2025 Analysis
| Assessment: Contested | Confidence Interval: Moderate |
| Direction: Contested | Confidence Interval: Moderate |
________
U.S. and China Jockeying for Leadership in AI
The United States continues to have a robust AI ecosystem led by an established private sector, where a handful of large companies such as Anthropic, Google, OpenAI, Meta, and Microsoft have produced most of the world’s foundation models in the last five years.[1] In 2023, private AI investment in the United States was almost nine times greater than the amount invested in China, the country with the second-highest investment, and 897 new U.S. AI companies were created.[2] The United States remains a top destination for top-tier AI talent,[3] but as big U.S. tech companies begin to turn more opaque in terms of research and development,[4] the number of AI patents[5] and publications produced by the United States are starting to slip.[6]
Source: Artificial Intelligence Index Report 2024, Stanford University (2024).
However, China has started to expand its AI ecosystem and has the most developers of large-language models worldwide.[7] The PRC’s AI advancements are largely driven by academia and a handful of national champions like Alibaba and Baidu.[8] PRC universities and labs have become leading AI research centers,[9] with Tsinghua University becoming a key hub for AI startups like Moonshot AI and Zhiphu AI.[10] Models produced by these startups and giants like Alibaba are competing against second-tier U.S. frontier models[11] with competitive edges like fluency in non-English languages[12] and various specializations, such as AI-generated text-to-video or vision capabilities. For example, PRC startup DeepSeek’s V3 model is reported to outperform OpenAI’s GPT-4o and Meta’s recent version of Llama.[13] This increase in model capability is translating into real-world AI applications, like AI-enabled industrial robots.[14]
Source: Notable AI Models, Epoch AI (2024).
Wildcards
- How Can China Utilize Open-Source AI to Overcome Restrictions on U.S.-Controlled Hardware? Beijing is squeezing whatever efficiency gains it can acquire from open source as the U.S. and other countries tighten its access to AI hardware.[15] Open-source AI models can be accessed and used by anyone without restrictions, providing an alternative to U.S.-controlled technologies. In spite of the country’s strict censorship requirements, China’s leading AI firms have built high-performing open-source AI models that are on par with closed-source U.S. AI models.[16] Qwen, a family of AI LLMs developed by PRC Internet giant Alibaba, is among the most downloaded and most popular models on online repository Hugging Face.[17]
- Will the United States Overcome the Energy Bottleneck in its AI Data Center Buildout? AI growth, particularly the buildout of data centers and AI training facilities that require immense computational power, is placing unprecedented demands on U.S. electric infrastructure and could consume as much as 8% of U.S. electricity by 2030.[18] While major tech companies and utilities are racing to quickly expand electricity supply, these efforts may not be enough to address systemic problems within the U.S. power system that could constrain further AI development in the United States, like the permitting and regulation of next-generation energy sources.[19]
What to Watch
- New Model Training Paradigms. Distributed training runs, like those pioneered by startups such as the U.S.-based Prime Intellect,[20] create decentralized GPU networks that democratize computational power across multiple organizations, which could potentially solve the current resourcing and access challenges to model development.
- A More General Form of AI Is on the Horizon. The path to artificial general intelligence (AGI)[21] is accelerating through continuous improvements in large language models,[22] development of more complex AI capabilities such as reasoning,[23] and transformative advances across the AI stack, like quantum computing.[24] AGI could arrive as soon as 2025-2027,[25] potentially disrupting entire economic sectors and representing a critical technological inflection point.
[1] Artificial Intelligence Index Report 2024, Stanford University (2024).
[2] Artificial Intelligence Index Report 2024, Stanford University (2024).
[3] The Global AI Talent Tracker 2.0, MacroPolo (2023).
[4] Nathan Benaich, State Of AI Report 2024, Stateof.ai (2024).
[5] Country Activity Tracker (CAT): Artificial Intelligence, Emerging Technology Observatory (2024).
[6] Eliot Chen, Chinese AI Companies Are Catching Up Despite U.S. Restrictions, The Wire China (2024); China May Soon Be the Top AI Innovator in the World, New Report Finds, Information Technology & Innovation Foundation (2024).
[7] Paul Triolo & Kendra Schaefer, China’s Generative AI Ecosystem in 2024: Rising Investment and Expectations, The National Bureau of Asian Research (2024).
[8] Eleanor Olcott, Chinese AI Groups Get Creative to Drive Down Cost of Models, Financial Times (2024).
[9] ChinAI, MacroPolo (last accessed 2024).
[10] China May Soon Be the Top AI Innovator in the World, New Report Finds, Information Technology & Innovation Foundation (2024).
[11] Nathan Benaich, State Of AI Report 2024, Stateof.ai (2024).
[12] Sam Eifling, China’s Biggest AI Model is Challenging American Dominance, Rest of World (2024).
[13] Kyle Wiggers, DeepSeek’s New AI Model Appears to Be One of the Best ‘Open’ Challengers Yet, TechCrunch (2024).
[14] Artificial Intelligence Index Report 2024, Stanford University (2024).
[15] Meaghan Tobin, China is Closing the A.I. Gap with the United States, New York Times (2024).
[16] Liza Lin, China Puts Power of State Behind AI—and Risks Strangling It, Wall Street Journal (2024); Wendy Chang, Large Language Model Development in China Thrives, But Geopolitics May Spell Trouble, MERICS (2024); Paul Triolo & Kendra Schaefer, China’s Generative AI Ecosystem in 2024: Rising Investment and Expectations, The National Bureau of Asian Research (2024).
[17] Arjun Kharpal, China Wants to Dominate in AI — and Some of Its Models Are Already Beating Their U.S. Rivals, CNBC (2024).
[18] AI is Poised to Drive 160% Increase in Data Center Power Demand, Goldman Sachs (2024).
[19] Fortifying American Energy Dominance in the Age of AI, Special Competitive Studies Project (2024).
[20] Jack Clark, 10B Distributed Training Run; China VS the Chip Embargo; And Moral Hazards of AI Development, Import AI (2024); INTELLECT-1 Release: The First Globally Trained 10B Parameter Model, Prime Intellect (last accessed 2024).
[21] AGI Will Arrive In Three Ways, Special Competitive Studies Project (2024).
[22] Robots Learn, Chatbots Visualize: How 2024 Will Be AI’s Leap Forward, New York Times (2024).
[23] Learning to Reason with LLMs, OpenAI (2024).
[24] Quantum AI: Harnessing the Power of Quantum Computing for AI, Special Competitive Studies Project (2024).
[25] In Good Company: Dario Amodei – CEO of Anthropic, Norges Bank Investment Management (2024).