How Power, Politics And AI Will Shape Data Center Development In 2026

Last year marked an inflection point for data centers. Artificial intelligence-driven development made the sector a primary engine of U.S. economic growth, turned the industry into a political issue and fundamentally changed how construction is financed. 

If 2025 was the year data centers became unavoidable, 2026 may be the year that the future of the sector’s AI boom cycle comes into clearer focus. 

More than 25 leaders from across the data center industry told Bisnow about their predictions for the trends, changes and challenges that will shape the data center landscape in the year ahead. While the details revealed in their crystal balls varied, common themes emerged about the industry's future.

Industry leaders expect that in 2026, questions about the future of AI demand — and the validity of AI bubble fears — may be answered, while a clearer picture will materialize of how power, politics and AI adoption are redrawing the data center map. 

While there was an unprecedented tsunami last year of announcements for planned multigigawatt megacampus projects far larger than anything built before, executives like DartPoints CEO Scott Willis and Applied Digital CEO Wes Cummins say the next 12 months will reveal which developers can actually execute these projects. 

Cummins called 2026 the year of “put up or shut up,” when the wheat will be separated from the chaff among the flood of new entrants to the data center landscape. 

“2025 was a big year of big deal announcements and big contract signings and a lot of splashy stuff, and now the hard work begins,” Cummins said. “2026 is a year of execution, and there's going to be a lot of people that do execute and a lot of people that don't.” 

Future Of AI Demand Coming Into Focus

Despite the record pace of data center development last year — with construction spending more than doubling — the supply of new data center capacity has remained well behind the AI-driven demand.

Developers haven’t been able to keep up with Big Tech’s desire for facilities to host the graphics processing unit clusters used to develop and run AI models, largely due to constraints on power that have pushed timelines for grid connections to more than five years in some markets. 

In this environment, industry insiders were widely dismissive last year when some prominent commentators warned the U.S. was overbuilding data center capacity. With vacancy rates near zero and new data centers largely preleased years in advance, there was a broad consensus that these data center oversupply fears were unfounded. 

Many within the data center industry anticipate there will be no change to this dynamic in the next 12 months. 

“The bigger risk heading into 2026 isn’t overbuilding,” said Sean McDevitt, a partner with Arthur D. Little. “It will likely be underestimating how fast AI-driven workloads scale.”

While this may be the prevailing majority opinion, a growing chorus of industry voices, including many who spoke with Bisnow, says these concerns about data center overbuilding may become more valid in the year ahead.

Jonathan Hatzor, CEO of Parametrix, said a flood of new data center projects launched since the start of the AI boom is due to come online in the next 12 months. Even if demand continues to grow, this wave of new capacity could flip the supply-demand equation.

“Heading into 2026, the biggest risk for the data center sector isn’t about a lack of demand but the overbuild versus the real demand in AI and data center infrastructure,” Hatzor said. “Just like the dot-com bubble in the early 2000s, massive amounts of infrastructure are being built ahead of proven, durable demand.”

For others, concerns over a potential data center overbuild in 2026 are rooted in uncertainty around the future of AI demand as computing needs shift from training to inference, a trend that is expected to accelerate in the months ahead.

Inference and training often have different site selection requirements, so even if overall demand continues to grow, there are fears that some of the training-focused capacity due to come online this year is ill-suited for the inference workloads increasingly deployed by tenants. 

Still, others say the industry needs to be prepared for the possibility of an overall demand slowdown. 

While there is debate over whether the AI hype cycle and investment wave constitute an economic bubble, the nascent industry is rife with companies with “frothy” valuations but little in the way of revenue or even a product.

With so much AI data center capacity set to come online this year, EdgeCore Digital Infrastructure Senior Vice President Tom Traugott said he anticipates firms like OpenAI and other AI startups will take giant leaps forward in terms of their capabilities. 

But this also makes 2026 something of a make-or-break year for these companies. Should their improvements fall short of expectations or should customer interest not materialize, it could trigger an AI downturn that causes a slowdown in data center demand, said Amanda Garcia, director of business development at TTK USA.

Such a downturn is inevitable, she said, even if the ensuing economic fallout is nothing like the dot-com crash of the early 2000s. 

“Some companies will overbuild. Others will overpromise,” Garcia said. “What we are likely to see is not a crash but a reset. A shakeout, not a collapse.”

Even in the most optimistic demand scenario, there will be unexpected bumps in the road for the data center sector in 2026, JLL Global Head of Data Center Research Andrew Batson said.

