China’s DeepSeek AI Shakes Up Tech Industry with Groundbreaking Chatbot
The release of a new AI chatbot by a small Chinese company, DeepSeek, has sent shockwaves through the technology industry. This app quickly surpassed OpenAI's ChatGPT as the most-downloaded free iOS app in the US and triggered a historic market loss, causing Nvidia to shed nearly $600 billion (£483bn) in market value in just one day.
What Makes DeepSeek AI Different?
The key factor behind this disruption lies in DeepSeek’s cutting-edge large language model (LLM). Unlike its American counterparts, the DeepSeek AI model boasts comparable reasoning capabilities while reportedly requiring a fraction of the cost to train and operate.
DeepSeek’s base model, R1 V3, was trained in 2.788 million hours using thousands of GPUs, yet its estimated training cost remained under $6 million (£4.8m). In comparison, OpenAI’s GPT-4 reportedly cost over $100 million (£80m) to develop. By implementing technical strategies that minimize computational demands, DeepSeek has managed to achieve unprecedented efficiency.
The Impact on Nvidia and AI Development
Despite the dramatic drop in Nvidia’s stock, DeepSeek's models were still trained using approximately 2,000 Nvidia H800 GPUs. These chips were designed to comply with US export restrictions and were likely stockpiled before further sanctions were imposed in October 2023. Faced with these limitations, DeepSeek has had to maximize the efficiency of its available resources, leading to breakthroughs in cost-effective AI training.
Another notable advantage of DeepSeek’s approach is its potential to reduce the environmental impact of AI. Running massive data centers consumes enormous amounts of electricity and water to keep servers from overheating. While many AI companies do not disclose their carbon footprint, ChatGPT’s monthly emissions have been estimated at over 260 tonnes of CO₂—equivalent to 260 flights from London to New York. By cutting computational costs, DeepSeek could pave the way for more sustainable AI development.
A New Challenger in the AI Race
The speed at which DeepSeek emerged with a competitive AI model has taken many by surprise. Founded by Liang Wenfeng in 2023, the company has already made waves, with Wenfeng being hailed as an “AI hero” in China.
One standout feature of DeepSeek’s model is its transparency. Unlike OpenAI’s closed-source models, DeepSeek has openly released the model’s “weights” (numerical parameters obtained during training) and a technical paper outlining its development. This allows researchers worldwide to run the model on their own hardware and adapt it for various tasks.
However, some critical details remain undisclosed, including the datasets and code used in training. This has sparked curiosity among researchers, who are now working to piece together the missing information.
Also Read: Alibaba Unveils New AI Model ‘Qwen 2.5,’ Claims It Beats DeepSeek-V
The Future of AI and Sustainability
DeepSeek’s innovations have raised crucial questions about AI’s future—not just in terms of cost and accessibility, but also sustainability. Will the efficiency of these models truly reduce overall energy consumption, or will lower costs lead to an explosion in AI usage that counteracts any environmental benefits?
As the world prepares for the upcoming Paris AI Action Summit, DeepSeek’s breakthrough may help push the conversation around sustainable AI to the forefront. One thing is clear: the AI race is evolving rapidly, and DeepSeek is proving to be a formidable new player.

Comments
Post a Comment