Behind China’s DeepSeek, the emergence of cheaper, more efficient AI models could reshape demand for data centers, thus facilitating a sector where investors have already bet heavily will continue to thrive.
Over the years, analysts have predicted exponential growth in data centers, the critical infrastructure needed to power the world’s digital transition and large language model (LLMS) training.
Chinese startup DeepSeek AI model sends investors In trouble In late January, with the launch of its R1 model, questions about the dominance of the AI ​​field were raised Improve efficiency Possible demand for data center capacity suppresses.
Data centers usually take at least two years to build, and orders have been taken into account in 2025, meaning the launch of a disruptive R1 model is unlikely to have any impact immediately. While the launch of DeepSeek’s R1 initially led some analysts to respond to their forecasts as they questioned whether the money pumping the industry was somewhat “misleading”, experts told CNBC that the model of CNBC is cheaper and has lower features. The chips may eventually become an accelerated chip market.
Bullish outlook
According to Barclays analysts, DeepSeek highlights how data centers are vulnerable to shifts in AI spending narratives. If efficiency requirements raised by Chinese startups are confirmed, development shows that “hundreds of billions of dollars dedicated to AI development seem misleading and can reconsider a large number of evaluating capital expenditure plans,” said Brendan Lynch analysts. Analysts led by Brendan Lynch, explains. exist Note published on January 27.
They added that if AI requires less infrastructure, it will be the “lowest quality facility” (the lowest energy efficiency) and may face weaker demand and weaker price facilities.
Meanwhile, UBS analysts pointed out that about one-third of the current data center market growth forecasts depend on the establishment and development of generative artificial intelligence – AI can create images from written prompts. UBS said in its comment on January 28 that these predictions do not take into account fundamental improvements in efficiency.
UBS initially predicted that the global data center equipment market would grow by 10-15% in April last year. This week, analysts at the bank said new data and calls from experts ultimately lead to a more lasting market outlook. Analysts said in a Wednesday note that the company now expects revenue from the division to grow by 20% in 2025 and believe that “the range will be towards a 10-15% growth range” starting at least during the 2026-2028 period. growth range. .
Andre Kukhnin, an equity research analyst at UBS, told CNBC that the jury had to reason about whether DeepSeek needed 20 to 30 times less computing power to reason about. Or solve the task.
“While each token is more efficient, each query requires more tokens because it is a reasoning model, not a ‘voice stream’…most importantly, we don’t think it greatly reduces reasoning The demand for power,” explained Cookin.
Goldman Sachs’ research department predicts that the balance of supply and demand in data centers will “tighten” in the next few years, reaching its peak by the end of 2026, and then adjusting from 2027.
If efficiency improvements promote lower capital expenditure (CAPEX) levels for major investors, that could “ease the risk of long-term oversupply that we see in 2027 and beyond – we think this is an important consideration in the data In the central market, James Schneider, senior equity research analyst at Goldman Sachs, pointed out in a February 4 report.
Less than three weeks since DeepSeek released the data, the impact of new technologies has not been determined. Andrew McMillan, a partner at RPC law firm, said R1 itself is not enough to “change the needle.”
“If it can be proven to be replicable, then investors’ appetites will be eased, so the demand for future data processing will be much lower than it is now, or at least not on the same growth path.” McKeep McMillan specializes in mergers and acquisitions and data governance.
“I think it’s really interesting to see if this structural approach can take over in the long run, and I think that will affect the shape of the market.”
“Fire Fuel”
stock Easy to change The data center market plummeted on January 27. Schneider ElectricAccording to UBS’s data centers, European companies that are most likely to be exposed to data centers lost more than 9%, Siemens’ energy Stocks are down 20%, abb Closed 6% on the same day.
Some stocks have since recovered from losses from the market, recovering from the market’s knee reaction. Revenue statements from Mega-Hyperscales, such as Alphabet Google and Yuan Also instilled with confidence, as both companies promise Billions of dollars in investment Follow the technology sell-off.
Kukhnin of UBS said there is no “room to go wrong” in the industry. “That’s why some stocks have fallen and are not immediately repurchased because people already have a lot of stocks and are now trying to figure out if this is an opportunity to add, or the other way around.”
He added that lower costs indicate potential Democratization of AIwhich could lead to the acceleration of adoption of technology – it’s “things that are difficult to quantify”.
The data center market will also continue to be driven by a digital transition that is separate from advances in AI. “The generated AI is almost icing on the cake, but it has certainly become a very thick layer of frosting in terms of future growth,” Kukhnin said.
Equinix’s EMEA president Bruce Owen said the company was “in good shape as the curve of AI technology is bent,” adding that he hopes the emergence of more effective models is “acceleration” for AI.
“The other dynamic we might see is ‘Jevons Paradox’, which believes that increasing the efficiency of the resource may lead to greater consumption of that resource,” he told CNBC.
Ryan Cox, head of AI at the AI ​​consulting firm Synechron, also expects the Jevons Paradox effect to see more effective technologies that will ultimately lead to more data center demand.
“It’s a very complex equation,” he told CNBC, noting that there are several headwinds and headwinds in determining potential shifts in demand. He shared that Synechron’s customers are seeking “safe” options to use DeepSeek indirectly, such as by embracing Face, a repository of AI models.
“Overall, I think efficiency will drive efficiency in adoption, and I think it will continue to drive usage even if those costs are reduced. Overall data center demand will increase without decreasing,” Cox noted.
