I'm not convinced the AI bubble exists

Matteo Raso, an independent software engineer, published an essay on his site, matteoraso.github.io, on November 8, 2026, questioning whether an AI bubble exists. <cite index="1-2">He notes that online discussion treats a bubble and an imminent pop as settled, a premise he does not accept.</cite> <cite index="1-5">He concedes that some AI companies may have poor finances, but argues that their failure would not amount to a bubble bursting.</cite> <cite index="1-11">If AI does prove to be a bubble, he expects it to resemble the internet bubble, in which the core technology kept its usefulness.</cite> He also suggests the talk is less about finances than about nostalgia for an earlier period, <cite index="1-10">a hope that things can return to how they used to be.</cite>

The essay's argument can be tested against Anthropic's public filing. <cite index="19-1">Anthropic filed its S-1 publicly on October 26, with audited quarterly revenue.</cite> <cite index="20-3">It shows $16.3 billion of first-half 2026 revenue and an $80.7 billion net loss.</cite> Revenue for all of 2025 was <cite index="21-4">$4.6bn</cite>. In the second quarter, the company spent <cite index="25-3">$5.6 billion on compute while generating $11.6 billion in revenue, which it said means the company gets $2.08 of revenue for every dollar it spends on compute.</cite>

Most of the loss is non-cash: <cite index="20-15">the $80.7 billion net loss is mostly a $72.4 billion non-cash charge tied to Anthropic's rising valuation.</cite> Risks remain on the revenue side. <cite index="24-2">About 20% of revenue came from just two customers.</cite> <cite index="24-1">Anthropic says it struggles to secure enough compute from AWS, Google and Microsoft because those hyperscalers increasingly treat it as a competitor rather than a customer.</cite>