By 2027, the five biggest tech companies will spend more building AI than the U.S. government spends on its entire military.
Analysts at The Kobeissi Letter project that combined AI capital expenditures from Alphabet, Amazon, Meta, Microsoft, and Oracle will exceed $1.1 trillion by 2027, surpassing defense spending as a share of U.S. GDP.
In 2026 alone, total AI infrastructure spending is projected to top $800 billion—roughly 2.5% of GDP, almost neck-and-neck with the Pentagon.
Yet, even with all that money, the world’s richest companies are running into a wall getting their projects up and running…
America lacks enough readily available electricity to power everything they want to build.
That growing mismatch between hyperscaler spending and real power capacity is reshaping the AI investment opportunity.
The wall nobody built for
Data centers require more than chips, servers, and construction crews. They need enormous amounts of continuous power… plus grid connections, substations, transformers, transmission equipment, and utility approvals that can take years to secure.
And the U.S. power grid wasn’t designed for this.
Roughly 16 gigawatts of large data center capacity is scheduled to come online in 2026. But industry researchers estimate that 30%–50% of that could be delayed. Power constraints are a major reason, alongside shortages of critical electrical equipment and growing local opposition.
Simply put, Big Tech has the money to build… but the grid still determines how quickly that money can become usable computing power.
What they’re doing to solve it
The clearest evidence of this shortage is how far technology companies are willing to go to secure long-term power.
Microsoft has signed a 20-year agreement supporting the restart of Unit 1 at Pennsylvania’s Three Mile Island site, now renamed the Crane Clean Energy Center.
The reactor closed for economic reasons in 2019. But, backed by a $1 billion federal loan, the plant is expected to come back online in 2027, adding approximately 835 megawatts of baseload power to the regional grid.
Consider what that means: Microsoft is helping bring a retired nuclear reactor back online to secure enough long-term, carbon-free electricity for its growing data center footprint.
Other hyperscalers are likewise signing power agreements, developing on-site generation, and evaluating new nuclear technologies.
The AI arms race is rapidly becoming an energy arms race. And that changes the economics of the entire buildout.
Where the money actually goes when giants hit a wall
When a hyperscaler can’t get the power it needs to build from scratch, it looks for someone who already has it running.
That’s the setup worth paying attention to.
DigiPower X (DGXX) is the perfect example of who stands to benefit. The company controls approximately 200 megawatts of power capacity across its portfolio. And it recently signed a colocation agreement with Cerebras Systems covering approximately 40 megawatts of AI computing capacity.
Under a colocation arrangement, Cerebras supplies its computing equipment while DGXX provides the facility, power, cooling, and supporting infrastructure. The initial 10-year contract is valued at approximately $1.1 billion, with renewal options potentially raising the total value to $2.5 billion.
And remember: That deal covers only a portion of DigiPower X’s broader power portfolio.
Execution still matters. DGXX must finance construction, deliver each phase on schedule, manage development costs, and secure customers for its remaining capacity.
But the market now has a tangible benchmark for what one portion of DGXX’s power portfolio is worth.
The bottom line
The first phase of the AI trade rewarded the companies selling chips and cloud computing.
The next phase will increasingly reward the companies capable of turning those chips into operating infrastructure.
Big Tech’s capital budgets may reach historic levels. Yet every AI facility ultimately depends on physical assets: electricity generation, transmission equipment, substations, land, cooling, permits, and grid access.
Those assets take years to develop. Companies that already control them, like DGXX, hold a growing advantage.
For more analysis on where the AI buildout is heading—and which companies are positioned to benefit—stay tuned to Wall Street Unplugged.
