Political Party Using
The $745 Billion AI Bet With No Clear Business Model
NYU professor Aswath Damodaran — whose valuation frameworks are used in business schools on every continent and who is sometimes called the Dean of Valuation — published analysis this week warning that Microsoft, Amazon, Meta, and Alphabet combined could spend $745 billion on artificial intelligence this year without a defined path to returns. That is not a marginal concern from a peripheral skeptic. Damodaran has spent decades valuing companies across sectors, and when he says he cannot find the business model, the warning carries institutional weight.
Bank of America is flagging AI valuations as approaching dot-com era extremes. Their specific data point is striking: all six Magnificent Seven stocks that reported earnings last week saw price swings exceeding what options markets had priced in — and this was apparently the first time that has occurred across all six in the same week during the ChatGPT era. The options market is conventionally regarded as sophisticated on volatility expectations. When reality consistently exceeds those expectations, either the market is structurally mispriced or something about AI earnings is fundamentally harder to model than previously assumed.
Nvidia is not waiting for that debate to resolve. The chipmaker committed up to $3 billion to Lancium and joined a $2 billion capital raise for Firmus this week, deepening its stake in physical AI infrastructure. Whether or not AI software business models ultimately pencil out, the infrastructure buildout generates real spending that flows through Nvidia's order books. The company appears to be constructing a vertically integrated position — selling chips, helping build the facilities that house them, and holding equity in the companies that own those facilities.
A stark data point frames the gap between ambition and execution: only twenty percent of planned U.S. AI data centers are actually under construction. Some of that gap reflects genuine bottlenecks in permitting and grid connection queues. But some of it also depends on revenue projections that have not yet materialized. Amazon is reportedly building a $2 billion data center in Gilroy, California, without residents' knowledge — the secrecy is the story, but the underlying fact is that the buildout continues at a pace that outstrips public awareness.
HR and finance software company Rippling reportedly burned through millions in unchecked AI token costs before launching what it is now calling an AI Spend Console — a dashboard for enterprises to track and control AI spending. The fact that a sophisticated technology company with experienced engineers and finance systems ran up millions in unintended AI costs is a micro-illustration of what Damodaran is describing at the macro level. The companies now selling cost-management tools for AI spending are themselves a new industry created by the inefficiency of the AI industry — a feedback loop worth monitoring.