The AI trade's Darwin moment: what seven podcast desks are actually saying

The $67.2B run rate that disappointed

Start with the number that should have been a triumph. On TBPN's Anthropic Investor Day episode, John Coogan reports that Anthropic is annualizing its quarterly run rate to $67.2 billion — and in his own words, that is "remarkable for a company that was doing a tenth of that a year ago." Yet the manifest's summary is blunt: the figure failed to meet aggressive market expectations. The bar for frontier labs is no longer growth; it is growth against expectations set by the largest capital expenditure cycle in technology history, and Anthropic came in under it. m:…2516

The more uncomfortable disclosure was the distillation line item. Coogan reports that "distillation is between 5% to 7% of revenue" — models trained on frontier outputs siphoning demand — which he notes is "a little bit lower than people were expecting." Lower than expected is still a tax on the core product: every percentage point of revenue that comes from unauthorized copying is a point that isn't coming from customers paying for the frontier model itself. Jordi Hays, his co-host, diagnoses the competitive response as "tall poppy syndrome in the frontier" — large tech firms forming what the manifest describes as an anti-anthropic alliance to prevent a monopoly on intelligence. m:…2516

Capital walks

The second TBPN episode, titled SELL EVERYTHING, documents the rotation. Investors are moving out of AI-exposed tech stocks into Treasuries and the Dow, the hosts report, with Altimeter Capital reducing its AI exposure. Brian Gerstner's firm has sold all core AI holdings and reallocated 10% of its book into Meta and Tesla. CoreWeave — the GPU-cloud name that has become the sector's stress indicator — is down 50% in the manifest's key facts. This is not a debate about AI's long-term importance; it is a repricing of who bears the leverage. m:…8668

The credit side of the story came through the NAB Morning Call. Host Phil cites the Financial Times' framing directly: a "Darwin moment" in which data-center bond prices are falling because investors are dumping them in favor of Treasuries. NAB economist Ken Crompton voices the underlying fear: "there's a concern that now increasingly AI is not going to deliver on the promises or even, you know, match the revenue that's needed to service the debt." When the concern moves from equity multiples to debt service, the conversation has changed categories — from valuation to solvency. m:…4628

CNBC's Squawk Box added the capex angle on the same day. Becky Quick said the spending major companies are doing around the technology "has investors concerned that it's maybe out of balance." Andrew Ross Sorkin argued: "If we're ever going to get the cost of this stuff down, you need a lot of competition. And by the way, that competition will ultimately hurt the valuations of all these companies." CoreWeave shares fell more than 15% in a single session on borrowing worries. And NYC Comptroller Mark Levine is calling for a delay of the Anthropic IPO, saying of the existential-risk language: "By their own admission, they're building a technology which poses an existential risk. Those are not my words. They're reportedly in the S-1." A municipal comptroller trying to gate a frontier IPO over pension index-inclusion is a new kind of headline risk. m:…9583

The other side of the tape

Not every desk is bearish. On Fast Money, BofA's Savita Subramanian cautioned that "the good news is priced into the market" and flagged a 20 gigawatt gap between hyperscaler demand and projected capacity delivery — with equity holders in the first-loss position on AI infrastructure debt. That is a warning wrapped in an observation: the demand signal is real, but the capital structure absorbing it is fragile. Supply-chain bottlenecks preventing overbuilding were even described as bullish, a sign of how far the framing has shifted from growth-at-all-costs. m:…2588

The outright bull case came from T. Rowe Price's Page on Bloomberg Surveillance: "Fundamentally, I am optimistic for the next six to 12 months on that trade." His evidence is demand-side — roughly 20% of enterprises now have at least one agentic process in place, with consultants estimating 80% eventually — and he concedes the trade is more levered than it was, with fragile players set to get in trouble. Active management and selective positioning, in his view, not index exposure. m:…8224

The historical coda came from author Liaquat Ahamed on Bloomberg Talks, who argues the AI buildout mirrors the 19th-century railroad boom: debt-financed overinvestment, eroding profitability, and a correction whose timing nobody can call. His numbers are worth sitting with: 80% of railroad money came through the bond market, and by 1872 only one in five railroads was paying a dividend. The railroads got built anyway. The question for this cycle is the same one Ahamed poses — not whether the infrastructure is real, but who finances it and on what terms. m:…5105