Heat Sink Solutions:1. Scaling for Data Centers and Chip Evolution

Computing Power Development and Chip Thermal Management

With the advancement of digital transformation, widespread adoption of IoT devices, expansion of cloud computing, and extensive applications of artificial intelligence (AI) and machine learning (ML) technologies, the total volume of newly generated data worldwide is growing rapidly annually alongside digital development.

The surge in data volume necessitates greater computing resources to process, store, and analyze this information, particularly placing higher demands on data processing technologies and real-time capabilities.

As a result, technologies such as cloud computing, big data, and AI require enhanced computing power to meet these challenges.

The global expansion of computing capabilities primarily stems from two key drivers: advancements in the performance of individual computing chips and the expansion in both the quantity and scale of data centers.

The development of AI computing power chips

Currently, NVIDIA’s GPUs hold more than 80% market share in the global AI computing power chip market. Over the past eight years, NVIDIA has improved GPU computing power by approximately 1,000 times through optimizing chip architecture, enhancing manufacturing processes, adopting HBM (High-Bandwidth Memory), and using advanced packaging technologies.

While computing power has been improving, chip power consumption has also been continuously increasing. According to statistics from the Asia-Pacific Core Valley Technology Research Institute, NVIDIA’s newly launched GB200 chip has a staggering power consumption of 2,700W, nearly 10 times that of its V100 chip.

For AMD, another major player in the AI computing power chip market, the power consumption of its GPU chips has also increased significantly during the iteration from 2020 to 2024, rising from 300W of MI100 to 750W of MI300.

Power Consumption Evolution of NVIDIA and AMD Chips

While the performance and power consumption of computing power chips have increased simultaneously, the evolution of AI technologies, innovative applications, growing data volumes, and computing power demands have collectively driven the growth of the AI computing power chip market. From 2015 to 2024, global data volume increased by 122 ZB, and the corresponding global GPU market size grew by $112 billion. GPU products have also evolved through five architectures, from Pascal to Blackwell.

Based on this trend, global data volume is expected to increase by 870 ZB from 2024 to 2030, and the global GPU market size is projected to grow by several times. Computing power chips will become the primary driver of growth in the global semiconductor market in the future.