One such bump was the January 2025 unveiling of the Chinese DeepSeek AI model, which sent tech and data center stocks into a tailspin amid fears that more efficient AI training would dramatically reduce the need for AI computing. In such an immature, rapidly evolving industry, it is inevitable that more DeepSeek moments — perhaps with greater consequence — will catch the sector off guard.

“The industry is changing so rapidly that there will be a number of curveballs this year, just like we had right out of the gate last year with DeepSeek,” Batson said. “There will be some turbulence this year. What exactly that is is unknown.”

Power And Politics Redrawing The Data Center Map

Hyperscalers’ hunt for power remains the primary force transforming the geography of data center development in the U.S. As the scale of data center projects skyrockets, developers are increasingly willing to build wherever they can access hundreds of megawatts of power quickly.

A fundamental shift in how tech and data center firms pursue power began last year, with more companies turning to on-site or behind-the-meter natural gas generation as the fastest route to electricity. This allowed them to energize facilities months or years before a grid connection could be secured.

Projects that rely on self-generation, microgrids and behind-the-meter power are going to become far more common in 2026, industry leaders say. This bring-your-own-power strategy will continue to accelerate development and construction starts in markets with ample natural gas resources like Texas and Pennsylvania. 

“As the grid struggles to keep pace, natural gas will continue to serve as a crucial bridge to sustainable baseload solutions like geothermal and new nuclear,” EdgeCore’s Traugott said. “The new geography of AI infrastructure will be defined not just by space but by speed and power.”

At the same time, Applied Digital’s Cummins said he anticipates that in the next 12 months, many companies planning gas-powered projects will discover that executing them is far more difficult than building a traditional data center. While there has been a wave of announcements of massive campuses with on-site or behind-the-meter power, Cummins said many of these firms — or their potential end users — may decide in the coming months that it just isn’t worth the effort. 

“Those projects are going to be harder than the actual data center projects,” said Cummins, whose firm built and sold a behind-the-meter data center project in Texas. “My experience with it was not a fun one, so I'm not really pursuing that again.”

Beyond power, community opposition to data centers will play a growing role in where developers look to build in the year ahead. Data center firms must now consider the risk that a project will be delayed or rejected due to community pushback as part of the site selection process.

While local efforts to block data center projects have been on the rise for years, that opposition is steadily becoming more organized and more effective, Milldam Public Relations' Adam Waitkunas said. 

“What were once isolated local zoning disputes are coalescing into a national opposition ecosystem, with environmental advocacy groups and litigators developing plug-and-play playbooks that local opponents can quickly adopt, spanning messaging, research, organizing tactics and emerging legal strategies,” Waitkunas said.

“As a result, opposition in 2026 will be less episodic and more systematic.”

Developers say the need to mitigate these political risks is pushing them away from mature data center markets where the local landscape tends to be more hostile and better organized. Secondary and tertiary markets with few data centers tend to represent the path of least resistance, with no existing opposition groups and local officials often eager to bring in data centers as a source of tax revenue. 

“They can get something developed quickly and there won't be as much NIMBYism, and you get a higher degree of certainty that the data center is going to get done on time,” Greenlight Data Centers CEO Alex Stoewer said. “We think that the Tier 2 and 3 markets are going to continue to get built out faster than the Tier 1 markets.”

Growing demand for AI inference is also reshaping the data center landscape. Exactly what these changes will be is a source of disagreement among data center developers. 

AI inference is generally far more latency-sensitive than AI training. As a result, many within the data center space see inference driving development toward the “edge,” smaller data centers close to Tier 1 markets and major population hubs. 

“The evolving rules around AI training and inference are putting new pressure on latency, making speed a deciding factor much like it was in the early days of search engines,” EdgeCore’s Traugot said. “I expect continued growth in regions like Phoenix, Northern Virginia and the Northeast, where proximity to key markets still matters.”

But other industry leaders disagree. While a few AI use cases like high-speed trading require such low latency that the compute needs to be located a stone’s throw from the end user, such applications are few and far between, Cummins said. 

He said it makes little business sense to build in these expensive Tier 1 markets when the workloads that account for the vast majority of demand could just as easily be located in a regional Tier 2 or 3 market or a regional data center cluster with strong connectivity at a far lower cost. 

“I do not see the push into primary, Tier 1 markets for inference,” he said. “I think that narrative is going to stick around, but I’m just not seeing it.”