OpenAI plans to release ChatGPT 5.0 by the end of this year and is advancing the development of customized ASIC chips. The company is in talks with chip design giants like Broadcom (AVGO) and Marvell (MRVL).

As AI algorithms mature and the demand for tensor computations increases, the value of customized chips is gradually becoming apparent. According to New Fortune Industry Research Institute, the focus of global AI development in the second half will be on customized ASIC chips, which brings uncertainty to NVIDIA's future.

With the continuous increase in computing power demand, NVIDIA's GPU chips are in high demand, and TSMC's (TSMC) wafer production has correspondingly increased. This year, the demand for computing power in large models has continued to rise, with NVIDIA's high-performance chips holding an important position in the market. However, the rise of customized chips may change this situation.

In the past, customized ASIC chips replaced general-purpose GPU chips as the main force in Bitcoin mining, and this history may repeat itself today. Broadcom and Marvell hold significant shares in the customized chip market, and the market size is expected to grow rapidly.

NVIDIA is actively expanding into the customized chip market, establishing a new department focused on designing custom computing power chips. Their partners include major cloud computing companies such as Google, Microsoft, Amazon, and Meta. NVIDIA also leverages its software and technical advantages, such as the CUDA ecosystem and NVLink technology, to maintain a leading position in the AI computing power training stage.

The development of customized ASIC chips is predicted to focus on three main directions: expanding chip area, improving memory bridging solutions, and enhancing compiler design. Customized chips will demonstrate high performance and cost-effectiveness in specific application scenarios, driving large-scale market demand.