Share this article on:
The AI market has expanded once more, with Taiwanese company Foxconn announcing the launch of its first large language model (LLM), FoxBrain.
Foxconn is the world’s largest contract electronics manufacturer, most notably responsible for producing NVIDIA AI servers and assembling iPhones for Apple.
In a statement, Foxconn, also known as the Hon Hai Technology Group, announced that the Hon Hai Research Institute launched FoxBrain earlier this month, marking Taiwan’s entry into the AI race.
“FoxBrain not only demonstrates powerful comprehension and reasoning capabilities but is also optimised for Taiwanese users’ language style, showing excellent performance in mathematical and logical reasoning tests,” said Foxconn.
Trained using 120 NVIDIA H100 GPUs, which are the standard for AI training, and was scaled using NVIDIA Quantum-2 InfiniBand Networking, FoxBrain was finished in four weeks with a lower-cost training method and greater efficiency compared to “inference models recently launched in the market”. It is unclear if this means it was cheaper to develop than DeepSeek’s R1 model, the last training run of which only cost US$5.7 million.
“In recent months, the deepening of reasoning capabilities and the efficient use of GPUs have gradually become the mainstream development in the field of AI,” said Hon Hai Research Institute AI Research Centre director Dr Yung-Hui Li.
“Our FoxBrain model adopted a very efficient training strategy, focusing on optimising the training process rather than blindly accumulating computing power.
”Through carefully designed training methods and resource optimisation, we have successfully built a local AI model with powerful reasoning capabilities.”
While Foxconn said that the performance falls slightly short of DeepSeek’s distillation model, it is very close to the high standard set by AI giants.
Additionally, while originally intended for internal applications, Foxconn said it intends to expand its applications for supply chain management, manufacturing and more and will eventually make the technology open source.