standford university
The race between the U.S. and China has been intensifying, not just in AI model development but also in practical AI applications. While OpenAI’s success has helped it establish a commanding position in the market, China is making significant strides in AI deployment across various industries. Major Chinese tech firms such as Alibaba, Tencent, ByteDance, and Meituan have already begun developing large AI models and services, rapidly integrating generative AI into diverse sectors. As of June 2024, China had 2.3 billion users of generative AI products, comprising 16.4% of the country’s population.
In response to China’s growing AI presence, the U.S. has stepped up its efforts to maintain its global AI leadership. The U.S. Congress, through the U.S.-China Economic and Security Review Commission (USCC), has called for the establishment of an “AGI Manhattan Project” — a large-scale initiative aimed at accelerating the development of artificial general intelligence (AGI). This plan seeks to consolidate efforts from government, private companies, and research institutions to secure U.S. dominance in AGI technology, which is seen as essential for shaping the future of industries like healthcare, finance, and national security.
The concept of an AI-focused “Manhattan Project” is not new to Altman, who, even before founding OpenAI, had suggested a similar initiative to kickstart the AI revolution. Now, as OpenAI edges closer to realizing AGI, this vision seems closer to fruition. OpenAI’s recent release of its AI Infrastructure Blueprint outlines plans to establish AI economic zones and invest in renewable energy sources such as solar and wind to meet the demands of AI infrastructure.
Altman has also proposed forming a “World-Class AI Alliance” with U.S. allies like the Britain, Germany, Japan, and South Korea to create a unified strategic approach to AI development, aiming to outperform rival nations with different values and ideologies. This global coalition is expected to not only foster collaboration but also ensure that the U.S. retains a dominant position in the AI field.
Despite the remarkable progress in AI, significant challenges remain. As AI models continue to scale up, some experts argue that the industry may be approaching a plateau. The so-called "Scaling Law," which says that increasing the size of a model leads to exponential improvements in performance, appears to be slowing down. AI researchers are concerned that the gains from scaling up models like GPT-4 and beyond may become less pronounced, as we approach the limits of computational power and available data.
Harry Shum, the founding chairman of IDEA Research Institute and foreign member of the U.S. National Academy of Engineering, told AsianFin that the slowdown in the Scaling Law is primarily related to data constraints.
Tian Yuandong, a researcher at meta AI's FAIR lab, echoed this view, suggesting that the root cause lies in the lack of significant advancements in data and algorithms.
Zhipu AI CEO Zhang Peng shared a nuanced perspective, saying that while Scaling Law has encountered challenges in a narrow sense—evidenced by the plateau in language capabilities—there remains room for exploration in multimodal capabilities and AI agents.
"The slowdown in Scaling Law is merely a phenomenon and the observed result," Zhang noted. "In essence, computational power is likely the critical factor—useful information drives progress. Significant breakthroughs are still being pursued by numerous researchers, and each new advance brings new opportunities. Concerns about hitting a ceiling or a wall are a bit premature."
According to KKR & Co., global data center investments are projected to reach $250 billion annually due to increasing AI computational demands.
Reports from Sequoia Capital and Bain & Company estimate that the global AI market will soar to nearly $1 trillion by 2027, with annual growth rates for AI hardware and services reaching 40%-55%. Of this, over US$ 600 billion will be invested in AI infrastructure, with generative AI computing scaling to $10.99 billion, and model training costs increasing by over 240% annually.
OpenAI has set ambitious targets for the next phase, aiming for one billion annual users within a year. However, with annual expenses exceeding $5 billion, the company remains far from achieving financial balance.
Giancarlo Lionetti, OpenAI's Chief Business Officer, stated that the company aims to generate $4 billion in revenue this year, primarily through individual subions to ChatGPT Plus. OpenAI’s ultimate goal is to reach $100 billion in revenue by 2029.
Cheetah Mobile Chairman and CEO, Fu Sheng, told AsianFin that 2025 will likely be a year of AI application prosperity. "OpenAI has enhanced product and application capabilities, and GPT-4 has introduced smoother new dialogue modes that are also applications in their own right. This continuous evolution of models shows that Scaling Law is indeed slowing. Hence, I believe AI applications will flourish significantly next year," Fu noted.
As the U.S.-China AI rivalry intensifies, both countries are pushing the boundaries of innovation in the hope of leading the world in this transformative technology. While the future of AI remains uncertain, one thing is clear: the race is on, and the stakes have never been higher.
