DeepSek Accelerates Launch Of New AI Model Amid Intensive Chinese Investments

DeepSek, preparing to launch the latest generation artificial intelligence (AI) model, R2 (photo: x @NasdaqTurk)

JAKARTA Chinese startup DeepSeek is preparing to launch the latest generation artificial intelligence (AI) model, R2, ahead of the planned schedule. According to several sources familiar with the company's operations, the launch, which was originally scheduled for May, has now accelerated, although with no specific details regarding its release date.

The move comes after the previous model, DeepSek R1, rocked the global market with performance that outperformed many Western AI models, despite being built using its lower computing power Nvidia chips. The launch of R1 even triggered a decline in market value of more than $1 trillion in various technology sectors.

DeepSek is in the spotlight because of its different approach than other Chinese tech giants. Founded by Liang Wenfeng, a former manager of a well-known closed quantitative fund, DeepSek is more research-oriented than commercial companies in general.

Liang, who was born in Guangdong in 1985 and graduated from communication engineering from Zhejiang University, runs DeepSeek with a flatter management structure. The company's working culture differs from China's technology industry, which is known for its hierarchical system and work culture 996 (work from 9 am to 9 pm, six days a week). In DeepSeek, employees, including new graduates, are given full control over their project with more flexible working hours.

"Liang gives us full trust and treats us as experts in their respective fields," said Benjamin Liu, a 26-year-old researcher who has worked at DeepSee. I have complete freedom in developing an important part of their AI system.

Instead of focusing on AI-based application development as Baidu and Tencent do, DeepSeek prefers to improve the quality of their AI model. This strategy proved successful, as their R1 model could compete with world-class AI models at a much lower cost.

DeepSek can develop thanks to a massive investment from High-Flyer, Liang's quantitative fund company. Since 2020, High-Flyer has allocated about 70% of its revenue for AI research and built two supercomputer clusters. One of them, Fire-Flyer II, has 10,000 Nvidia A100 chips and plays an important role in the development of their latest AI model.

The decision to invest heavily in AI computing had attracted the attention of Chinese regulators, especially due to the accumulation of large amounts of chips. However, authorities ultimately did not intervene, a decision that was shown crucial when the United States imposed a ban on exporting A100 chips to China in 2022.

Now, the Chinese government fully supports DeepSek and has even instructed companies to remain humble in interacting with the media to avoid excessive attention from Western countries.

One of the main factors of DeepSek's success is their approach in designing more efficient AI architectures. Their models use Mixture-of-Experts (MoE) and Multihead Latent Attention (MLA) techniques, which allow data processing with lower computational power consumption. This approach makes the price of DeepSek 20 to 40 times cheaper than OpenAI, according to analyst estimates in Bernstein.

However, DeepSek's success has also raised concerns in the West. Several countries, including South Korea and Italy, have removed DeepSek apps from their national app stores due to privacy concerns. Meanwhile, several AI entrepreneurs in the West, including Scale AI CEO Alexandr Wang, have accused DeepSek of possessing up to 50,000 high-end Nvidia chips banned from being exported to China, although there has been no concrete evidence of such claims.

DeepSek has now been adopted by various Chinese government and corporate entities, including 13 city governments and 10 state-owned energy companies. Tech giants such as Lenovo, Baidu, and Tencent have also integrated DeepSek models into their products.

The Chinese government sees DeepSek as proof that their country is capable of rivaling the dominance of global AI, especially amid the tensions of trade and technology embargoes imposed by the United States.

Meanwhile, DeepSek competitors in the West are starting to adjust their strategy. OpenAI recently lowered the price of their services, while Google Gemini introduced cheaper subscription packages. DeepSek also encourages innovation in the AI industry, with some companies starting to adopt their more cost-effective approach.

With plans to accelerate the launch of R2, the world is now waiting to see how DeepSekwill continue to shake up the global artificial intelligence industry.