Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
The appearance of cheaper and more efficient AI models in the back of Deep Speek of China could remodel the demand for data centers, which promotes a sector that investors were already betting strongly still booming.
For years, analysts have predicted exponential growth in data centers: the critical infrastructure required to boost the digital transition of the world and the training of large language models (LLM).
China Depseek Ai Model startup sent investors In nervousness At the end of January, as the launch of its R1 model raised questions about the mastery of the United States in the AI sector and if the developer efficiency gains Could the demand for the ability of the data center?
Data centers often take at least two years to build and orders have already been taken into account in 2025, which means that it is unlikely that the launch of the disruptive R1 model has some immediate impact. While the launch of DEEPEEK R1 initially conditioned some analysts who tempered their forecasts by questioning whether the money pumped to the sector could have been something “wrong”, experts told CNBC that the models built cheaper and with less powerful chips could become an accelerator. The market.
Alcista perspective
Deepseek highlights how data centers are vulnerable to changes in narratives around AI spending, according to Barclays analysts. If the efficiency claims made by the Chinese startup are confirmed, the development shows that “the hundreds of billions of dollars dedicated to the development of AI, therefore, seem wrong and the capital spending plans of the hyperscalers They could reconsider, “analysts led by Brendan Lynch said, Brendan Lynch analysts said, they said in A note published on January 27.
They added that, if AI requires less infrastructure, they will be the “lower quality facilities”, which are the least efficient in energy, which could face a weaker demand and a weaker price.
Meanwhile, UBS analysts pointed out that around a third of current data center growth projections are based on the construction and development of generative artificial intelligence, which can create images from written indications. These forecasts do not take into account a radical improvement in efficiency, UBS said in a note on January 28.
UBS initially predicted in April last year that the Global Data Centers market would grow 10-15% in a period of three years until 2028. This week, bank analysts said that the new data and calls of experts finally They lead to a more bullish perspective on the market. The firm now expects the income of the sector to grow 20% in 2025 and see that “the scope is towards the upper end” of the growth range of 10-15% for at least the beginning of the period 2026-2028, analysts said In a note on Wednesday. .
The jury is “even” if Depseek needed 20 to 30 times less computer power for consultation for inference, Andre Kukhnin, an up -to -variable investigation analyst at UBS, told CNBC, referring to the process of executing data through a AI model to make a prediction or solve a task.
“While it is more efficient by Token, it needs more tokens for consultation because it is a reasoning model instead of ‘speech flow’ … The conclusion is that we do not believe that it substantially reduces the demand for inference power.” Kukhnin explained.
The Goldman Sachs Research Department predicts the balance of the supply and demand of the data center “will harden” in the coming years, reaching a peak at the end of 2026 and then moderating from 2027 onwards.
If the efficiency profits drive the lowest levels of capital expenses (CAPEX) of the main investors, that could “mitigate the risk of an excess of supply in the long -term market that we see in 2027 and beyond, what we believe That it is an important consideration that could promote more durability and less cyclicity in the data centers market, “James Schneider, senior investigation analysts of Variable Goldman Sachs, said in a report on February 4.
It has not yet been determined on the impact of nascent technology, less than three weeks since Depseek published its data. R1 is not enough for “changing the needle” when it comes to sue, according to Andrew McMillan, partner of RPC Law Firm.
“The appetite of investors will be tempered if it can be demonstrated that it is replicable and, therefore, there will be a much smaller demand for data processing in the future of what exists now, or at least it will not continue on the same growth route “McMillan said, who specializes in mergers and acquisitions and data governance.
“I think that in the long term, it will really be interesting to see if that structural approach is able to establish itself, and I think that can affect the form of the market.”
‘Fuel to fire’
Stocks vulnerable to changes in The data center market collapsed on January 27. Schneider ElectricThe European company most exposed to data centers according to UBS, lost more than 9%, Siemens energy The actions show 20% and TISSUE closed 6% lower in the day.
Since then, some actions have recovered their losses, recovering from the instinctive reaction of the markets. MEGA-HIPERSCAL GANNICAL DECLARATIONS AS THE ALPHABET Google and Goal He also infused confidence, since both companies pledged to multimillionaire investments After the sale of technology.
There has been “much margin of error” in the sector, said Kukhnin of UBS. “That is why some of the actions have fallen and are not immediately returned, because people already have many of the actions and are now trying to discover if this is the opportunity to add or if it is the opposite.”
He added that the lowest costs indicate a potential Democratization of AIwhich could lead to an acceleration in the adoption of technology, which is “something that is very difficult to quantify.”
The data center market will also continue to be fed by the digital transition, which takes place separately from advances in AI. “The generative AI was almost glazed in the cake, but it has become a very thick layer of glaze, certainly in terms of future growth,” Kukhnin said.
Bruce Owen, president of Emea de Equinix, said that the company is “well positioned as the AI technology curve is bent”, adding that he hopes that the advent of more efficient models will be an “accelerator”, for AI.
“An additional dynamic that we could see is the ‘Paradox of Jevons’, which postulates that the greatest efficiency of a resource can lead to greater consumption of that resource,” he told CNBC.
Ryan Cox, head of AI, at the Synechron consultation firm, also expects the effect of Jevons’s paradox to see the most efficient technology, ultimately, lead to a greater demand from the data center.
“It is a really complex equation,” he told CNBC, noting that there are several winds against and tail winds when it comes to determining possible changes in demand. He shared that Synechron customers are looking for “safe” options to use indirectly Deepseek, as through Hugging Face, a repository of AI models.
“In general, I believe that efficiency will feed adoption, and I think it will continue to increase use, even when these costs decrease. The race towards these more advanced models and the broader applications, the use of AI, means that the general demand The data center will increase and not fall, “Cox said.
