Technological Innovation and Application Implementation of Artificial Intelligence: AI Computing Power Demand Shows Exponential Growth
-
The demand for AI computing power is enormous. AI serves as a crucial productivity tool, empowering various industries through integration. In emerging sectors such as autonomous driving, smart homes, security monitoring, robotics, medical devices, and smart classrooms, technological innovation and application implementation of artificial intelligence are driving forces behind industry intelligence. Additionally, application scenarios like AI interaction and AI creation are developing rapidly. For instance, the emergence of natural language processing tools like ChatGPT is expected to further enhance the level of industry intelligence.
At the COMPUTEX conference, NVIDIA CEO Jensen Huang unveiled the computing power "game-changer"—the DGX GH200 supercomputer. This supercomputer is specifically designed for large-scale generative AI workloads, comprising 256 GH200 superchips. It boasts 1 exaflop of extraordinary AI performance, 144TB of memory (nearly 500 times that of NVIDIA's current DGX A100 system), 150 miles of fiber optics, and over 2,000 fans.
AI Computing Power Demand Shows Exponential Growth
Faster transmission rates and higher coverage are required for optical modules, which are the fastest and most direct indicators of the global AI industry's rapid development. Since March, North American manufacturers have placed four additional orders for 800G optical modules, with the total annual demand for 800G optical modules now exceeding 1.2 million units.
As one of the leaders in this wave of AI enthusiasm, NVIDIA's recent financial report has been instrumental. Public data shows that NVIDIA achieved revenue of $7.19 billion in the first quarter of fiscal year 2024, a 13% year-on-year decline but a 19% sequential increase. Net profit was $2.04 billion, up 26% year-on-year and 44% sequentially. Additionally, the company expects second-quarter revenue to reach $11 billion, with a gross margin of 68.6%, significantly higher than Wall Street's expectation of $7 billion in revenue.
NVIDIA CEO Jensen Huang previously stated, "The computer industry is undergoing two simultaneous shifts—accelerated computing and generative AI. As companies race to apply generative AI to every product, service, and business process, the trillion-dollar global data center infrastructure will transition from general-purpose computing to accelerated computing."
In fact, amid the AI frenzy sparked by ChatGPT, NVIDIA's GPUs for AI computing are in short supply, driving the company's market value to soar, approaching the trillion-dollar mark.
Building on this, J.P. Morgan updated its forecast in a recent report, predicting that NVIDIA will capture 60% of the AI product market in 2023, primarily from its graphics processing units (GPUs) and network interconnect products. Broadcom ranks second, with its application-specific integrated circuits (ASICs) expected to account for 13% of revenue. TSMC ranks 17th, with a 3% revenue share.
Since ChatGPT's groundbreaking debut last year and its rapid rise in monthly active users, AI has not only drawn global investor attention but also become a hot topic across social networks worldwide. Unlike the distant, high-tech perception of AI in the past, this wave of innovation has made people realize its potential to integrate into daily life. NVIDIA's stellar earnings report confirmed the accelerating trend of AI from the computing power source, while U.S. chip designer Marvell emphasized that AI-related revenue is expected to double annually over the next two years.
Over the past decade, AI computing power has primarily relied on GPUs from foreign chip manufacturers, widely used in cloud-based AI products. Meanwhile, edge-side embedded AI computing has evolved from CPUs and GPUs to DSPs and ASIC architectures, driving the widespread adoption of deep learning technologies like speech recognition, facial recognition, image-text recognition, AIGC, object detection, super-resolution, and ADAS.
Computing power is the productivity of the digital economy era. "In a few years, computing power will be like electricity today—you won’t need to care where it comes from; just connect to the computing network, pay, and use it."
China is building a national computing network, treating it like a power grid and operating it like the internet, making computing power as accessible as electricity. Currently, the China Computing Network has aggregated over 3 exaflops of collaborative computing power, achieving preliminary nationwide coordination and efficient scheduling, laying the strongest foundation for the digital economy.
Capitalizing on the "East Data, West Computing" initiative, major players like China's three telecom operators, Huawei, ZTE, Tencent, and GDS have established operations in Shaoguan, becoming key forces in this national project. Central state-owned enterprises have shown particular interest, with China Mobile, China Telecom, and China Union all setting up data centers in Shaoguan as part of the Greater Bay Area national hub.
Flexible computing resource allocation was a key focus at the conference. "Our network enables full connectivity for computing power, available on demand," said Wang Xiaoyun, Chief Scientist at China Mobile. The company has developed a "4+N+31+X" computing power layout, with Shaoguan as part of the "N" nodes to meet diverse business needs.
As a hub, Shaoguan serves both as a cluster for data centers and a radiating center for computing resources.
Tang Xiongyan, Chief Scientist at China Unicom, highlighted that the company has deployed a national-level computing center in Shaoguan, integrating general, super, and intelligent computing. Through the "Greater Bay Area Computing Power Scheduling Platform" in Shenzhen, it enables dual-center computing services, extending its reach nationwide. "We’re exploring deep integration of computing and networks. Future 6G networks will have built-in intelligence, where computing power will play a vital role," Tang said.
Facing the surging demand for computing power, synergistic effects have become increasingly prominent. Huang Hongbo, Deputy General Manager of China Telecom's Tianyi Cloud, explained that with the rapid development of the digital economy, the demand for computing power across society is expected to grow by 55% annually. Over 70% of this demand comes from first-tier cities, where energy consumption quotas are scarce. By transmitting data at high speeds over networks, the 'East Data West Computing' and 'East Data West Storage' initiatives can effectively alleviate the imbalance between computing power supply and demand by enabling cross-regional resource allocation.
The 'East Data West Computing' project has set higher requirements for promoting green and intensive development of computing power. Wu Jianbin, Chairman and Chief Scientist of Changxing Taihu Energy Valley Technology Co., Ltd., also mentioned that Shaoguan's abundant clean energy resources, such as hydropower, wind power, and photovoltaics, can effectively meet the demand for 'green electricity' in computing power. 'We aim to provide green energy support and services for Shaoguan's supercomputing centers, where there is enormous potential for collaboration,' Wu Jianbin said.
As telecom operators ramp up their investments in computing power infrastructure, communication equipment manufacturers see significant market opportunities. 'The 'East Data West Computing' project will build a high-quality, low-latency computing power network across the country, serving as the main artery for China's future computing power circulation. We will work with operators to ensure computing power and data permeate every corner and scenario of urban governance,' said Hu Xuemei, Vice President of ZTE.
Over the past year, the Shaoguan data center cluster has attracted 12 large-scale data center projects, with plans to build 361,000 racks and a total investment of 39.5 billion yuan. The accelerated construction of supporting infrastructure and the establishment of upstream and downstream industries have laid a solid foundation for the development of computing power infrastructure and industrial layout in Guangdong.