Nvidia 'preparing for a GPU-less AI era'

Nvidia is said to have started preparing for the AI ​​phase powered by ASICs instead of GPUs, facing competition from Broadcom and Marvell.

According to Commercial Times, Nvidia has established a new division specializing in ASIC (application-specific integrated circuits) and is actively recruiting talent. CEO Jensen Huang also confirmed last year that the company plans to hire 1,000 ASIC engineers in Taiwan. The source said that this is considered a "preparatory measure to deal with fierce competition from Broadcom and Marvell when the AI ​​industry stops using GPUs and switches to ASIC chips".

Nvidia H100 chip card. Photo: Serve The Home

The explosion of generative AI and large language models (LLMs) has dramatically increased the demand for graphics processing units (GPUs). Nvidia 's dedicated GPUs , such as the H100 and H200, are widely used to train AI models.

However, the market is starting to shift away from GPUs and towards ASICs in the near future. This trend is driven by the need for chips optimized for real-world AI applications. ASICs are considered to outperform GPUs in inference applications, such as AI training and cryptocurrency mining.

According to Verified Market Research , the AI ​​inference chip market is expected to grow in value from $15.8 billion in 2023 to $90.6 billion in 2030.

Major tech companies have already started using custom-designed ASICs, like Google, whose AI chip called “Trillium” is expected to be widely available by December 2024. The shift to custom AI chips is also intensifying competition among semiconductor companies. Broadcom and Marvell have become increasingly important players in the AI ​​craze, partnering with cloud service providers to develop specialized chips for data centers.