Hon Hai Kaohsiung Super Computing Center
Taiwan’s fastest AI supercomputer built by Foxconn and Nvidia
Foxconn and Nvidia are constructing Taiwan’s most powerful AI supercomputer at the Hon Hai Kaohsiung Super Computing Center. The system uses Nvidia Blackwell GB200 architecture to deliver over 90 exaflops of performance for cancer research, large language models, and digital twin simulations.
- Delivers over 90 exaflops of AI performance
- Uses 4,608 Nvidia Blackwell GPUs in 64 racks
- First phase operational by mid-2025
- Supports cancer research and LLM development
- Taiwan’s fastest supercomputer upon completion
What it is
The Hon Hai Kaohsiung Super Computing Center hosts a new artificial intelligence supercomputer developed through a partnership between Foxconn and Nvidia. Located in Kaohsiung, this facility will house Taiwan’s fastest computing system. The machine is designed to handle high-performance computing tasks and advanced AI workloads. It aims to support industries ranging from healthcare to manufacturing by providing massive computational power. The project highlights Foxconn’s shift from pure contract manufacturing to owning significant AI infrastructure.
How it came about
Foxconn, also known as Hon Hai Technology Group, announced the project during its Tech Day event in 2024. The deal was initially finalized at the Computex trade show in Taipei earlier that year. Unlike previous national efforts such as the Taiwania series managed by the National Center for High-Performance Computing, this system is privately owned. Construction started immediately after the announcement. The first phase is scheduled to become operational in mid-2025, with full deployment expected by 2026.
The compute
The supercomputer utilizes Nvidia’s Blackwell architecture, specifically the GB200 NVL72 data center platform. It comprises 64 racks containing a total of 4,608 Tensor Core GPUs. Each rack integrates 36 Nvidia Grace central processing units and 72 Blackwell graphics processing units. These components connect via Nvidia NVLink technology, which provides 130 terabytes per second of bandwidth. The system delivers a peak theoretical AI performance of more than 90 exaflops and 207.4 petaflops for traditional floating-point calculations.
Where the GPUs are
The hardware is housed at the Hon Hai Kaohsiung Super Computing Center in Kaohsiung, Taiwan. This location serves as the primary site for Foxconn’s high-performance computing initiatives. The facility supports the physical infrastructure required for the 64-rack cluster. While Foxconn operates factories globally, including in mainland China and Mexico, this specific supercomputing resource is centralized in Kaohsiung to support local and global digital twin and AI development needs.
Customers and deals
Foxconn intends to use the supercomputer for its own internal operations. Key applications include cancer research, the development of large language models, and the creation of digital twins for smart manufacturing. The system will also support robotics platforms and advanced driver-assistance systems for electric vehicles. By owning this infrastructure, Foxconn can accelerate its own product development without relying on external cloud providers or public research networks.
Money and ownership
Neither Nvidia nor Foxconn disclosed the exact cost of the supercomputer. Industry estimates suggest a street price of approximately $3.4 million per rack for the GB200 NVL72 systems. This would place the total hardware cost around $220 million for the 64 racks. Foxconn reported revenues of NT$6,162 billion in 2023, indicating it has the financial capacity to fund such a large-scale infrastructure investment independently of government subsidies.
Its place in AI
This supercomputer represents a significant milestone for Taiwan’s computing capabilities, surpassing previous national systems like Taiwania 2. It positions Foxconn as a major player in AI infrastructure, not just hardware manufacturing. The system’s performance rivals international peers such as Germany’s Jupiter supercomputer. It demonstrates how large electronics manufacturers are integrating exascale-class computing into their industrial strategies to maintain competitiveness in the AI era.
Questions people ask about Hon Hai Kaohsiung Super Computing Center
Who owns and operates the supercomputer?
The supercomputer is owned and operated by Foxconn, also known as Hon Hai Technology Group. It is built in partnership with Nvidia, which supplies the underlying Blackwell GPU architecture and networking technology.
What is the computing power of the system?
The system delivers more than 90 exaflops of AI performance at full capacity. It also provides 207.4 petaflops of peak FP64 performance, making it the fastest supercomputer in Taiwan.
When will the supercomputer be fully operational?
Construction has already begun. The first phase is scheduled to be operational by mid-2025. Full deployment of all 64 racks and 4,608 GPUs is expected to be completed in 2026.
What hardware does the supercomputer use?
It uses Nvidia’s GB200 NVL72 platform. This includes 64 racks, each with 36 Grace CPUs and 72 Blackwell GPUs, connected via NVLink technology for high-bandwidth communication.
Good reading and sources
There is no English Wikipedia article on Hon Hai Kaohsiung Super Computing Center yet, so the facts on this page come from the sources listed above: datacenterdynamics.com/en/news/nvidia-and-foxconn-to-build-taiwans-fas, nextplatform.com/ai/2024/10/09/taiwans-fastest-ai-supercomputer-goes-t. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.
Like Hon Hai Kaohsiung Super Computing Center
All AI supercomputers
🇲🇾 Malaysia and 4 moreYTL AI Cloud
Malaysian AI cloud and supercomputer run from a solar-backed Johor data centre parkNeocloud · Nvidia Blackwell · 2024
🇺🇸 United StatesPrometheus
Meta's first multi-gigawatt AI training clusterSupercomputer · 2026
🇺🇸 United StatesHorizon
Largest academic supercomputer in the United StatesSupercomputer · Nvidia Blackwell · 2026