SAN FRANCISCO: A small Chinese startup named DeepSeek has set the AI world buzzing with its latest innovation—a breakthrough artificial intelligence system, DeepSeek-V3, that rivals the capabilities of renowned chatbots developed by tech giants like OpenAI and Google. This feat, revealed on the day after Christmas, is not just a technological milestone; it also raises critical questions about the unintended consequences of trade restrictions imposed by the United States on Chinese AI companies.
What makes DeepSeek’s achievement even more remarkable is the fact that it has managed to build this powerful system with only a fraction of the highly specialized computer chips typically used by leading AI companies. These chips, such as those produced by Silicon Valley’s Nvidia, are at the heart of the intense competition between the U.S. and China in the AI space. As the U.S. government continues to tighten controls on advanced semiconductor exports to China, companies like DeepSeek have had to innovate in creative ways.
In a research paper detailing the development of DeepSeek-V3, the company’s engineers highlighted that they used just 2,000 Nvidia chips—far fewer than the 16,000 or more chips typically required by U.S. tech companies for training advanced AI models. DeepSeek’s team spent approximately US$6 million on computing power for their project, a stark contrast to the hundreds of millions of dollars invested by tech giants like Meta to build similar systems.
This development challenges the conventional wisdom that only the largest tech companies with deep pockets can afford to build state-of-the-art AI systems. “The number of companies who have US$6 million to spend is vastly greater than those with US$100 million or US$1 billion,” remarked Chris V. Nicholson, an investor at Page One Ventures, highlighting the significance of DeepSeek’s cost-effective approach.
The DeepSeek chatbot, powered by its cutting-edge AI, is capable of answering questions, solving complex logic problems, and even writing computer programs—performing at levels comparable to those of market leaders. According to benchmark tests, it meets the high standards set by the U.S. AI industry. This success comes despite the limitations imposed by the U.S. government’s restrictions on the export of advanced chips to China. These restrictions, designed to prevent China from gaining access to military-grade technologies, have pushed Chinese researchers to rely on freely available tools and open-source software, contributing to unexpected breakthroughs.
The success of DeepSeek underscores the global shift toward open-source AI development, which has gained considerable momentum in recent years. The Chinese startup has made its system open-source, enabling other businesses and researchers to build upon it. Open-source software has been at the core of the AI ecosystem, particularly with releases like Meta’s LLama system. However, DeepSeek’s successful use of open-source platforms demonstrates that even smaller, resource-constrained companies can produce competitive AI technologies.
Despite concerns from U.S. lawmakers about the potential risks of open-source AI—citing fears about misinformation and security—experts argue that restricting open-source technologies could give China a substantial edge. If U.S. firms stifle the open-source movement, China could potentially lead AI research and development globally, leveraging its growing open-source community.
Notably, DeepSeek operates without any consumer-facing products, allowing its engineers to focus solely on research without having to comply with China’s stringent AI regulations, which restrict certain information for public consumption. This has allowed the team to explore innovative uses of AI, including in fields like poetry generation and tackling the challenging Chinese college entrance examination.
More Read
Meanwhile, the broader AI industry continues to evolve. OpenAI recently released a new reasoning system, o3, which is poised to push the envelope of AI capabilities. But DeepSeek’s innovations, alongside open-source contributions from companies like Alibaba, suggest that the race for AI supremacy is far from over, and smaller players can still make a significant impact.
This trend is illustrated by other projects, such as Sky-T1, an AI system built with just US$450 in computing power. By leveraging open-source technologies, developers managed to create a system that challenges the performance of OpenAI’s o1 model on specific benchmarks, demonstrating that innovation does not always require enormous resources.
In conclusion, DeepSeek’s ability to compete with Silicon Valley’s titans despite limited resources marks a significant shift in the AI landscape. It challenges the conventional notion that only large corporations can afford to develop advanced technologies, while also highlighting the growing importance of open-source AI in shaping the future of global innovation.
Reuven Cohen, a technology consultant in Toronto, has been using DeepSeek-V3 since its release and believes it offers comparable capabilities to those of OpenAI and Google’s latest systems—at a fraction of the cost. “DeepSeek is a way for me to save money,” he said, underlining the growing appeal of affordable, high-performance AI systems in a rapidly evolving market.
