The success of DeepSeek again brings the debate over research openness versus research security to the forefront of how to best achieve and maintain technical leadership. (Express File Photo)
Jan 30, 2025 18:10 IST First published on: Jan 30, 2025 at 18:04 IST
Written by Norbert Holtkamp
Much has been written already about the severe impact the Chinese AI model release had on the confidence of the United States to maintain its position as the leading AI powerhouse. The US’s overwhelming access to capital, better chips and cheaper energy for future network and data centres provided what seemed to be an “unsinkable Titanic”, but as with the ship, the US’s stronghold turned into an illusion last Monday. The DeepSeek release and its demonstrated performance developed into a crisis at the stock market, which burned $600 billion USD for Nvidia alone (the world’s leading AI chip maker). Come Tuesday, the market began to recover, and for good reason.
Story continues below this ad
After all, the DeepSeek release with its stunning aftermath questions two things. First, it has long been the business model of leading US companies to be highly protective of their strategies and software. It is called into doubt whether this is more successful in the long term over a well-proven strategy of open source development to drive innovation. Second, there is a question over what other surprises can we expect from China as a country committed to driving AI technology forward, so much so that it educates nearly half of all AI-trained engineers and scientists with a bachelor’s degree worldwide.
Evaluating the first question, there is enormous opportunity with the future of AI development. DeepSeek was developed by Liang Wenfeng and a team of capable human beings, not by another AI system. The principles of Wenfeng’s approach were clearly accessible to any AI chatbot, yet it was their research team that remained foundational to creating DeepSeek. It required a well-versed, committed, and visionary leader with an equally competent workforce to implement these ideas. At least for the time being, it demonstrates to the world that a self-propelling AI is not as disruptive as well-educated people can be. It also points to the fact that even within AI, talent is the ultimate resource needed, more so than money, chips, and power. While AI development requires some access to computing, talent can make a difference anywhere in an interconnected world.
Open science and open source software developments have long been the driver of fast success in science, technology, and computing. The success of DeepSeek again brings the debate over research openness versus research security to the forefront of how to best achieve and maintain technical leadership. Given the unexpected and staggering success of the DeepSeek release, top AI development companies should consider revisiting the debate on whether their development environment is more constraining than need be. Still, having better access to capital, chips, energy, alongside open talent will immediately supercharge any model that is out there. Research groups around the world have worked on similar approaches. It will take little time to adapt novel elements of DeepSeek to existing infrastructure with superior hardware. Just how much better the next generation of models can be will soon be known and access to more resources will always provide a superior platform in the end. This has allowed for the quick and advancing recovery of the stock markets since the release.
Story continues below this ad
More time is needed to fully answer the second question. It is currently unclear if it was funded by just a few million dollars or developed by a comparatively small number of experts. In any case, DeepSeek’s funding is dwarfed by the AI investments made elsewhere in the world, in particular in the US, which sees large private and public spending. It is also unclear how their approach might reshape the AI ecosystem. Immediate threats remain: Given the integration of the private, public, and military sectors in China, it is safe to assume that DeepSeek’s underlying developments have already been used elsewhere. It is important to understand where and what capabilities it can supply beyond existing models. Outside of China, it is clear already that another “TikTok” moment can be expected very soon, given the growing worldwide use of the Chinese-developed chatbot. The fact that DeepSeek links back into Chinese hardware and already diverts uncomfortable questions about China’s history to meaningless answers is only the beginning. These early experiences provide a glimpse into how far-reaching into society AI and chatbots can quickly become. Still, the power of the new model and its simplicity attracts users, proven by the millions of downloads per day with little regard to those concerns.
most read
It seems apparent that more disruptive ideas and products can be expected given the sheer size of the AI talent that is available in China. The opportunity is obvious and it has been demonstrated right under our nose: Talent is key and talent is the driver, more so than anything else, at least for now.
Holtkamp is a Science Fellow at the Hoover Institution at Stanford University
