DeepSeek’s $1.6 Billion AI Model Raises Eyebrows

DeepSeek’s AI model remains a subject of increasing debate due to its budget-friendly approach that outperforms exclusive models such as OpenAI’s o1 reasoning model in numerous areas like programming, science, and mathematics. This is all achieved for a significantly lower expense compared to the development cost of these proprietary models.

The model powered by DeepSeek’s R1 V3 was allegedly trained at an estimated cost of $6 million. This comes amid concerns that advancements in artificial intelligence have stagnated due to a scarcity of top-tier information for training purposes. However, fresh insights about DeepSeek’s AI model have emerged from a report by SemiAnalysis, as reported by Tom’s Hardware.

It appears that the Chinese AI company invested more resources than initially expected for training its R1 model. Not only did it go beyond the $6 million budget and 2,048 AI GPUs, but also splurged a staggering $1.6 billion on hardware, specifically 50,000 NVIDIA Hopper GPUs. Moreover, operating costs amounted to approximately $944 million for the project.

I witnessed a significant stir in the AI sector when DeepSeek’s innovative approach caught everyone’s attention. This development sparked investor apprehension, leading to a staggering potential loss of up to $600 billion for NVIDIA in market value within a day. Notable figures in the AI industry like Microsoft CEO Satya Nadella have expressed admiration for DeepSeek’s technology, emphasizing the need for all key players to closely monitor advancements, particularly those originating from China, with great caution.

According to Meta’s top AI scientist, Yann Lecun, there seems to be a significant confusion regarding the $1 billion investment in artificial intelligence (AI). The funds, as stated by this executive, are earmarked for inference, rather than training AI models.

According to Nadella, DeepSeek brings some genuine advancements. When the value of tokens decreases, the cost of inference computing also goes down, which results in more consumption and the creation of additional applications.

DeepSeek hires from within

Even though it needs substantial funds to progress in AI technology, this Chinese startup operates independently, avoiding external meddling and potential acquisitions. It prioritizes efficiency and refining algorithms rather than expanding models at a fast pace.

openai, a competitor in AI technology, faces the necessity of transitioning into a profit-driven company or risk returning funds to investors who invested during the October fundraising round, which propelled its market value surpassing $157 billion and prevented potential bankruptcy.

Nevertheless, this action has faced criticism from significant figures, including Elon Musk, a former co-founder of OpenAI, who took legal action against the company twice. He argued that it represented a severe deviation from its original purpose and possible engagement in unscrupulous activities such as racketeering. Musk alleged that he was deceived into investing in the AI company under false pretenses of a charitable mission.

To this end, DeepSeek has reportedly invested up to $500 million in its AI advances.

In other locations, the rapidly growing AI company is said to only hire staff who are graduates from prestigious Chinese universities like Peking University and Zhejiang University, prioritizing their problem-solving abilities and skills. Moreover, this company, DeepSeek, is known for generously rewarding its employees, with some of its AI researchers reportedly earning as much as $1.3 million annually.

Read More

2025-02-04 16:09