I'm not convinced the AI bubble exists

400 points · 675 comments on HN · read original →

Points and comments are a snapshot, not live.

AI industry spending is high but not unsustainable, with real revenue growth justifying valuations.

The author argues AI companies' massive infrastructure spending (up to $5 trillion committed) is largely one-time, not ongoing. Anthropic's S-1 filing shows $2.08 revenue per $1 compute cost and revenue growing from $41M (2023) to $16.3B in first half of 2026. Major AI companies like Alphabet and Meta have stable P/E ratios around 20, unlike dot-com bubble companies. The lack of a moat is compared to the profitable chicken sandwich wars. Capital remains available through IPOs and public debt, with many AI companies not yet public.

What commenters are saying

Commenters split: some agree usability and potential profits justify investment over a decade, noting AI's impact on markets affects everyone via 401ks. Others highlight parallels to 2007 housing crisis with GPU-backed securities and compute futures. A key correction notes GPU lifespan (5 years) rather than 20-year amortization, multiplying required annual revenue by 4. Meta's margins are reportedly crushed by AI capex. One camp sees Anthropic's $1.2T valuation as bimodal gamble between success and default. Another argues bubble rhetoric reflects wishful thinking that AI will disappear rather than sound financial analysis.