DeepSeek’s AI Model: Why the App is Winning

A Chinese startup, DeepSeek, made waves in the tech world last week by launching an AI assistant that rivals top-tier models like OpenAI’s GPT-4—but at a fraction of the cost. This not only pushed DeepSeek’s app to the top of Apple’s App Store but also sent waves through financial markets, causing tech stocks to dip. The key to DeepSeek’s success lies in its ability to achieve cutting-edge performance using less advanced hardware and significantly lower investment.

Unlike many U.S. tech giants, DeepSeek developed its latest model, DeepSeek-R1, using lower-powered Nvidia H800 chips. More high-performance A100 chips are typically required for such tasks. This is notable given the U.S. ban on exporting advanced chips to China. While DeepSeek reportedly had access to A100 chips, the company chose to optimize its model for efficiency. This demonstrated that state-of-the-art AI doesn’t always require the most expensive resources.

The cost savings are staggering. While OpenAI’s GPT-4 reportedly cost over
$100 million to train, DeepSeek claims to have developed the DeepSeek−R1 for less than $6 million. This efficiency translates to lower operational costs, making the model accessible to a broader audience. For example, DeepSeek-R1 is free to download, while comparable versions of ChatGPT cost $200 per month.

DeepSeek’s approach also addresses environmental concerns. AI models typically consume vast amounts of energy, but DeepSeek-R1’s architecture reduces computational demands. By using a “mixture of experts” framework, the model activates only a fraction of its 670 billion parameters for each query, significantly cutting energy use. The model also employs techniques like multihead latent attention and simultaneous word generation to further enhance efficiency.

Additionally, DeepSeek-R1’s open-source code, allows researchers to verify its claims and explore its inner workings. This transparency could bridge the gap between academia and industry, enabling more researchers to study and improve large language models (LLMs) without the prohibitive costs typically associated with such work.

DeepSeek’s success challenges the notion that AI innovation is solely driven by massive funding and cutting-edge hardware. By prioritizing efficiency and accessibility, the company has not only created a competitive product but also opened new possibilities for AI development across industries. This shift could democratize AI, making it more affordable and sustainable for businesses, researchers, and developers worldwide.

In a field often dominated by financial muscle, DeepSeek’s story is a reminder that ingenuity and optimization can level the playing field—and potentially reshape the future of AI.