Intel (INTC.US) CEO Chen Liwu praised China's cutting-edge AI models for improving efficiency and firmly supporting open source

The Zhitong Finance App learned that Intel (INTC.US) CEO Chen Liwu said on Monday that China's cutting-edge artificial intelligence (AI) model costs less per token and has been applied “very quickly.” According to reports, Chen Liwu claimed that the performance of models such as DeepSeek, Kimi, and GLM was “good enough.” He also said, “I'm a strong supporter of open source. It's the best way to drive efficiency.” At the same time, he added that in the end, which models are suitable for which application scenarios will be decided by the market.

Chen Liwu's views are in the same camp as NVDA.US (NVDA.US) CEO Huang Renxun. Huang Renxun has defended China's open source model many times. In July of this year, Hwang In-hoon said that the models from DeepSeek, Alibaba, and Moonshot AI are all “excellent” and should be applied. He also said that the open source model can expand the scope of AI applications rather than reduce the demand for Nvidia chips.

An important feature of the open source AI model is that the cost of using the model is relatively lower. For Intel, more widespread adoption of low-cost AI models may drive AI to be deployed more widely, thereby increasing market demand for the CPUs, servers, and data center infrastructure needed to run these workloads, and this is the direction Intel is currently working towards.

Meanwhile, Chen Liwu has positioned Intel's data center business as a beneficiary of intelligent AI. According to the data, Intel's second-quarter revenue increased 25% year-on-year to US$16.13 billion, the fastest growth rate since 2011, better than market expectations of US$14.42 billion. Among them, the revenue of the data center and AI business division reached 6.3 billion US dollars, a sharp increase of 59% over the previous year. The growth rate was more than double the company's overall revenue, which was higher than the market's forecast of 5.37 billion US dollars.

As the industry shifts from training large models to large-scale deployment of intelligent AI that can perform tasks autonomously, demand for general-purpose CPUs has experienced a blowout. Intelligent AI can perform multi-step tasks, so it requires a lot of computational power. Traditional AI inference can rely heavily on GPUs to complete matrix computation, but the intelligent workflow will split a task into multiple stages such as planning, retrieval, file reading and writing, rule verification, code execution, tool calls, and result verification. Naturally, these processes rely more on general CPU, memory access, I/O, and system scheduling. In traditional inference scenarios, GPU and CPU resource requirements are about 7:1. In the intelligent workflow, the CPU may be closer to 1:1 or even heavier. This does not mean that GPUs are unimportant, but rather that AI data centers have moved from “GPU silos” to an era of heterogeneous systems of “CPU+GPU+ASIC+optical interconnect network infrastructure+storage”.

Furthermore, some analysts believe that the development of an open source model can also help Intel reduce its dependence on a single AI ecosystem. Currently, the AI infrastructure market is largely built around the Nvidia CUDA ecosystem. The development of the open source model has an opportunity for enterprises to form more diversified AI infrastructures. For example, an enterprise may choose Nvidia GPU+Intel CPU, Intel CPU+other AI accelerators, or the CPU is responsible for general computing+dedicated accelerators for AI inference. For Intel, the most important thing is not to have to become the “second Nvidia,” but to make itself an indispensable general computing infrastructure supplier for AI data centers as much as possible.