The world's largest data center has failed, and the winner is actually the photovoltaic industry?

What infrastructure is hottest in the AI ​​era? The answer is undoubtedly data centers. However, recent news has poured cold water on the ever-growing boom in computing power infrastructure.

On July 2nd local time, QTS, a data center operator under Blackstone Group, officially announced its withdrawal from a large-scale data center project in Prince William County, Virginia.

According to the initial plan, the project covered 2,100 acres and planned to include 37 data center buildings with a total construction area of ​​22 million square feet, making it the world's largest single data center campus upon completion.

Just recently, Blackstone sold three other data center assets in Virginia for $3.5 billion. This withdrawal and sale clearly demonstrates the shift in the investment strategy of this top global data center operator regarding its core computing power cluster in North America.

In recent years, the explosive growth in AI computing power demand has made data centers a target of global capital competition. Tech giants are expanding rapidly, Wall Street hot money is pouring in, and the entire sector is expanding at a high speed. However, amidst this rapid growth, power shortages are becoming a bottleneck restricting the expansion of computing power.

I. Data Center Power Shortage

Northern Virginia, known as the "Global Data Center Corridor," is home to approximately one-third of the nation's hyperscale data centers, handling over 70% of global internet traffic.

For the past few years, its relatively low electricity prices, lenient tax policies, and well-developed network infrastructure have made it a prime location for global tech giants and investors to deploy computing power infrastructure. However, the explosive growth in computing demand driven by the AI ​​revolution has pushed the local power grid's capacity to its limits.

Dominion Energy, responsible for powering Northern Virginia, currently has over 70GW of pending high-load power connection applications, nearly three times the historical peak load of the region.

In contrast, large-scale transmission and substation expansion projects involve multiple stages, including planning, approvals, and land acquisition, with construction cycles lasting 3 to 7 years. Projects often take several years from application submission to full power supply.

At the current construction pace, data center projects submitting connection applications now will not achieve full-load power supply until around 2030 at the earliest.

As grid queuing times continue to lengthen, electricity costs are also skyrocketing. In the 2027/2028 capacity auction of the PJM electricity market, a capacity shortfall of approximately 6.5GW occurred for the first time, with settlement prices directly hitting the regulatory price cap, representing an increase of more than tenfold compared to market lows a few years ago.

Virginia will also implement a new electricity pricing mechanism starting in 2027, requiring users of large data centers exceeding 25MW to bear at least 85% of the contracted capacity cost, shifting more grid expansion costs onto computing power users and further increasing long-term operating expenses for projects.

Grid expansion remains a distant prospect, access costs continue to rise, and the social environment is tightening. A Gallup poll in March 2026 showed that about 70% of Americans oppose building AI data centers near their homes, with 48% expressing strong opposition. About one-fifth of respondents are concerned that the massive electricity consumption of data centers will drive up local residents' electricity bills.

Therefore, the outcome of Blackstone Group's data center project was already expected. Rather than forcing its way forward and getting into trouble, it was better to proactively withdraw the project and cut losses.

II. Power Bottlenecks Are Reshaping the Data Center Industry

Virginia's predicament is merely a microcosm of the global power supply and demand imbalance driven by the rapid expansion of AI computing power.

For a long time, data center site selection and construction revolved around land, network bandwidth, and capital costs, with power supply assumed to be a readily available basic necessity.

However, the arrival of the AI ​​era has completely rewritten this logic. An AI server equipped with a high-performance GPU consumes several times, or even more, the power consumption of a traditional general-purpose server, leading to an exponential increase in the power demand of high-density computing clusters.

According to industry estimates, by 2030, data centers in the United States will account for more than one-tenth of the total electricity consumption of the entire society, with AI computing power contributing more than half of the incremental demand.

However, the expansion speed of the global power grid is mismatched with the growth period of computing power. A 500 kV transmission line typically takes five years or more from planning to commissioning, while a large data center takes only one to one and a half years from groundbreaking to operation.

This speed difference is creating increasing power congestion in major computing clusters worldwide. From Texas in the US to Ireland and Singapore, insufficient power capacity has become a common hurdle for data center deployment.

This means that the development of the data center industry is shifting from incremental expansion based on "land acquisition" to resource-constrained development based on "power availability." Whoever can first solve the problem of stable, low-carbon, and deployable self-sufficient power supply will gain the upper hand in the competition for next-generation computing infrastructure.

III. Is Solar Power the Optimal Solution?

There's a widely circulated saying in the data center industry: in the second half of AI, it's not about computing power, it's about power.

In the solution to overcoming power bottlenecks, proactively deploying distributed photovoltaic (PV) power stations and energy storage power stations, with their advantages of short construction cycles, high technological maturity, and continuously decreasing costs, is becoming a widely accepted breakthrough path.

For data center projects stuck in the grid connection phase, the most direct value of adding PV and supporting energy storage is reducing grid connection waiting time, effectively smoothing peak electricity loads, and correspondingly reducing the access capacity requested from the grid.

In some computing clusters with scarce power resources, the smaller the requested access capacity, the shorter the queuing approval cycle tends to be. Compared to the grid expansion cycle that often takes several years, the commissioning cycle of photovoltaic (PV) and energy storage (ESS) projects has a significant time advantage.

Besides accelerating project implementation, the hedging effect on electricity costs is equally crucial. Electricity costs have always been a major expense for data center operations, and in recent years, in addition to the continuous rise in electricity prices, the increases in capacity prices and transmission and distribution prices have been even more significant.

Hedge against electricity costs is also indispensable. Electricity costs have always been a major expense for data center operations, and in recent years, in addition to the continuous rise in electricity prices, the increases in capacity prices and transmission and distribution prices have been even more significant. The cost per kilowatt-hour of PV power generation already has a clear advantage; combined with energy storage for charging during off-peak hours and discharging during peak hours, peak-valley arbitrage can be achieved, and additional revenue can be obtained by participating in the demand response and ancillary services markets.

More importantly, the initial construction cost of PV-ESS projects is relatively fixed, essentially providing a safety net for the project's electricity costs for the next twenty years or so, effectively mitigating the risk of long-term electricity price increases.

At the same time, increasingly stringent global green electricity requirements are also forcing the computing industry to accelerate its embrace of new energy sources. Domestically, this year's Government Work Report included "computing and electricity synergy" in the new infrastructure framework for the first time. Previously, the National Development and Reform Commission and other departments issued the "Special Action Plan for Green and Low-Carbon Development of Data Centers," which explicitly requires that newly built data centers at national hub nodes must have a green electricity ratio of over 80%.

This means that photovoltaics has almost become a necessity.

Of course, integrated photovoltaic and energy storage for data centers is not a panacea. Data centers operate at high loads 24 hours a day, while photovoltaics only generate electricity during the day. To cover the entire day's demand, long-term energy storage is needed. Currently, energy storage for up to 4 hours is economically viable, but the cost of longer-term storage remains high.

Site space is another hurdle. Hyperscale data centers have extremely high electricity loads. To increase green electricity self-sufficiency, large-scale ground-mounted power plants are often required. However, in areas with dense computing power clusters, land resources are generally scarce, and land acquisition costs remain high.

These practical constraints objectively exist, but the direction of the industry's progress is clear. Blackstone's withdrawal of its project is not a sign of the decline of the computing power infrastructure boom, but rather a turning point towards moving away from extensive expansion and returning to the fundamentals of electricity. This industrial transformation spurred by power shortages will continue to open up new growth opportunities for photovoltaics.