The appearance of cheaper and more effective AI models from the back of Chinese deep can reshape demand for data centers, enhancing the sector for which investors are already strong betting and further boom.
For years, analysts have predicting exponential growth in data centers – the critical infrastructure needed to power the world digital transition and training of large language models (LLMS).
Chinese startup Deepseek AI model sent investors in Jitters At the end of January, as a launch of its R1 model asked questions about the US dominance in the AI sector and whether it was a developer Gain efficiency Can it alleviate demand for data center capacity.
Data centers often need at least two years to build, and the orders have been largely introduced for 2025 years – which means that the launch of the devastating R1 model is probably a direct impact. While the launch of Deepseek R1 initially led to some analysts alleviated their forecasts while questioning whether the money was pumped into the sector could be somewhat “wrong,” said CNBC experts to build cheaper is less powerful chips in the end of the end an accelerant for the market.
Bull look
Deepseek emphasizes that data centers are vulnerable to shifts in narratives around AI consumption, according to analysts at Barclays. If the claims of the efficiency made by Chinese startups are confirmed, the development shows that “hundreds of billions dedicated to the development of AI, therefore, make it wrong, and the plans of capital consumption of hyperscalers could be reconsider,” said analysts conducted by Brendan Lynch in Note published on January 27.
They added that if AI requires less infrastructure, it would “” plant with lowest quality ” – which are the least energy efficient – which could face weaker demand and weaker prices.
In the meantime, UBS analysts have noticed that about a third of the market growth of the current data center of data relies on the construction and development of generative artificial intelligence-Ai that can create paintings from written inquiries. These forecasts are not a factor of radical improvement of efficiency, UBS said in note 28 January.
In April last year, UBS initially predicted that Global Data Center’s equipment market increased 10-15% over a three-year period until 2028. This week, the bank analysts said that new data and calls of experts ultimately lead to the prospect of Bullish on the market. The company is now expecting that the sector revenues will increase 20% in 2025 and see that the “scope will be according to a higher end” growth range of 10-15% for at least the beginning of the 2026-2028 period, analysts said Wednesday.
The jury is “still” about whether Deepsek needed 20 to 30 times less computer for asking for conclusion, Andre Kukhnin, analyst to explore the UBS stake, he told CNBC – referring to the data launching process through AI models in order to predict for prediction or resolved a task.
“Although it is more effective in token, it needs more tokens upon asking, because it is a model of explanation, not” the flow of speech “… The essence is that we do not think that it significantly reduces the demand for strength to conclude,” Kukhnin explained.
The Goldman Sachs Research Department envisages that the balancing of the offer and demand of “demand” in the coming years will be “tightened”, peaking at the end of 2026, and then moderating from 2027 onwards.
If efficiency increases lower capital expenditures (Capex) than the main investors, it could: “Mitigate the risk of long-term too much offering that we see 2027 and wider I think it is important to consider more durability and less cyclical in the data center market , “James Schneider, senior analysts for capital research at Goldman Sachs, noted in a report on February 4th.
Much has yet to be determined about the impact of initial technology, less than three weeks since Deepseek has published his information. R1 is not enough to “move the needle” when it comes to demand, according to Andrew McMillan, a partner at the RPC law firm.
“The appetite of the investor will be alleviated if this can be shown to be repeated, so in the future there will be a much lower demand for data processing than now, or at least it will not continue the same way,” said McMillan, who specializes in the management of M&A and data.
“I think in the long run it will be interesting to see if this structural approach is capable of retaining and I think it can affect the market shape.”
‘Fuel to fire’
Supplies vulnerable to shifts in The data center of the data center fell on January 27th. Schneider electricThe most commonly exposed European company in data centers according to UBS, has lost more than 9%, Siemens Energy stocks spilled 20% and Abb closed 6% lower a day.
Some shares have returned their losses since, recovering from the knee reaction from the market. Earnings of Mega-Hiperscales such as alphabet Google and Target also instilled in confidence, such as both companies dedicated Investments in more billions of dollars After a technical sale.
There was not “much room for mistakes in the sector,” Kukhnin said. “That is why some shares have collapsed and are not purchased immediately, because people already have a lot of shares and now they are trying to figure it out if it is an opportunity to add or it is the opposite.”
Added that lower costs indicate the potential Democratization of AIwhich could lead to acceleration of technology adoption – which is “something that is very difficult to quantify.”
The data center market will continue to encourage a digital transition, which takes place separately from the progress to the AI. “The generative Ai was somehow almost glaze on the cake, but it became a very thick layer of glaze, certainly in terms of future growth,” Kukhnin said.
Bruce Owen, President of Emea from Equinix, said the company “is well positioned because the curve of AI technology is bent”, adding that he expects the appearance of more effective models to be “accelerated” for AI.
“An additional dynamics we could see is” Jevons Paradox “, which believes that increased resource efficiency can lead to greater consumption of this resource,” he told CNBC.
Ryan Cox, head of AI -ja, at the Synechron consulting company, is also expecting that Jevon’s paradox effect is to see more efficient technology to ultimately lead to a greater demand of data center.
“It’s a really complex equation,” he told CNBC, noting that there are several winds and tail winds when it comes to determining potential demands in demand. He shared that Synechron’s clients follow “safe” opportunities for indirectly using Deepseek, such as Face hugs, AI models storage.
“Generally, I think efficiency will encourage adoption, and I think it will continue to encourage use, even if these costs fall. Race according to these more advanced models and wider applications, the use of AI, means that this means the total demand of the data center, the growth will grow and Don’t fall, “Cox remarked.
