ON January 27, 2025, a seismic event in the tech world sent shockwaves through the stock market, with Nvidia—a US-based chipmaker powering the AI industry—suffering the largest one-day market value loss in US history, amounting to an astounding US$600 billion.
This sharp decline was sparked by the sudden rise of a little-known startup from Hangzhou, China: DeepSeek.
DeepSeek, a company focused on large language models (LLMs), made waves by unveiling a model that rivals the best AI systems in the world, yet at a fraction of the cost.
The company claimed to have completed the critical “pre-training” phase of its model in just two months, with a budget of less than US$6 million—far below the costs associated with comparable US models.
This raises a fundamental question: Is this the beginning of a new AI paradigm in which costs are significantly lower than previously estimated? If so, it could have far-reaching implications for industries ranging from data centres and microchips to energy grids.
Changing Landscape for Data Centres
Southeast Asia, a region that has seen rapid growth in AI adoption, may find its tech economy facing two key challenges: demand and supply.
On the demand side, the promise of cheaper AI could spur wider adoption, much like the region has benefited in the past from affordable Chinese products and technology.
However, the more pressing concern is supply—particularly the development of data centres.
Over the last few years, Southeast Asia has attracted billions of dollars in foreign investment, with major players like Nvidia, Google, and AWS establishing data centres across countries such as Malaysia, which is set to become the third-largest data centre hub in Asia by 2029.
Malaysia’s growth has been fuelled by its strategic advantages: ample land, low construction costs, a robust submarine cable network, and an abundance of water resources.
Malaysia’s ambitions have been aligned with its government’s focus on fostering a thriving digital economy, with Prime Minister Datuk Seri Anwar Ibrahim aiming to drive AI as a key growth engine.
In December, he announced the formation of the National AI Office, projecting that AI could contribute up to 25.5% of Malaysia’s GDP in the near future.
However, DeepSeek’s entry into the market forces a reconsideration of whether these investments will prove worthwhile. If AI can be trained at a significantly lower cost, the need for massive data centre infrastructure might not be as urgent as previously thought.
This scenario mirrors the disruption seen in the past, such as when cloud storage solutions made physical servers obsolete.
The Great Catch-Up
Despite the challenges posed by DeepSeek’s sudden success, there are several reasons to believe that AI’s infrastructure investments may not be on the verge of decline.
First, it’s important to consider that DeepSeek’s reported US$6 million cost was just for the pre-training phase of its model, and not the total R&D expenditure.
Analysts have speculated that DeepSeek’s access to tens of thousands of Nvidia chips, while not the highest-end models, could have pushed its overall infrastructure spending to well over US$500 million.
This means that the total cost of developing its AI model may not be as low as advertised.
Second, industry experts like Dario Amodei, CEO of AI company Anthropic, argue that lower training costs for AI models don’t necessarily lead to lower infrastructure spending.
In fact, as AI systems become more powerful and capable, companies tend to reinvest any cost savings into further development rather than cutting back on infrastructure.
The argument is that the demand for higher-performing AI will always outweigh the need for savings in building capacity.
Lastly, while DeepSeek’s cost efficiency could lower the barrier to entry for many companies looking to build or use AI models, it could also spur greater demand for AI applications.
If DeepSeek’s model—and its smaller variants—become embedded in everyday consumer devices, the demand for AI infrastructure may rise rather than fall.
In short, DeepSeek’s rise has sparked what some are calling the "great catch-up" in the global AI race. After challenging Alibaba in 2024 for the top spot in the AI field, the company has now closed the gap with OpenAI, making its disruptive impact just the beginning of a much larger shift.
As Southeast Asia and Malaysia in particular continue to bet on AI-driven infrastructure, the region’s investments will soon be put to the ultimate test—whether DeepSeek’s efficiency is a game-changer or just a temporary disruption in an ever-evolving tech landscape.
James Chai, a political analyst and columnist, is the author of Sang Kancil (Penguin Random House). - February 14, 2025
