Amazon, Meta, Microsoft, and Google are gambling $320 billion on AI infrastructure. The payoff isn't there yet. (www.businessinsider.com)

Big tech is in the middle of a data‑center and GPU arms race: Business Insider estimates Amazon, Meta, Microsoft and Google could spend roughly $320 billion on capex this year alone for AI infrastructure, while Meta projects $600 billion through 2028 and a joint OpenAI–Oracle “Stargate” project is pegged at $500 billion. Amazon has signaled more than $30 billion in capex each of the next two quarters; firms are raising capital via bonds, private credit and securitizations to fund sprawling GPU farms. If the bet pays off, hyperscalers could turn raw compute into high‑margin, subscription‑style AI services and spark a new tech growth cycle. The buildout is already reshaping power grids, water use and real‑estate markets. But the business case is far from proven. The boom rests on the scaling assumption that larger models trained with exponentially more compute and data yield commensurate performance gains; critics like Gary Marcus and recent reactions to GPT‑5 argue that returns are increasingly marginal and hallucinations persist. Academic and industry studies show most early corporate AI pilots haven’t yielded clear ROI (MIT found ~95% of initiatives lagging), and Bain warns that meeting demand could require $500B annual capex and ~$2T in new revenue by 2030. That mismatch raises bubble risks and the prospect of stranded assets—parallels to the railroad and dot‑com buildouts—while financiers keep recycling rental income to fuel further construction.