During Computex 2026 in Taipei, Intel, Foxconn, and AI accelerator chip maker SambaNova jointly announced that the three companies will work together to build rack-scale AI infrastructure based on Intel Xeon processors, targeting inference deployment scenarios in data centers, hyperscale data centers, and AI computing centers. The rack integrates Intel Xeon processors with the SambaNova SN-50 RDU, focusing on high cost-performance and energy-efficient AI inference. Foxconn will be responsible for full system integration and will separately launch rack variants with a high CPU density core, targeting inference, data processing, and hybrid AI workloads that do not require additional hardware acceleration. Foxconn Chairman Young Liu said the three-way collaboration will combine the strengths of computing platforms, system integration, and the global supply chain to accelerate the global deployment of AI technologies.
The strategic logic behind this collaboration lies in the shift of AI workloads from model training acceleration to inference. Analysts point out that during the training era, the ratio of CPUs to GPUs was around 1:4. However, in the agentic AI phase, this ratio approaches 1:1 or even lower, significantly elevating the role of CPUs in AI architecture. This provides a critical window for Intel, which is advancing its strategic transformation. At the expo, Intel CEO Lip-Bu Tan stated that Intel is pushing forward innovation “from chip to system level” to meet the accelerating needs of inference, agentic AI, and physical AI. On Foxconn’s side, its AI server revenue grew approximately 170% year-over-year last year, with a market share of about 40%, and the cloud and networking division has become the company’s largest business segment. The collaboration will also extend into areas of physical AI such as robotics, autonomous driving, smart cities, and smart manufacturing, and will explore joint development of custom chip design services.