Foxconn Builds FoxBrain, Its Own AI Model
Nvidia provided support through its Taiwan-based supercomputer and technical consulting
The world’s largest contract electronics maker, Foxconn, said Monday it has built its own large language model with reasoning capabilities, developed in-house and trained in just four weeks.
Initially designed for internal use within the company, the artificial intelligence model, called FoxBrain, can serve functions including data analysis, mathematics, reasoning and code generation, the company said.
Foxconn said Nvidia NVDA 1.92%increase; green up pointing triangle provided support through its Taiwan-based supercomputer and technical consulting, enabling successful model training.
The company said it intends to open-source the model for collaborations with industry partners. It envisions FoxBrain driving advancements in manufacturing and supply chain management.
The model “prioritized optimized training strategies over simply throwing computing power” at the problem, said Yung-Hui Li, director of Foxconn’s artificial intelligence research center.
Using 120 Nvidia H100 graphics processing units, Li’s team finished training FoxBrain in about four weeks, the company said.
Foxconn, best known for assembling Apple’s iPhones, has released some parameters of FoxBrain. The company said further information would be revealed at Nvidia’s annual technology event in mid-March.
Based on Meta’s Llama 3.1 structure, the company said FoxBrain was Taiwan’s first large language model with advanced reasoning, designed and optimized for traditional Chinese, the form of the language used in Taiwan and some overseas Chinese communities.
The company said its model’s performance was slightly behind some models of China’s DeepSeek but was approaching world-class levels.
Foxconn, facing challenges in its core electronics manufacturing business due to industry shifts and declining profitability, has been diversifying into areas such as AI and electric vehicles.