloads between local devices and the cloud. By processing some of the AI tasks locally and reserving the cloud for more complex workloads, this architecture maintains performance while significantly reducing overall inference costs and increasing deployment flexibility across different device form factors.
“As enterprises expand their AI deployments, balancing performance and cost has become a key challenge,” said Brian Zhang, general manager of ASUS Commercial PC BU. “With Hybrid AI, we aim to move more AI processing to the device, reduce dependence on cloud resources, and improve both efficiency and practicality. Extending this capability to laptops, desktops, and mini PCs reflects our commitment to evolving AI PCs from standalone devices to scalable real-world solutions.”
At the technology level, Phison’s aiDAPTIV is integrated into the architecture.™ Memory expansion technology enables devices with limited hardware resources to support medium to large language models locally. This removes traditional memory constraints and enables AI workloads that previously required high-end infrastructure to run on commercial PC platforms. Additionally, the gateway-based routing mechanism intelligently allocates tasks based on complexity, favoring local processing wherever possible to further optimize efficiency.
Results from PinchBench, a benchmark system for evaluating LLM models as OpenClaw coding agents, show that a hybrid inference approach can reduce inference costs by up to 70% for medium and large models (such as 26B and 35B) while maintaining performance, providing a more sustainable and cost-effective model for enterprise AI deployments.
“aiDAPTIV enables local execution of larger AI models by overcoming traditional memory limitations and significantly reduces overall compute costs through hybrid inference,” said KS Pua, CEO of Phison. “Our collaboration with ASUS demonstrates that this technology can be effectively deployed across commercial platforms to provide a scalable and cost-effective solution for enterprise AI deployments.”
With a diverse portfolio spanning laptops, desktops, and mini PCs, ASUS is expanding its hybrid AI capabilities across a wide range of enterprise use cases. These include multilingual translation, business email writing, summarizing meeting notes, summarizing contracts and long documents, internal knowledge base Q&A, automated customer service and answering FAQs, as well as CRM recordkeeping and sales support. By handling these tasks locally, enterprises can reduce their reliance on cloud token consumption while improving data privacy and response efficiency, creating a more cost-effective approach to AI deployment. Combined with scenarios such as on-device AI assistants, log analysis, and troubleshooting support, hybrid AI architectures give organizations more flexibility in deploying AI according to their operational needs, accelerating the transition from pilot projects to large-scale implementations.
ASUS continues to advance AI PC and commercial computing platforms by integrating hardware, software, and ecosystem partnerships. The company aims to provide an AI infrastructure that balances performance, cost, and flexibility, enabling enterprises to scale AI adoption and drive long-term digital transformation.
