Hon Hai Technology Group, better known as Foxconn, is showcasing its manufacturing and system-integration capabilities for next-generation AI infrastructure at Hon Hai Tech Day 2026 in Taipei. The exhibition puts the NVIDIA Vera Rubin platform and modular NVIDIA DSX POD design at the center of its “Token Factory” strategy, highlighting how the electronics manufacturer is expanding beyond server assembly into rack-scale computing, power delivery, cooling, networking and precision components for AI data centers.
Foxconn Puts Token Factory Strategy at the Center of HHTD26
The seventh annual Hon Hai Tech Day, running October 30–31 at Taipei Nangang Exhibition Center, reflects Foxconn’s effort to position itself as a broader AI infrastructure manufacturing and technology partner. The company presented 321 exhibits, with approximately one-quarter of the exhibition dedicated to its Token Factory zone. The demonstrations covered technologies ranging from precision-engineered circuit boards and cooling systems to complete AI computing infrastructure.
The strategic shift is significant for a company widely associated with contract electronics manufacturing. As AI workloads expand, demand is increasing for integrated systems that combine accelerators, high-speed networking, power management and thermal engineering. Foxconn is seeking to capture a larger role in delivering those systems, rather than supplying only individual components or assembled servers.
Chairman Young Liu described the company’s direction as a transition toward becoming a technology manufacturing platform service provider. Its stated approach combines vertical integration and precision manufacturing across AI, electric vehicles, consumer electronics, robotics and other technology markets.
NVIDIA Vera Rubin Moves From Components to Rack-Scale Systems
A central focus of the exhibition is Foxconn’s work on the NVIDIA Vera Rubin platform. Rather than treating AI servers as standalone machines, the platform brings together processors, GPUs, networking, storage and infrastructure technologies to operate as an integrated AI supercomputer.
NVIDIA’s Vera Rubin architecture combines its Vera CPU and Rubin GPU with NVLink 6 switches, ConnectX-9 networking, BlueField-4 data processing units and Spectrum-6 Ethernet switching. The design is intended to coordinate compute and data movement across large AI systems, supporting demanding training and inference workloads. NVIDIA Technical Blog+1
Foxconn’s exhibition emphasized the manufacturing requirements behind that architecture. Its demonstrations included rack-level AI supercomputer integration, densely populated printed circuit boards with precision-engineered components, and compute tray designs intended to reduce cabling, cooling complexity and mechanical constraints.
The company also highlighted co-packaged optics (CPO), a technology that brings optical connectivity closer to switching silicon. As AI clusters grow, moving data efficiently between processors and systems becomes increasingly important. Optical interconnects are one approach to addressing bandwidth, power and signal-integrity challenges in large-scale infrastructure.
These demonstrations point to the breadth of manufacturing work involved in building AI factories. GPU performance is only one part of system capability; power distribution, cooling, interconnects and component reliability also influence how effectively infrastructure operates at scale.
NVIDIA DSX POD Highlights Modular AI Factory Deployment
Another major exhibit was the NVIDIA DSX POD. The reference design enables partners to integrate Vera Rubin rack-scale infrastructure inside modular containers, complete key validation before shipment and connect the systems to prepared site infrastructure after delivery.
NVIDIA introduced DSX POD in October as a way to accelerate deployment of complete Vera Rubin AI factories. The design moves more integration and validation work into factory environments, potentially reducing the amount of on-site assembly required. NVIDIA has said its partner-built modular systems could shorten deployment at prepared sites from months to weeks.
For Foxconn, the design creates an opportunity to apply its manufacturing scale and system-integration expertise to a broader infrastructure product. Standardized, factory-built systems could help customers deploying AI capacity manage complex installation requirements across multiple locations.
The company also demonstrated an AI data hall simulation showing how compute, networking, power and cooling need to be designed together. The presentation included NVIDIA’s 800 VDC power distribution architecture, which is intended to support the increasing power requirements of next-generation AI infrastructure. NVIDIA
Foxconn’s AI infrastructure subsidiary, Visionbay.ai, is planning facility designs around NVIDIA DSX MaxLPS principles. These are intended to coordinate workload placement and power use to improve AI output within a fixed power envelope. The broader objective is to evaluate computing, electrical and cooling configurations before facilities are built.
Power, Cooling and Manufacturing Become AI Infrastructure Priorities
The exhibition underscores a wider challenge for data center operators: scaling AI compute requires more than procuring additional GPUs. Power availability, thermal management, networking capacity and facility design can all constrain deployment.
NVIDIA’s DSX platform addresses these issues through technologies for AI factory simulation, power management, operations and infrastructure coordination. Its 800 VDC architecture is designed to reduce power distribution losses and support future high-density racks, while DSX MaxLPS focuses on maximizing tokens per watt through dynamic power management. NVIDIA
Foxconn’s demonstrations of cooling systems, high-speed data solutions and precision-engineered components illustrate how manufacturers are adapting their production capabilities to these requirements. The company’s emphasis on cable-free, hose-free and fanless compute tray designs also points to efforts to simplify system assembly and servicing.
The commercial opportunity is extending across the AI infrastructure supply chain. Cloud providers, AI developers and enterprises need systems that can be deployed and operated reliably, while manufacturers are competing to supply increasingly integrated hardware platforms.
Agent Safety and Robotics Extend the AI Ecosystem
Foxconn also used HHTD26 to demonstrate technologies beyond data center hardware. Its AI safety exhibits included NVIDIA’s Open Agent Safety Platform, which is designed to govern AI agent behavior across software and infrastructure.
The platform combines NVIDIA OpenShell, an open-source runtime that establishes boundaries around agent activity, with NVIDIA Sentry, a reference design for monitoring and enforcing security policies through hardware infrastructure. The architecture is intended to help organizations control what agents can access and do, including when they behave unexpectedly.
Foxconn demonstrated potential applications in smart healthcare, manufacturing and smart cities. The exhibition also featured robotics and humanoid systems, alongside precision manufacturing technologies used in products such as AI companion robots and laptop enclosures.
Taken together, the exhibits show how Foxconn is connecting AI infrastructure manufacturing with the systems that use AI, from industrial automation to robotics and smart-city applications. Its long-term positioning will depend on its ability to translate manufacturing scale into reliable, integrated infrastructure and deliver it as AI demand grows.
Market Landscape
AI infrastructure is evolving from individual server procurement toward integrated AI factories, where compute, networking, power, cooling and facility operations are engineered as a coordinated system. NVIDIA’s Vera Rubin and DSX initiatives reflect this shift, while Foxconn is using its manufacturing footprint and vertical integration to expand into rack-scale systems and modular data center infrastructure. The competitive opportunity extends beyond chip supply to system integration, precision components, deployment speed, energy efficiency and operational reliability.
Top Insights
- Foxconn showcased 321 exhibits at HHTD26, with approximately one-quarter focused on its Token Factory AI infrastructure strategy.
- The company highlighted its manufacturing and system-integration work around NVIDIA Vera Rubin rack-scale AI computing systems.
- NVIDIA DSX POD enables partners to factory-integrate and validate modular AI infrastructure before connecting it to prepared data center sites.
- Foxconn’s demonstrations covered co-packaged optics, 800 VDC power distribution, cooling, GPU infrastructure and precision-engineered components.
- NVIDIA Open Agent Safety Platform featured in demonstrations spanning smart healthcare, manufacturing, robotics and smart-city applications.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
