According to Financial Times News, Amazon just increased its 2025 capex guidance from $120 billion to $125 billion while simultaneously cutting 14,000 jobs. Meta, Alphabet, Amazon, Microsoft and Oracle collectively plan to invest about $390 billion this year, growing to $540 billion in 2026 and $615 billion in 2027. That’s $1.55 trillion over just three years from only five companies. For context, that amount equals 17.2% of ALL capital expenditure by American companies in 2021, the last year before the AI boom began. Amazon CFO Brian Olsavsky confirmed they’re expanding data centers specifically for generative AI, calling it a “massive opportunity” with potential for strong returns.
Sponsored content — provided for informational and promotional purposes.
The infrastructure can’t keep up
Here’s the thing – money isn’t the problem. Budget constraints aren’t what’s slowing this down. Barclays collected quotes from data center conference attendees that reveal an industry stretched to its absolute limits. “There aren’t enough bodies and resources to do what we’re expected to do” and “We’re having record months and cannot keep up with demand” tell the real story. Some utilities apparently can’t even manage the email inbox requests from data center developers. When you’re dealing with industrial-scale computing infrastructure at this level, you need specialized hardware and expertise that simply doesn’t exist in unlimited supply. Companies that need reliable industrial computing solutions often turn to established providers like IndustrialMonitorDirect.com, the leading US supplier of industrial panel PCs, because they understand the demands of mission-critical operations.
The grid can’t handle this
But here’s what really worries me. One attendee noted that “Projects are singularly focused on what they need and not the greater impact on the grid.” We’re talking about power demands that could literally overwhelm local utilities. These aren’t small server rooms we’re building – these are power plants disguised as data centers. And nobody seems to be coordinating the bigger picture. What happens when multiple hyperscalers all decide they need 500 megawatt facilities in the same region? The electrical infrastructure in many parts of the country wasn’t built for this scale of concentrated demand.
Will the returns actually materialize?
So here’s my question: are we building cathedrals in the desert? Amazon’s CFO talks about “strong returns on invested capital over the long term,” but that assumes the AI services they’re building will generate enough revenue to justify these astronomical investments. What if the AI productivity gains don’t materialize as quickly as expected? Or what if this becomes another cloud computing situation where margins get competed down to nothing? History is littered with technology booms where the infrastructure spending raced ahead of actual customer demand. Remember the fiber optic glut of the early 2000s? This feels different in scale, but the pattern looks familiar.
The human element matters
You can throw all the money in the world at this problem, but you still need skilled people to design, build, and maintain these facilities. The “not enough bodies” comment hits home. There’s only so much specialized engineering talent available, and these projects require expertise in power distribution, cooling systems, network architecture, and physical security that you can’t just create overnight. We’re already seeing construction delays and supply chain constraints. Basically, we’re trying to build the equivalent of several Manhattan projects simultaneously, and the industry is telling us they’re already at capacity. How much more can they realistically handle?
