Prime Highlights:
DeepSeek’s R1 AI model gained significant attention, dethroning OpenAI’s ChatGPT as the most-downloaded app on the U.S. App Store.
DeepSeek revealed that the training costs for its R1 model totaled $5.6 million, focusing on Nvidia GPU rentals, though this figure excludes additional research and data preparation costs.
Key Background:
A recent report estimates that DeepSeek, the Chinese AI startup, has spent significantly on hardware, with total expenses likely surpassing $500 million over its history. The company’s emergence in the tech world has caused a stir, particularly with the release of its groundbreaking AI model, R1, which is already being compared to the advancements of Google and OpenAI.
DeepSeek’s AI model garnered significant attention this week after its AI Assistant reached the top spot on the U.S. App Store, overtaking OpenAI’s ChatGPT. The company claimed that the training costs for R1 amounted to approximately $5.6 million, based on Nvidia’s GPU rental rates. However, this estimate only includes official training and excludes prior research, architecture experiments, and data preparation costs.
The report noted that DeepSeek’s hardware spending is likely well above $500 million, considering the extensive investment in research and development, as well as the substantial computing resources required to generate the synthetic data used in training. The report also mentioned that while the Claude 3.5 Sonnet from Anthropic cost tens of millions to train, the difference in funding sources—Anthropic raised billions from Amazon and Google—highlights the immense financial requirements for AI companies.
DeepSeek, founded in 2023 by Liang Wenfeng, originally grew out of the AI research unit of the quantitative hedge fund High-Flyer. Despite limited funding, the company has managed to achieve remarkable advancements in large language models and artificial general intelligence (AGI). Experts in the field have praised the R1 model for its impressive reasoning capabilities, though there are concerns over the company’s use of OpenAI’s data to train its system. The rapid rise of DeepSeek, which is entirely funded by High-Flyer, has sparked discussions about the potential geopolitical implications of AI competition, particularly with the U.S. lagging behind China in the burgeoning AI sector.
