Chinese upstart DeepSeek’s arrival on the artificial intelligence world stage has upended technology giants, at one point slashing more than $1 trillion in value from stock market values in a single trading day.
Yet to some real estate developers, DeepSeek’s release of a new AI model — far less costly than what's produced by publicly traded, U.S.-based big tech firms — could turbocharge already strong demand for data storage facilities in the years to come.
That demand could bring a new round of challenges and opportunities. Data centers developers are already in a race to line up enough power to support new projects as availability of storage capacity nears zero in top-tier markets, property professionals say.
“There is almost no scenario where a proof of concept like DeepSeek that makes AI more tangible, accessible, and cheaper to a wider audience is a bad thing,” Gina Szymanski, real estate investment trust portfolio manager for AEW Capital Management, told CoStar News.
DeepSeek released an AI model that it said in the past week used older-generation Nvidia chips and cost a fraction of the multibillion-dollar investments that behemoths such as Facebook parent Meta, Microsoft, Google, Oracle and ChatGPT’s OpenAI are pouring into the nascent technology. That caused stocks such as Nvidia to plunge in the first day of trading this week after DeepSeek’s platform gained global attention.
DeepSeek’s project, reportedly led by a young team of Chinese engineers using years-old versions of Nvidia chips, could have profound effects on specific companies jockeying to lead the industry. In some cases, that has meant pledging tens or even hundreds of billions of dollars — in the case of the recently announced $500 billion Stargate venture, including the U.S. government, Oracle, SoftBank and OpenAI — to build waves of new data centers and other infrastructure.
The property industry's optimism over cheap AI comes with qualifiers. Some professionals expect no immediate effects, while others express skepticism as to whether the lower-cost approach will pan out.
Even so, DeepSeek’s open-source innovation could actually lower what has long been presumed to be a sky-high barrier of entry to AI innovation, said Rahul Mewawalla, CEO of Mawson Infrastructure Group. The publicly traded firm’s products include modular AI data pods that can be set up on brownfield sites such as former steel plants near nuclear power plants.
“While the industry is still debating what this means, our view is that this is a positive tailwind for the industry because it will mean a much broader spectrum of companies will be able to build AI models for a variety of AI needs,” Mewawalla said. “Before, you were going to need to be one of the big tech firms like Meta or OpenAI to be in this space. This shows you can be out of the box in how you can develop AI models. We think that will increase overall demand for compute.”
More property development
Mewawalla compared today’s AI landscape with the early days of the internet, which once required massive investments to form the core infrastructure.
“Initially, there were just a handful of companies in the internet space, but over time the internet became a core part of every company’s strategy,” he said.
The potential of lower-cost AI developments will likely lead Mawson to increase its development of brownfield sites such as former steel plants in Pennsylvania and Ohio, he said.
“From our perspective, it really drives the view around building more modular data centers that can be much more cost-effective and that we can take to market much quicker,” Mewawalla said. “Agility now is the key to success."
The launch of DeepSeek is a wakeup call to those in the AI industry, with the open-sourced AI expected to catalyze even more data center development, said Christopher Tang, a professor of decisions, operations and technology at the University of California, Los Angeles, and the Edward W. Carter chair in business administration.
"DeepSeek is using new techniques to be cleverer with a more expert learning system that can be faster without the latest chips," Tang told CoStar News, saying he has taken a dive into the mathematical systems of DeepSeek and it seems to perform better, in his opinion, than OpenAI. "DeepSeek opens a new window and creates doubts on if we really need that many data centers and chips to run AI systems. It opens a new set of questions."
China developed this AI system as the country faces bans on chips and needed a different system to work around not having access to new chips, Tang said. The result: DeepSeek showed that low-end chips could handle an AI job, he added.
If DeepSeek's move to open source its AI technology helps influence OpenAI or big tech firms like Microsoft and Meta to do the same, Tang said it could drive down the cost of innovation and invite more brains to the development table, without the need to open a big wallet. The launch of DeepSeek also shifts the catalyst and could cause more demand, in turn, driving more demand for data center campuses, he added.
Innovation and limitations
"This is a Sputnik moment that encourages more people to be engaged and improve AI," Tang said. "This brings more competition. It's a wakeup call for the industry."
