Plan for both: AI reshapes the economy and most of this wave's capital is lost

A Business Insider column on the AI spending boom refuses the usual binary: asked whether this is rational investment or speculative bubble, the writer's answer is "both" ^[1]. That is the more useful position for anyone allocating capital, because it converts a market prediction into a planning assumption - the technology reshapes the economy ^[2] while much, possibly most, of the money going in now is lost ^[3].

The mechanism offered is a two-factor test. Bubbles need growth, meaning an innovation that produces enormous real demand, and leverage, meaning debt, credit, circular financing or anything else that amplifies it ^[6]. Both were present in the late 1990s: usage genuinely exploded, with millions of people connecting every month and five years of fundamental growth at dozens of companies including Amazon, AOL, Yahoo and Cisco ^[7], while companies borrowed hundreds of billions of dollars to build telecom networks and buy gear, and internet service providers raised tens of billions in equity ^[8]. The bust did not arrive because the story was false. It arrived when supply caught up with demand, rising interest rates raised the cost of borrowing, and hundreds of IPOs and follow-on offerings exhausted investor appetite ^[9]. Housing rhymed: by 2008 there were no lending standards left to relax and anyone who wanted a house had one, so returns plunged and stretched borrowers stopped paying ^[10].

On the growth side, the current numbers are not ambiguous. The column reports that Anthropic's Claude Code drove revenue to $4.8 billion in the first quarter of this year ^[11] and to $11.6 billion in the second ^[12], roughly 2.4 times the first-quarter figure ^[1], with July results implying a $65 billion annualized run rate, up 7x from last year ^[13]. The arithmetic does not line up cleanly: four times the second-quarter figure is $46.4 billion, so either growth accelerated steeply again in July or the two numbers are not stated on the same basis ^[2]. The Wall Street Journal reports Anthropic is now profitable, which the column treats as evidence against the claim that leading model companies cannot make money, with the caveat that it may not stay profitable ^[14]. OpenAI is reported at $40 billion annualized on possibly different accounting ^[15], about 62 percent below Anthropic's stated run rate ^[3]; the column's view is that OpenAI becomes the Netscape of this era ^[16].

The planning consequence follows from the exposure being systemic rather than sectoral. The column cites estimates that AI investment accounts for a third of US economic growth, so a bust would likely bring a bear market and a major recession ^[5]. Declining to buy AI is therefore not a hedge. What is actionable is the line between assets that survive a write-down and commitments that require the boom to continue: contracted multi-year capacity, debt-funded buildout, headcount and pricing premised on today's cost of compute. The forecast in the column is a bust that clobbers the first wave of companies and investors, followed by a long boom that builds large new companies ^[4], which is an argument for solvency through the gap rather than for absence from the market.

Watch the three historical triggers rather than the sentiment: supply catching demand, the cost of borrowing, and saturation of investor appetite for new issues ^[9]. Watch whether Anthropic's profitability holds ^[14]. And note the provenance: this is one columnist's read, and he concedes an interest, writing that he hopes there is never an AI bust and would be hurt along with everyone else ^[17].