The sell-off in data centre Singapore REITs (S-REITs), triggered by concerns over artificial intelligence (AI) development costs and US export controls, is “overdone and unjustified”, says Citi Research analyst Brandon Lee.
This is given that the pace of AI adoption is likely to accelerate in future, especially if cheaper models and, or lower costs do work out eventually, he adds, referring to Chinese AI start-up DeepSeek’s release of its open-source model, R1.
Year-to-date (ytd), units in Keppel DC REIT are down by 2.71% based on its $2.15 closing price on Tuesday (Jan 28). Similarly, units in Mapletree Industrial Trust (MINT) are down by 4.02% to close at $2.15 on Jan 28. Units in Digital Core REIT are down by 5.17% ytd based on its 55 US cent (75 cents) close on Jan 28.
The declines were largely attributed to the latest export quotas on AI graphics processing units (GPUs) announced by the US Department of Commerce’s Bureau of Industry and Security (BIS) on Jan 13, as well as DeepSeek’s announcement on Jan 20.
“Given that R1 was built at a significantly lower cost (supposedly not using the more/most advanced GPUs) than its competitors’ AI models, we believe this has led to investors’ concerns over hyperscalers’ ongoing pace of sizeable spending on AI and consequently, demand trajectory/outlook for data centre space/capacity,” Lee writes in his Jan 28 report. According to tech publications, DeepSeek’s model is estimated to be 90% to 95% cheaper than OpenAI’s frontier reasoning LLM.
Despite these concerns, Citi analyst Atif Malik, who covers Nvidia, says he does not expect leading AI companies to move away from more advanced GPUs. Fellow analyst Ronald Josey, who covers Meta Platforms, believes DeepSeek’s open-source model should enable other platforms like Llama, Gemini and OpenAI to “learn from its efficiency gains”.
See also: RHB trims CDLHT target price yet again after FY2024 results miss estimates
To Lee, there is a potential upside for data centres in terms of demand for space and capacity should the cheaper AI models gain traction. He adds that less costly GPUs could also lead to lower development costs and less energy-consuming AI data centres.
“It is important to note that the rising demand for data centres is also heavily driven by cloud computing, with AI being an additive and only coming into the conversation the past two years,” Lee writes.
“More specifically to Singapore’s data centre sector, AI inference has been the key driver of data centre demand, with Keppel DC REIT’s assets here all on AI-inference and hence from GPU compute perspective, it won’t rely too much on AI GPUs,” he adds.
See also: DBS names Seatrium and other regional names as top picks in oil & gas, energy sector
That said, the analyst acknowledges potential risks, particularly in the capital values or rental rates for data centres in different geographies due to the uncertain regulatory environment as well as swift developments in the sector.
‘Good buying opportunity’ for data centre S-REITs
To this end, Lee believes investors should leverage this “good buying opportunity” for all three REITs, Keppel DC REIT, MINT and Digital Core REIT, due to their unit price weaknesses.
The analyst has maintained his “buy” calls for Keppel DC REIT, MINT and Digital Core REIT with unchanged target prices of $2.50, $2.65 and 76 US cents respectively.
Malik has also kept his “buy” call on Nvidia in his Jan 26 note, with a target price of US$175 ($237.20). Josey has maintained his “buy” call on Meta Platforms with a target price of US$753.
As at 1.54pm, units in Keppel DC REIT, MINT and Digital Core REIT are trading at $2.18, $2.11 and 55.5 US cents respectively.
Shares in Nvidia and Meta closed at US$124.91 and US$687 as at Jan 30.
