I’ve seen a lot of bubble finally popping takes this weekend and I don’t think they’re reading the week right, so I want to lay out what I think actually happened. The chip index fell 11.4% last week, its worst since April 2025. But look at where the damage started, September CPI came in at 3.7% against 3.6% expected, the Fed already hiked in September, and the 10 year touched 5.36% the week before, a 24 year high. In that environment the first things to break were the companies funding AI infrastructure with debt rather than operating cash flow. CoreWeave has over $51B of debt and obligations, and its 2031 bonds went above 15%. Goldman has AI related high yield spreads above 350bp, and data centre debt specifically above 400bp, against roughly 250bp for comparable non AI credit. The popular bear case deserves a fair hearing, the IMF flagged roughly $650B of debt tied to $1.4T of AI revenue commitments, much of it circular, where backers fund the companies that then commit to buy from them. If rates stay above 5%, a lot of projects that penciled out at 4% stop working. That’s a legitimate reason to de rate the whole complex. Where I disagree is the indiscriminate selling, Ciena and Coherent each fell 26% on the week. Micron fell 15% while trading on about 12x trailing earnings, against Marvell near 78x. The selloff treated cash generating suppliers the same way it treated the leveraged buyers of their products. Those are different risks, and I think the market will separate them once it gets one clean data point. That data point is Meta on Wednesday, revenue is guided at $61-64B and consensus sits around $63.3B, so revenue probably isn’t the story. The story is that Meta’s Q2 free cash flow fell to $784 million, from $10.6B in Q3 last year, as capex ran at about $51B through the first half. 2026 capex is guided at $130-145B, and the Street has 2027 somewhere around $185-195B. If Meta guides 2027 capex above $200B without Q3 free cash flow recovering meaningfully, the debt funded names have another leg down and the suppliers follow. If cash flow rebounds and the 2027 number lands near consensus, I expect the cash funded part of the trade to recover first, while the neoclouds stay broken until rates come down. The thing I’m genuinely unsure about is the Fed, hike odds for the late October meeting are only around 15% now, but another hot energy driven print in November changes everything above. Meta isn't the only one reporting next week, the stack is full with SPCX, TSLA, IBM, BE, IBM and many more providing read throughs for my book.
CryptoAlerta — análise de criptomoedas e mercado em tempo real