Most investors still think the AI race runs through chips. The Pentagon disagrees.
The Wall Street Journal reported Thursday that the Department of Defense is in talks to lend roughly $5 billion to AI cloud startup Fluidstack, with the money coming from the Pentagon’s Office of Strategic Capital and earmarked to shore up U.S. supply chain and manufacturing capacity for certain data center-related components, rather than funding a new AI facility outright.
Read that again carefully. The U.S. military is not paying to build another warehouse full of GPUs. It is paying to ensure the transformers, power distribution units, and cooling equipment needed to fill those warehouses can be manufactured domestically, at speed. That is a significant signal for long-term investors who know where to look.
Why Manufacturing Capacity Is the Actual Constraint
U.S. data centers are projected to consume between about 6.7% and 12% of U.S. electricity by 2028, according to the Department of Energy, making power infrastructure the first place AI earnings leverage appears, ahead of compute itself. The chips get the headlines. The power gear makes the chips useful. Every GPU cluster needs transformers, switchgear, uninterruptible power supplies, and liquid cooling before a single model can train. Rather than directing billions into another data center, this loan would strengthen U.S. manufacturing and supply chains for equipment needed to build them, including the less glamorous kit required to feed power into enormous GPU clusters and stop them overheating.
The timing follows a Trump executive order in August 2026 that declared a national emergency to secure the bulk-power system and directed restrictions on certain foreign-sourced bulk-power system electric equipment. Washington is connecting the dots between AI capacity, grid infrastructure, and national security in ways that should reshape how investors think about the sector’s industrial backbone.
Fluidstack Is the Vehicle, Not the Story
Fluidstack itself has assembled an impressive roster of strategic relationships. Anthropic announced a $50 billion investment in American computing infrastructure with Fluidstack, including data centers in Texas and New York. Apollo and Blackstone have also announced financing tied to Anthropic’s previously announced expansion of more than 1 gigawatt of compute infrastructure expected to deploy at Fluidstack-based sites starting in mid-2026. Google has been reported to be in talks to invest approximately $100 million at a $7.5 billion valuation, with discussions tied to Google’s broader effort to expand distribution of its TPU accelerators through third-party data center partners. Separately, Palmer Luckey is behind a proposed bank called Erebor, but there is no verified public reporting that Erebor is advising Fluidstack’s Pentagon loan application.
But Fluidstack is the conduit. The suppliers who manufacture the physical components this loan is designed to support are the durable compounders worth studying.
The Industrial Suppliers Worth Owning
Vertiv, headquartered in Westerville, Ohio, produces power and thermal solutions for data centers, including liquid cooling. Eaton, which is incorporated in Ireland, designs and manufactures electrical equipment such as switchgear and uninterruptible power supplies used in data center power systems. Both are already embedded deeply in the AI buildout.
Vertiv said its order backlog was about $15.0 billion as of December 31, 2025, driven by a massive surge in data center orders. In Q1 2026, Vertiv reported revenue of $2.65 billion, up 30% year over year, with the company expecting full-year revenue of $13.5 billion to $14.0 billion. Eaton has reported strong demand as well, including 48% year-over-year total backlog growth in its Electrical sector in Q1 2026.
The competitive moat here is underappreciated. AI data center infrastructure creates a specific switching-cost structure: liquid cooling architecture decisions made for a hyperscaler’s 2024 to 2026 buildout cycle are not reversible on a 12-month timeline, with rack layout, thermal management, and facility design all structured around the chosen supplier. Once Vertiv or Eaton is designed into a facility, they are designed in for years.
What Could Go Wrong
The Pentagon loan is still in talks, not closed. The Office of Strategic Capital operates with a specific mandate, and not every negotiation reaches completion. Beyond that, both Vertiv and Eaton trade at valuations that already reflect strong expectations, meaning any deceleration in hyperscaler capital spending would pressure their multiples before it pressures their backlog. Tariffs on imported components add a cost variable that neither company fully controls.
The deeper risk is that Washington’s push to onshore component manufacturing takes years to translate into domestic production capacity, potentially creating a period where demand outruns supply at precisely the moment the loan is meant to ease that constraint.
The Long-Term Verdict
When a defense office historically focused on critical minerals and other defense supply chains decides its most urgent lending priority is AI power equipment, the message is clear: the physical supply chain for artificial intelligence has become a strategic asset class. Investors who spent the last three years concentrated in chip designers and hyperscalers may want to ask whether the businesses that manufacture the hardware those chips cannot run without deserve a permanent seat in a long-term portfolio.
The Pentagon just answered that question with $5 billion.
