Data centre euphoria starts to ebb after DeepSeek

NEW YORK – The recent market turmoil sparked by DeepSeek’s chatbot has left some rethinking the credit frenzy around artificial intelligence (AI).

Corporate giants told money managers that

the Chinese start-up’s cheaper models

will only spur more demand for the technology and the sprawling infrastructure it requires.

While big investors such as Blackstone president Jon Gray say that “digital infrastructure remains essential”, behind the scenes, landlords and credit providers say that the situation is more nuanced, and some are starting to fret.

A senior executive at one of the major data centre landlords expects the firm’s borrowing costs to increase because lenders will seek to protect themselves against the risk of the properties becoming obsolete from a DeepSeek-style disruptor.

Another asset manager, which both acquires and provides credit for the real estate, said that its lending ambitions are now less certain after the market panic.

The danger, the person added, is that AI mimics the green investing boom, when a rapid influx of money into clean energy created an oversupply of assets.

Both executives asked not to be identified as they were not authorised to speak publicly.

“The market is still exploring the impact of the new AI models, but it could be a healthy correction,” said Mr Timo Buijs, senior director of project and infrastructure finance at ABN Amro Bank. “Even if it has an impact on demand for data centres for AI, there is still growth expected,” he added, noting it may be “just slightly lower growth”.

The explosion of interest in AI since ChatGPT’s 2022 debut has set off a global building spree for data centres that house the servers and chips powering the tech. Private equity, real estate firms and sovereign wealth investors have pledged enormous amounts for the facilities, making them central components of economic growth across the globe.

Apollo Global Management, for example, sees a more than US$2 trillion (S$2.73 trillion) opportunity in data centres and associated infrastructure over the next five years.

Despite DeepSeek’s emergence, Meta Platforms chief executive Mark Zuckerberg predicted a “really big year” in AI for his firm. Khazna, a data centre developer that is part of Emirati tech conglomerate G42, told Bloomberg that DeepSeek’s emergence “further underscores the growing demand” for the facilities.

Wider credit markets have also been comparatively sanguine about last week’s turmoil, suggesting that they see the issues as less of an existential risk than stock investors.

“A big difference between equity and credit valuations was highlighted, with many names losing significant market value (Broadcom and Nvidia the largest), yet bond-spread widening was limited to a narrow range of 10 bps (basis points) or less,” Mr Robert Schiffman, Bloomberg Intelligence senior credit analyst, wrote on Jan 28.

In the commercial mortgage-backed securities market, Barclays strategists said that the downside risk to securitised data centres is low compared with stocks.

“The biggest risk to our view is a paradigm shift around data centres, which can significantly change investor sentiment, akin to what happened with the office sector over the past few years,” they wrote.

If DeepSeek’s claims about its spending are accurate, the start-up showed that it could train an AI model at a fraction of the amount that Meta, OpenAI and other major developers spend. However, these models still need data centres and equipment to make functional chatbots that people use or AI tools that companies purchase.

During this stage, even developers like DeepSeek will need to spend heavily on chips and computing resources, explained Mr Narry Singh, an executive partner with the consultancy AlixPartners. “I find the idea that this is going to depress massively the demand for infrastructure tenuous at best,” he added. BLOOMBERG