The central flaw in the AI investment narrative is not technological — it is arithmetic. For every dollar flowing into artificial intelligence infrastructure, the financial returns arriving on the other side are proving stubbornly, structurally inadequate. This is not a temporary lag. It may be the defining economic story of the next several years.
Consider Google, the undisputed industry leader. Since ChatGPT’s launch in late 2022, the company has deployed an additional $85 billion in cumulative capital expenditure on AI-related infrastructure. The annualised revenue attributable to that investment stands at roughly $21 billion — representing a modest 5% uplift on total revenues. Worse, this revenue carries low margins, a sharp contrast to the high-margin search and advertising franchise that built the company’s legendary profitability. Google’s capex now exceeds its net profit — a threshold once unthinkable for a business celebrated as the archetype of capital-light software economics.
If the industry leader cannot generate compelling returns on AI capital deployment, the implied trajectory for competitors is considerably darker. This matters enormously for investors who have priced AI infrastructure players — chipmakers, data centre operators, hyperscalers — as though the demand curve runs permanently upward.
The deeper structural risk is commoditisation. Large language models are converging toward functional equivalence for most enterprise applications. When models become “good enough” across a sufficient range of tasks, competition inevitably shifts to price. This is not a novel dynamic: it is precisely what happened to cloud storage, telecommunications bandwidth, and semiconductor memory. The economic gravity of commoditisation is relentless, and it compresses margins across entire ecosystems, not just individual players.
History offers an instructive parallel. The late 1990s telecommunications boom was sustained by projections of perpetually doubling bandwidth demand. Those projections were briefly accurate, then catastrophically wrong. Capital poured in anticipating exponential growth; the growth curve normalised; the capital was largely destroyed. AI’s “killer app” narrative carries recognisable echoes.
The sophisticated question for investors is not whether AI matters — it clearly does — but whether the current valuation of AI-adjacent equities reflects a durable earnings reality or a collective suspension of financial discipline. The arithmetic, examined carefully, suggests the latter. Knowing where the flaw lies is not a reason to abandon the thesis. It is, however, a reason to price the risk honestly.
Source: Raw/trigger-why-we-worry-part-1.md