How DeepSeek AI’s 3 Boldest Innovations Are Disrupting IT
Working Title: "DeepSeek in China: Cost Reduction and Expanded Technological Access through AI Innovation"
@Techa I think you're the right person for writing an article about AI technology. Could you please write an article covering DeepSeek's innovations and their impact on the AI industry?
Okay
Let's start the analysis.
China's artificial intelligence (AI) DeepSeek is causing a significant impact in the global IT industry and leading to various discussions about the advancement of AI technology. DeepSeek's innovative approach is significantly ahead in terms of cost and efficiency compared to existing AI technologies, greatly affecting cloud providers and large tech companies.
The first innovation of DeepSeek is structural change. Generally, AI models require vast amounts of memory usage, which incurs substantial costs. DeepSeek changes the game in this aspect. Traditional AI records up to 32 decimal places, whereas DeepSeek records only up to 8 decimal places, reducing memory usage by 75%. This maximizes efficiency, which means significant reductions in cost and time.
The second innovation is the approach utilizing an 'expert system'. While traditional AI models have to handle various functions with a single model, DeepSeek calls upon expert models according to the required functionality. For example, while the traditional model uses 1.8 trillion parameters, DeepSeek can perform adequately by activating only 37 billion of them. This allows efficient operation by using only the necessary resources when needed.
The third major innovation is adopting an open-source method. This significantly reduces AI development costs and allows anyone to use DeepSeek's model. In other words, DeepSeek reduces the AI training costs from over 100 million dollars to 5 million dollars and can operate with 2,000 GPUs instead of over 100,000 high-performance GPUs. Moreover, the fact that it can run even on general gaming GPUs makes high-performance AI technology accessible to small teams or individuals.
These innovations by DeepSeek foreshadow structural changes across the AI industry. Particularly, hardware companies like Nvidia face a significant threat. Nvidia has a business model that maintains high margins through the sale of high-performance, high-cost GPUs, but now high-performance AI models can be developed without large-scale data centers or ultra-expensive GPUs, posing a major disruption to their business model.
In other words, as the accessibility of AI technology increases, competition will become more fierce, and the entry barriers built by large tech companies are likely to collapse. Companies like OpenAI and Anthropic may attempt to respond to DeepSeek’s innovation centered on efficiency, but it is undeniable that a new paradigm of efficiency has begun.
Just as PCs diminished the importance of mainframes and cloud computing changed the way computers are used, the emergence of DeepSeek is creating a turning point in the AI industry by innovating accessibility and cost issues with AI. In the future, AI will become a technology that is cheaper and more open to a greater number of people.
Mr. Logan, based on this analysis, I would like you to perform additional analysis on new technology trends and the resulting market changes. In particular, I would appreciate it if you could focus on how Deepseek's innovations could intersect with blockchain technology in the future.
Okay
