Opportunities from AI earthquakes

Following the DeepSeek craze, major Chinese tech companies quickly raced to launch large language models (LLMs) to compete and gain market share.

Global technology is accelerating

On January 29, the first day of the Lunar New Year 2025, Alibaba Group launched a new version of its artificial intelligence (AI) model Qwen 2.5 and announced that this model is more powerful than DeepSeek. According to Alibaba, Qwen 2.5 "surpasses in almost every aspect" the most advanced AI models from OpenAI and Meta (USA) such as GPT-4o and Llama-3.1-405B.

Previously, DeepSeek launched AI assistants using the DeepSeek-V3 and DeepSeek-R1 models at a cost only a fraction of that of products from American companies.

Two "earthquakes" DeepSeek and Alibaba have created an AI upgrade race right in China. Notably, the low-cost AI chatbot ChatGLM was developed by a research team from Tsinghua University and Zhipu AI Company. ChatGLM is an open-source AI chatbot capable of generating text, translating, and answering questions.

Tech giant Baidu's Ernie Bot also launched an upgraded version, Ernie 4.0 Turbo, capable of performing various tasks, including text generation, translation, question answering, image creation, etc. Meanwhile, ByteDance - TikTok's parent company - updated its flagship AI model and claimed it surpassed OpenAI's o1 in the AIME test - a standard that measures AI's ability to process and respond to complex instructions.

DeepSeek also threatens the global market, forcing leading technology giants such as OpenAI, Meta, Google, Microsoft, etc. to adjust their investment strategies. American technology corporations previously invested in AI factories and data factories, but from now on, they will have to focus on improving algorithms to reduce costs and operate faster and more efficiently. Recently, OpenAI introduced the Deep Research feature, which helps ChatGPT synthesize and research information. The highlight of this feature is the ability to display detailed research processes, including citations and summaries of the methods used, making it easy for users to track and verify information.

Several other countries such as India, South Korea... are also participating in the development of low-cost AI chatbots. For example, in India, billionaire Mukesh Ambani's Reliance Jio Company launched the first AI chatbot that supports local languages called BharatGPT. This is a significant step forward in the AI field in India, helping technology reach the non-English speaking population more easily.

DeepSeek and Qwen 2.5 interface Photo: Future/Qwen/Shutterstock

DeepSeek and Qwen 2.5 interface Photo: Future/Qwen/Shutterstock

What opportunities for Vietnam?

Mr. Lu Vincent The Hung - founder and CEO of EduX Global Institute Joint Stock Company - said that in the past, the technology industry often announced the cost of building AI models up to billions of USD. However, the launch of R1 by technology start-up DeepSeek with a cost of only 6 million USD - if true - could be the driving force for countries and start-ups to develop their own AI with high performance and low cost. From there, businesses, especially small and medium enterprises, can access and apply AI without investing too much initial cost; promote digital transformation in the fields of production, trade, services, education ...

However, DeepSeek's newly launched R1 model has faced many controversies related to data security and copying AI models of other technology companies. Therefore, in order to create their own AI models at low cost, countries and businesses need to carefully understand the operating principles, algorithms, source code, etc., instead of copying or stealing. "A successful AI model is not only about processing capacity but also about cost. Vietnamese businesses should follow the direction of building solutions that can be expanded and customized according to the requirements of many different industries" - Mr. Hung suggested.

According to Mr. Phan Tan Quoc - Deputy Director of KPMG Vietnam Innovation Program, fierce competition in the AI market will be both a pressure and a motivation for Vietnamese enterprises to quickly create useful and prestigious models to serve domestic users and customers. Mr. Quoc recommends that the government have policies to encourage enterprises and individuals to participate in AI development and database building; at the same time, support investment in AI research centers.

"The trend of sharing open source code in the AI community can help developers take advantage of and improve the model. In the coming time, many startups and individuals will easily access and apply AI to their operations when many new AI models appear, meeting specific needs," said Mr. Quoc.

According to experts, when developing AI, businesses must clearly determine whether the model can solve a specific problem, bring real value to users, and have differences in features or outstanding highlights. DeepSeek's model is a lesson for global technology companies with self-declared costs at super low levels but performance is said to exceed ChatGPT. Next, it is necessary to clearly understand the demand for AI in the market to ensure that AI is competitive enough and can be profitable.

Is DeepSeek really cheap?

In its report, DeepSeek said the total cost of training the AI model was less than $6 million, mainly due to renting graphics processing units from NVIDIA. However, recently, semiconductor research and consulting firm SemiAnalysis calculated the cost of developing DeepSeek to be much higher than stated.

DeepSeek is estimated to operate a massive computing system of around 50,000 Hopper GPUs, with 10,000 H800 GPUs and 10,000 more powerful H100 GPUs, along with additional H20 GPUs. The total capital expenditure for the servers is estimated at around $1.6 billion and operating costs are around $944 million.

Source: https://nld.com.vn/co-hoi-tu-nhung-con-dia-chan-ai-196250204200528423.htm