The AI “crazy train” that Silicon Valley boarded this summer is barreling forward and now, Wall Street has jumped aboard with it.
Financial institutions are flooding artificial intelligence projects with capital, fueling the boom through increasingly complex borrowing structures and circular financing deals. That infusion of money is powering the construction of massive data centers and GPU clusters at an unprecedented scale but it’s also spreading risk throughout the financial system in ways that worry even seasoned observers.
Newspapers reporter Dakin Campbell, who’s covered Wall Street for nearly two decades, describes what’s happening as both impressive and unnerving. “We’ve seen this movie before,” he said. “Structured credit isn’t inherently dangerous, but it distributes risk in a way that makes it harder to track and understand.”
That opacity, he added, “makes the job harder for investors, regulators, and journalists the people who are supposed to act as counterbalances to excess.”
The risk Wall Street can’t see
Structured credit the same kind of financial engineering that fueled the pre-2008 mortgage boom is now financing AI infrastructure. Banks and private lenders are packaging loans and selling off risk slices to investors hungry for exposure to the booming sector.
The danger isn’t necessarily imminent collapse. It’s the invisibility of where risk resides, and how quickly overexposure can spread if AI valuations start to wobble.
For Campbell, the warning signs are familiar. “It’s not that AI itself is the problem,” he said. “It’s the financial scaffolding that’s being built around it.”
Founders chasing glory and maybe history
Campbell also sees a psychological component driving the frenzy. Tech titans like Sam Altman and Mark Zuckerberg aren’t just chasing profits they’re chasing a place in history.
“At some level, I do believe that Zuckerberg and Altman think there’s a lot of money to be made,” Campbell said. “But their egos are involved in the belief that they could be the ones to usher in AGI and become the legends of history. These are men who grew up reading science fiction. We can’t overlook that.”
That mix of profit motive and messianic ambition has created a feedback loop where AI innovation, hype, and capital seem to feed endlessly off one another a cycle that has defined past technological gold rushes, from railroads to dot-com fiber networks.
The wrong kind of infrastructure?
Many have compared today’s AI buildout to the railroad boom of the 1800s enormous up-front investment followed by lasting infrastructure. Campbell isn’t convinced.
“Railroad tracks and locomotives are long-lived assets,” he said. “That’s not true of GPUs.”
Citing data from tech blogger Paul Kedrosky, he noted that roughly 60% of the cost of a data center goes toward GPUs expensive hardware with a depreciable life of three to six years. By contrast, the physical shells of those data centers buildings, cooling systems, and power infrastructure represent a much smaller portion of the long-term value.
“If Kedrosky is right, less than half of the spending is going into assets that could be considered true long-dated infrastructure,” Campbell said. “It’s more like the fiber overbuilding in the first dot-com boom and that didn’t end well for investors.”
When real products matter more than hype
At some point, Campbell believes, the AI frenzy must translate into tangible, repeatable products that solve real-world problems not just large-scale models chasing the dream of Artificial General Intelligence.
“Inference getting AI models to deliver answers for users is the same goal as creating end products,” he said. “But we’re still several advancements away from AGI. The markets and investors will eventually force these companies to focus less on the science-fiction ideal and more on real, usable outcomes.”
Without that shift, the AI industry risks overbuilding expensive, short-lived infrastructure for products that don’t yet exist a classic symptom of speculative excess.
The value question
Even those enthusiastic about AI acknowledge the limits of its current usefulness. Campbell said he personally sees value in tools like Grammarly or AI systems that assist in research or ideation. But he noted that when asked to perform complex, document-based reasoning, these systems still fall short.
“I do see the potential,” he said. “But when I talk to people, what I hear again and again is that they want AI to help them solve problems easily and repeatably without needing to know the perfect prompt to type in. It doesn’t feel like we’re there yet.”
That gap between promise and practicality is the central tension in today’s AI economy one that’s being amplified by the very financial engines meant to sustain it.
As Campbell put it, Wall Street’s entrance into the AI boom might not mark the beginning of the end but it could mark the end of the beginning.
