With the exponential growth in computing power demands driven by the AI wave, Silicon Valley giants are accelerating their R&D and investment in basic hardware. On Wednesday, Intel announced plans to release 15 new FPGAs this year, breaking its annual product release record for this category.

In early March, Intel released the Agilex7 FPGA F-Tile with FPGA transceivers. Compared to the previous generation of Intel FPGAs, Agilex 7 F-Tile doubles the bandwidth of each channel while reducing power consumption, enabling it to meet the computational requirements of applications such as data centers that require higher data throughput.

FPGA stands for Field Programmable Gate Array, which is a chip that can reconfigure circuits. Through programming, users can change its application scenarios at any time. It can simulate various parallel operations of hardware such as CPUs and GPUs, making it also known as the "universal chip" in the industry.

Currently, AI chips mainly include GPUs, FPGAs, and ASICs. With the success of large language models (LLMs) like ChatGPT, representing advanced AI technologies, data and computational requirements have risen sharply, intensifying the AI chip arms race.

According to Zhejiang Securities research reports, GPUs or CPU+FPGAs support the computational requirements of ChatGPT. GPUs are widely used because of their parallel computing capabilities, making them compatible with both training and inference.

In addition to GPUs, CPU+FPGA solutions can also meet the massive computational requirements of AI. As a programmable chip, the FPGA can be extended for specific tasks, providing some flexibility in the second phase of AI model construction. When combined with CPUs, FPGAs can achieve deep learning capabilities and be applied to deep learning models.

Therefore, institutions believe that the increase in demand for high-end chips due to ChatGPT will drive up chip prices, leading to an explosive growth in demand for chips. Facing the exponential growth in computing and data transmission requirements, manufacturers of GPUs or CPU+FPGA chips will soon enjoy a blue ocean market.

Regarding the future price trend of FPGA products, Hu Guopeng, an analyst at Guohai Securities, believes that the triple demand driver of data centers, artificial intelligence, and autonomous driving, coupled with tight supply, will result in high prices for FPGA chips in 2023.

Currently, the localization level of FPGAs is relatively low. The Chinese market is mainly dominated by Xilinx (now acquired by AMD) and Intel, which occupy over 70% of the market share, while domestic manufacturers such as ANLogic Technology, Ziguang Micro, and Fudan Microelectronics account for about 15%. Haitong Securities believes that as the localization process of FPGAs advances, the Chinese market has vast room for improvement.