The Case Against an AI Bubble
Original: I'm not convinced the AI bubble exists
Why This Matters
AI's financial sustainability shapes investment, hiring, and infrastructure bets industry-wide.
A personal finance analysis argues AI's $5T in spending is largely one-time infrastructure cost, and that companies like Anthropic already earn $2.08 per $1 of compute, suggesting the sector may be more solvent than critics claim.
The conventional wisdom in online finance circles treats the AI bubble as settled fact—a collapse waiting to happen. This essay pushes back on that assumption, though it stops well short of certainty, noting 'nobody can see the future.'
The core argument centers on the nature of AI spending. The industry has committed to at least $3 trillion in future outlays and has already spent an estimated $1.9 trillion, bringing total exposure to roughly $5 trillion. That figure sounds alarming, but the author argues most of it is capital expenditure on data centers—one-time builds, not recurring costs. Once that infrastructure is in place, operating costs drop sharply.
The supporting evidence comes from Anthropic's S-1 filing, which shows a $2.08 revenue return per $1 of compute spend. Anthropic has also begun generating more than its combined compute and infrastructure investments, undermining the argument that data center costs are inherently unsustainable.
On the debt side, a rough calculation suggests $5 trillion spread over 20 years equals $250 billion annually—a large but not impossible figure given that Alphabet alone earned $162 billion in revenue in 2019. Some debt is also structured as convertible notes, repayable in equity rather than cash. Anthropic's $72 billion in share-based losses in H1 2026 reflects its $1 trillion valuation, giving it substantial equity headroom.