Wall Street is fueling the AI 'crazy train' (www.businessinsider.com)

Wall Street is pouring new forms of capital into the AI buildout, using structured credit, elaborate borrowing and circular financing to bankroll massive GPU-heavy data-center projects. Reporters and market veterans warn this isn’t inherently novel but is worrying because structured credit disperses and obscures risk across the financial system, making it harder for investors, regulators and journalists to gauge exposure and rein in excess. High-profile founders still chase both profits and prestige, but the financing surge is increasingly divorced from clear, repeatable product revenues. That mismatch matters technically and economically. Analysts estimate GPUs account for roughly 60% of data-center costs, and GPUs depreciate in 3–6 years—far shorter than traditional infrastructure like railroads or fiber—so much of the capex is into quickly obsolescing hardware. The industry’s sustainability therefore depends on whether inference demand and real-world, repeatable AI products can monetize that capacity; many researchers say AGI remains distant and current models still struggle with reliable, bounded answers. The net implication: markets and financiers are accelerating AI deployment, but the durability of returns and systemic risk hinge on productization, revenue models for inference, and clearer risk transparency.