From Edge AI to Enterprise On-Premise AI — Skymizer HTX301, the world’s first on-premise AI inference chip, powered by HyperThought IP and scaling from Edge to On-Premise to Mini Data Center
TAIPEI, May 27, 2026 /PRNewswire/ — The Institute for Information Industry (III) today announced a strategic collaboration with Skymizer to lead Taiwan’s transition from Edge AI to enterprise-grade On-Premise AI. The partnership’s first deployable product is Skymizer HTX301 — the world’s first AI inference chip purpose-built for on-premise environments — powered by Skymizer’s proprietary HyperThought IP. Combined with III’s standardized middleware and heterogeneous-resource orchestration, HTX301 scales out in every direction: from a single Edge device, to enterprise on-premise racks, to plant-wide Mini Data Centers and group-level clusters. The flagship reference deployment — supporting knowledge management for the New Taipei Metro — launches in parallel and demonstrates the data-sovereignty, millisecond-latency, and cluster-grade scalability advantages of a Taiwan-built on-premise stack.
Background: From Consumer Electronics to B2B Edge AI
Taiwan’s silicon industry has historically excelled in consumer electronics. As global enterprise demand shifts from cloud AI toward Edge AI, building Taiwan-based on-premise solutions has become the pivotal opening for local chipmakers to enter the B2B market. Under a program led by the Industrial Development Administration (IDA) of the Ministry of Economic Affairs (MOEA), III is breaking the historical lock-in to single-vendor ecosystems by developing a standardized middleware layer and a heterogeneous-resource orchestration interface that resolves the integration problem across chip architectures.
When a single device’s compute is insufficient, the management layer can horizontally federate multiple heterogeneous chip nodes — automating scheduling, balancing workloads, and improving distributed execution efficiency to deliver true scalability.
How the Partnership Splits the Work
III brings deep field experience: it audits real-world enterprise pain points during AI transformation and translates them into application specifications and concrete scenario requirements. Taiwan’s chip vendors, in turn, supply the underlying compute foundation; integrated with the IDA program’s enabling technologies, that foundation is packaged into deployable industry services such as Mini Data Centers and AI Boxes — and tuned for performance in domains like manufacturing.
For B2B use cases, on-premise architectures offer advantages the cloud cannot match. Core processes and commercial secrets stay inside the customer’s environment, preserving data sovereignty and security at the highest tier. And millisecond-level response times make these solutions essential for latency-sensitive workloads such as industrial automation and real-time occupational-safety monitoring.
Flagship Product: Skymizer HTX301 — The World’s First On-Premise AI Inference Chip
Skymizer HTX301 is the partnership’s first deployable product and the world’s first AI inference chip purpose-built for on-premise environments. HTX301 is powered by Skymizer’s proprietary HyperThought IP, which redefines large inference workloads — historically assumed to require cloud GPUs — for the enterprise’s own premises. On a single chip, HTX301 delivers enterprise-class inference throughput and low-latency response while keeping sensitive data entirely inside the customer environment.
Combined with III’s standardized middleware and heterogeneous-resource orchestration interface, HTX301 expands from a single Edge node (an Edge AI Box or smart device), to enterprise on-premise racks, and on to Mini Data Center-grade inference clusters built from multiple HTX301 chips. The result is a true “start with one chip, scale out in every direction” path for Taiwan-built on-premise AI — letting enterprises grow compute capacity in step with their business momentum, with no forklift upgrade to existing infrastructure.
Executive Commentary
“III has spent years on the factory floor and in operations centers, and we know exactly where AI deployments stall. HTX301 — the world’s first chip purpose-built for on-premise AI inference — fills a long-standing gap: until now, Taiwan-built solutions have lacked deployable on-premise inference silicon. Combining HTX301 and HyperThought IP with our standardized middleware and orchestration, we will deploy in real environments such as the New Taipei Metro and take Taiwan’s on-premise AI from Edge to On-Premise to Mini Data Center-grade commercial scale.” — Calvin Tsai, Director, Institute for Information Industry
“HTX301 is the world’s first AI inference chip purpose-built for on-premise environments, and the embodiment of years of HyperThought IP development. Together with III, we will scale HTX301 from a single Edge device, to enterprise on-premise racks, to Mini Data Center-grade inference clusters — giving Taiwan a complete path from silicon to system in the global enterprise AI market.” — William Wei, EVP & CMO, Skymizer
Looking Ahead
Through deep industry-academia-research integration and the consolidation of key technologies, the partnership aims to break free from the rigid hardware constraints of the past and give enterprises the ability to expand elastically with their business momentum — propagating Taiwan-built on-premise solutions from single-point applications to plant-wide and group-wide cluster architectures, and realizing the commercial-scale value of Taiwan silicon in the B2B market.
About the Institute for Information Industry (III)
Founded in 1979, the Institute for Information Industry (III) is Taiwan’s principal think tank and technology accelerator for the ICT and software industries. III supports the government in shaping digital policy, drives industry innovation and talent development, and works alongside enterprises on their digital transformation journeys.
About Skymizer
Skymizer is a Taiwan-based technology company specializing in AI inference silicon, compilers, and heterogeneous compute orchestration. Skymizer’s proprietary HyperThought IP is purpose-built for on-premise AI inference and is realized in HTX301 — the world’s first AI inference chip designed for on-premise environments — enabling enterprises to run large inference workloads previously bound to cloud GPUs entirely on their own infrastructure, and to scale out from a single Edge device, to enterprise on-premise racks, to Mini Data Center-grade inference clusters.
