Intel Corporation (INTC.US) CEO Lip-Bu Tan Praises China's Cutting-Edge AI Models, Firmly Supports Open Source for Efficiency Gains

Intel Corporation (INTC.US) CEO Lip-Bu Tan said on Monday that China's cutting-edge artificial intelligence (AI) models have a lower cost per token and have already achieved "very rapid" adoption. According to reports, Tan said that models such as DeepSeek, Kimi, and GLM are "already good enough" in terms of performance. He also said: "I am a firm supporter of open source. This is the best way to drive efficiency improvements." He added that ultimately, which models are suitable for which application scenarios will be decided by the market. Tan's view aligns with that of NVIDIA Corporation (NVDA.US) CEO Jensen Huang. Huang has repeatedly defended China's open-source models. In July this year, Huang said that the models from DeepSeek, Alibaba Group Holding Limited Sponsored ADR, and Moonshot AI are all "very excellent" and should be adopted. He also said that open-source models can expand the scope of AI applications rather than reduce demand for NVIDIA Corporation chips. An important feature of open-source AI models is their relatively lower cost of use. For Intel Corporation, broader adoption of low-cost AI models could drive wider deployment of AI, thereby increasing market demand for the CPUs, servers, and data center infrastructure needed to run these workloads, which is exactly the direction Intel Corporation is currently pursuing. At the same time, Tan has positioned Intel Corporation's data center business as a beneficiary of agentic AI. Data shows that Intel Corporation's second-quarter revenue increased 25% year over year to $16.13 billion, the fastest growth rate since 2011 and better than the market expectation of $14.42 billion. Among this, revenue from the Data Center and AI business segment reached $6.3 billion, surging 59% year over year, more than twice the company's overall revenue growth rate and above the market expectation of $5.37 billion. As the industry shifts from training large models to deploying at scale agentic AI capable of autonomously executing tasks, demand for general-purpose CPUs is surging. Agentic AI can perform multi-step tasks, thus requiring substantial computing power. Traditional AI inference can rely heavily on GPUs for matrix computations, but agentic workflows break a task into multiple stages such as planning, retrieval, file reading and writing, rule validation, code execution, tool invocation, and result verification. These steps are naturally more dependent on general-purpose CPUs, memory access, I/O, and system scheduling. In traditional inference scenarios, the ratio of GPU to CPU resource demand is about 7:1, while in agentic workflows it may approach 1:1 or even be more CPU-heavy. This does not mean GPUs are unimportant, but rather that AI data centers are moving from "GPU islands" into an era of heterogeneous systems comprising "CPU + GPU + ASIC + optical interconnect network infrastructure + storage." In addition, some analysts believe that the development of open-source models can also help Intel Corporation reduce dependence on a single AI ecosystem. Currently, the AI infrastructure market is largely built around NVIDIA Corporation's CUDA ecosystem. The development of open-source models has the potential to enable enterprises to form more diversified AI infrastructure. For example, enterprises may choose NVIDIA Corporation GPUs + Intel Corporation CPUs, or Intel Corporation CPUs + other AI accelerators, or CPUs handling general-purpose computing + specialized accelerators handling AI inference. For Intel Corporation, the most important thing is not necessarily to become the "second NVIDIA Corporation," but to make itself as indispensable as possible as a supplier of general-purpose computing infrastructure for AI data centers.