Will Deepseek S Efficient Ai Model Reduce Or Boost Data Centre Demand

---

 
CLAIMS by DeepSeek’s large language model costing just USD5.6mil to train, a fraction of the billion-dollar investments that US companies have made, may reduce the need for data centres (DCs) in the future. 
“However, we envisage DeepSeek’s advancement could, in contrast, catalyse the DC industry as a whole, if its claim is proven to be true,” said RHB in the recent Malaysia Sector Update Report.
If DeepSeek’s claims are true, artificial intelligence (AI) training gains are likely to produce a 30% increase in efficiency but does not translate into a 30% reduction of chips or DCs. 

Instead, it means that the AI model gets 30% more power. Certain companies have lamented that AI is not able to deliver targeted ROIs and, hence, a more efficient AI model could enable such aimed returns to be met.
The paradox states that when technological progress makes using a resource more efficient, overall consumption of that resource tends to increase. 
Assuming demand for AI is relatively elastic, falling prices due to efficiency improvements create higher AI adoption. 

“We understand that one factor that slowed AI adoption within big organisations so far has been how expensive the AI models are to run,” said RHB.
Hyperscalers such as Meta and Microsoft  mentioned that the emergence of DeepSeek has not changed their plans to invest heavily in AI hardware in DCs in 2025. 
Both MSFT and META defended their USD80 bil and USD65 bil AI-related investments for the current fiscal year, saying it is crucial to remain strategically competitive in AI over time.

With tech giants staying put on AI investments, we think Gamuda and Sunway Construction’s DC orderbook and tenderbook will not face substantial scale-downs, as most of their clients are multinational corporations from Tier-1 countries (the US, the UK, Netherlands) that are eligible for the universal validated end-user status. 
If AI adoption continues growing amid democratisation, it may also bode well for contractors aiming to build DCs, for example, Kerjaya Prospek, with more DC providers potentially coming to Malaysia and providing more jobs for contractors.
Malaysia remains a magnet for DC investments due to its cost competitiveness in land, labour, electricity and proximity to Singapore. 
As AI enabled services may increase in the long run amid better affordability and efficiency, demand for underlying infrastructure is expected to grow in tandem. 

Hence, DC investments may continue coming on the assumption that Jevons paradox materialises in light of potential AI democratisation.
“Notwithstanding the above, we acknowledge the risks stemming from the lingering uncertainty with President Donald Trump having a 120-day window to comment on the AI chip restrictions,” said RHB.
Any major adverse impact may warrant a change in valuation multiples for key DC contractors later on.
Rerating catalysts could be the potential of chipmakers and policymakers in the US to overturn or tone down the AI chip restrictions. A key risk includes a major slowdown in DC development in Malaysia. — Focus Malaysia

---