By: Isabella Mazzi
Since OpenAI ignited the artificial intelligence boom in late 2022—which saw a surge in generative AI like DALL-E and ChatGPT—the prevailing notion had been that AI systems could not be built without billions of dollars worth of AI chips, computer chips used to optimize the inference and training of AI algorithms. This meant that the mega companies with the chips, like Microsoft, Google and Meta, held all the power, keeping the United States at the forefront of the AI race. The recent release of DeepSeek, a highly efficient Chinese Large Language Learning Model (LLM), has unraveled US control over the AI world and placed China on a pedestal in true Sputnik-like fashion. DeepSeek shook the world not only because it was able to match the abilities of the best chatbots from US companies like OpenAI and Google, but because the LLM could do so for a fraction of the cost. DeepSeek claims its model costs about $6 million, far less than the $80 to $100 million required by GPT-4 or LLaMA3.
The startup’s engineers demonstrated a more efficient way of analyzing data using the chips, significantly reducing the need for Nvidia chips, which DeepSeek uses. This made DeepSeek AI a more cost effective model compared to other LLMs. DeepSeek uses only 2,000 Nvidia Chips, a fraction of the standard 16,000+ chips required by any other chatbot. This lowers costs, due to improved memory efficiency and quickened processing time. Driven on the promise of cheaper AI training, markets have significantly devalued chip makers like Nvidia. However, the industry does have doubts on its reported costs, as the numbers have left out key information on research and development, which would have included the cost to design, explore and develop the base model, the purchase costs of the hardware and GPUs, and the AI researchers.
The impact of DeepSeek having Chinese origins is one not to overlook. Because Nividia is an American company, The Biden administration had the power to put rules in place limiting the number of powerful chips being sold to China. The US first banned the sale in 2022, telling the company it could no longer ship its A100 and H100 processors, prompting Nvidia to develop two less powerful chips for the Chinese market, the A800 and H100. This was in an effort to keep them out of the hands of China’s military and to maintain the US lead in the global AI race. The presence of these trade restrictions forced DeepSeek to adapt and optimize its model. Additionally, DeepSeek is subject to the authority of the ruling Chinese Communist Party (CCP) over what information and images can and cannot be shown. The CCP likely also has access and control over global user data input into the LLM. Given that AI firms set various rules to trigger set responses when words or topics the platform doesn’t want to arise, DeepSeek may have significant implications on free speech and global opinion due to its Chinese ties. The data that users input into DeepSeek may also be accessed by the CCP, leading to significant security and privacy concerns.
Compared to other AI models, DeepSeek AI uses an open source system, similar to software like R. This means it is one of the most unrestricted large-scale AI models yet. Anyone, from independent researchers to private companies, can fine-tune and deploy the model because the code is publicly available for modification. Without any monitoring or restriction, concerns are raised regarding mass disinformation, cyberattacks, or social manipulation. Despite this, its impact on society is undoubtedly more significant. By making a powerful, low cost AI open-source, Deepseek democratizes AI innovation by giving researchers, startups and organizations access without being part of a major tech firm or government-backed research lab. The open-source aspect of the LLM encourages usage and advancement of the model, and facilitates further global AI innovation. DeepSeek’s open source decision breaks the monopoly of large AI firms in innovation and offers a powerful alternative to paywalled AI models. This could lead to faster breakthroughs in fields like science, healthcare and business. If this model scales, it will redefine how AI is developed globally.
Similarly, DeepSeek’s low cost nature is groundbreaking. It demonstrates that an upstart team can enter cheaply and develop a competitive AI model. This newfound accessibility will likely attract other entrants, translating to an upcoming wave of involvement and innovation in the AI sector. Its low cost also makes advanced AI tools more accessible to a larger audience. This democratization allows for the leveraging of AI by individuals and businesses that were previously cost-prohibited. This could result in innovation, and advancements in different fields as mentioned previously, like science, healthcare and business. In order to maintain US leadership in AI, policymakers must promote competition by funding of startups, and antitrust enforcement.
While DeepSeek’s impact on the market will likely stabilize, its influence on the AI industry is poised to reshape the landscape for years to come. Not only will the LLM foster innovation through its low-cost and open source, but it will also democratize AI access. Although DeepSeek raises concerns regarding its geopolitical influence, the concept challenges the assumption that AI dominance is exclusive to companies with near-unlimited resources and reinforces the necessity for competition in the AI world.





Leave a Reply