It’s very early in the AI technology and application cycle, AT&T CEO John Stankey said in a call with analysts on Monday, and the DeepSeek announcement won’t be the last shock in the cycle.
“We should expect there's going to be days we wake up like this one when somebody comes in and says they figured out a way to get as much benefit out of the model by consuming less power or using less processing capability or they fine-tuned the models to work in a particular domain area more effectively or they can run locally as opposed to in the cloud, which is going to open up and facilitate new applications and business models,” Stankey said.
Cushman & Wakefield's Executive Managing Director Bo Bond said there's no immediate real estate-related effect from DeepSeek, with the data center market "being just fine" despite what he views as a blip temporarily impacting the stock market.
Bill Stein, chief investment officer and co-founder of Primary Digital Infrastructure, is among industry experts viewing DeepSeek’s claimed project cost and methodology with some doubt.
"I don't see how four-year-old processors are going to compete with the newest processors from Nvidia,” he said. “It doesn't make sense."
Data center demand
Stein is working on the Oracle-led data center campus in Abilene, Texas, and said he's been impressed by the national attention garnered after President Donald Trump and the federal government announced its plans to invest more in the nation's data center capabilities. He expects even more interest in the Texas city as a result.
DeepSeek shows what is possible now with that future still yet to be determined, said Michael Rareshide, a partner at Site Selection Group, who specializes in data center site selection throughout the country.
"The AI and high-performance computing market is such in its infancy right now, we can't even project beyond two or three years out," Rareshide told CoStar News.
Decreases in data-center leasing are not expected in the meantime, according to Gina Szymanski.
“Could there be a pause while firms evaluate their computing needs? Sure,” she said. “The source of the computing demand could change. Instead of the long runway of demand expected to stem from hyperscalers’ training models, that runway might be shortened and replaced with demand coming from companies who will utilize AI for new applications.”
More efficient AI algorithms will not destroy the value of tech firms in the long term and could reduce their capital expenditures, or capex, in the process, Szymanski said.
“In fact, investors have been concerned for a while that tech companies were engaging in heavy AI capex spending without sufficient evidence of payback,” she said. “If the news from DeepSeek means that the amount of capex required by tech firms might be less and shorten the length of time it will take to monetize their investments in AI, that actually increases their value and opportunity set.”
Long-term implications
If DeepSeek’s advances can significantly reduce the amount of energy needed for AI data centers, it could relieve pressure on the nation’s power grid, said Andy Cvengros, a Chicago-based managing director at JLL and co-leader of the brokerage’s national data center practice.
It also could reduce the size of eye-popping projected investments, such as Meta earlier this month saying it would invest $60 billion to $65 billion this year to upgrade its AI capabilities, or the even larger Stargate project recently announced by the Trump administration. Last year, Meta announced it would pour $10 billion into a single campus in Louisiana.
“AI is unproven in terms of how much load it’s going to take,” Cvengros said. “AI could take an astronomical amount of power, but we haven’t really seen the leasing trends behind it. How much power is really being consumed is really the metric we’re trying to get to. Right now, it’s just numbers being thrown out there. We haven’t seen these projects actually happen yet.”
If power requirements and project costs wind up being far below early projections, more investors could enter the data-center space, industry professionals said.
DeepSeek's seemingly highly cost-effective model has led to questions about the need for the hundreds of billions of dollars currently being committed by AI firms spurred data center and other related demands, bond rating firm Morningstar DBRS noted in a report Wednesday.
If that proves to be the case, then it could hurt the ratings on debt tied to data centers, at least initially, Morningstar said.
“There could be some short-term disruption affecting credit quality of data centers in the development pipeline and not yet deployed,” the report noted. “If DeepSeek's purported capabilities are actually confirmed independently, a more cost-effective way of performing AI calculations could be beneficial to the credit underpinnings of the entire AI data center ecosystem in the long run.”
Overall, though, Morningstar said it expects existing data centers with strong credit fundamentals to remain largely unaffected.
“The demand for AI — which continues to increase — will require energy infrastructure and significant investment to support it,” said Angie Gildea, KPMG’s U.S. energy lead. “AI remains the disruptive force in the energy industry and the need for AI data centers is not going away.”