The AI Buildout Found Its Price Tag — And a Fed Chair Willing to Argue About It

The week of October 26 belonged to Anthropic, though not in the way a single IPO filing usually dominates a week. The company filed for a US listing at a valuation approaching $2 trillion, and within forty-eight hours that document had done more to reprice global markets than any central bank communication. Asian chip stocks rallied in coordination. Analysts published maps of which cloud, networking and power names would benefit. Broadcom's bulls and bears, who agree on nothing, converged on the view that Anthropic-related funding risk sits unpriced in the stock. The filing's headline $80.7 billion six-month loss turned out to be mostly an accounting entry tied to how stock compensation and prior funding rounds are carried on the books, and the market read through it within a day — a telling detail, because it shows investors were not reading the S-1 for Anthropic's profitability. They were reading it as a purchase order. That is the week's central theme: the AI capital-spending cycle stopped being a narrative and started producing auditable numbers across an improbably wide set of companies. Celestica's revenue rose 83% on AI networking and compute. Amkor's computing revenue climbed 37% as TSMC raised advanced packaging prices by up to 20%, identifying packaging rather than fabrication as the tightest link in the chain. SK Hynix posted a sevenfold jump in operating profit, topped 100 trillion won in revenue at an 80% margin, and said its output is sold out through 2029 — a disclosure that pulled Samsung, Kioxia and Micron higher on sympathy alone. India's Netweb reported revenue up 336%. Advantest raised its outlook. Then the evidence spread past the semiconductor complex entirely: a glass maker, a building-controls firm and a cell-tower REIT all posted double-digit growth tied to data-center construction, and Kirby, Littelfuse, Terex, Flex and IDEX each reported quarters pointing the same direction. When five unrelated industrials confirm the same story, it is no longer a sector trade. The financing side of that buildout is where the week's anxiety concentrated. SpaceX closed $40 billion in debt to fund its chip-rental business, and Musk used the occasion to predict that rival neocloud providers would collapse under their own spending — a striking forecast from a man who had just borrowed $40 billion to compete in the same business. One analysis put the potential Nvidia revenue from SpaceX's plans at $400 billion. Then Bloomberg reported that Anthropic accounted for more than a third of SpaceX's third-quarter revenue, which recasts the whole arrangement: debt raised by one private company, flowing to chipmakers, to serve capacity rented largely by another private company that is itself about to ask public markets for money. Oracle's response was the week's most interesting counterprogramming — it capped future fundraising for its data-center buildout at $20 billion and projected $30 billion in free cash flow by 2029, and the stock rose 12%. Investors rewarded a company for promising to stop borrowing. Meta, by contrast, saw its reported $95 billion AI infrastructure plan cited as a drag on the same Nasdaq session that Anthropic's filing lifted, and Steve Eisman's warning that an OpenAI failure could trigger a recession found an echo in the Meta downgrades issued for precisely that reason: capital spending with no margin of safety if growth disappoints. The hyperscaler results split that question down the middle. Microsoft's cloud revenue growth accelerated to 19%, answering the market's core worry, and the stock rose. Alphabet fell hard enough to post its worst one-day drop in years and drag index futures with it. Amazon's AWS results came in modest with weak advertising alongside, and the stock fell despite management reiterating that cloud capacity is effectively sold out. Three companies, one thesis, three verdicts. The second macro story was Kevin Warsh's first meeting as Fed chair, which produced a hold on a 10-2 vote — the widest dissent in recent memory — alongside a reduction in the policy calendar from eight meetings to seven. Warsh said the economy is "not running too hot," and third-quarter GDP growth of 3.4% gave the hawkish dissenters cover to argue no further support is needed. Gold, which usually gains when policy looks set to loosen, sold off on the split, then spent the rest of the week pinned near $4,000 as hot PCE data revived talk of hikes rather than cuts. The policy divergence widened elsewhere: the ECB signaled a December hike followed by a stop, and New Zealand's governor put December odds at "high," while the BOJ's hold was read as hawkish enough to keep a December move alive — a reading that helped push the yuan to a four-year high. For any multi-currency portfolio, these are not abstractions; the Buy List 2026 book's gap between price return and total return reflected exactly this mix of income and currency effects. Underneath the AI boom, the consumer economy looked considerably worse. Reckitt cut guidance and Beiersdorf did not, despite both missing on the same weak demand, and the softness spread to US appliances and meat processing with no supply-chain link between them. Mortgage rates above 7% squeezed Builders FirstSource and Forestar at opposite ends of the homebuilding chain, and produced the week's sharpest real-estate contrast: Blackstone Mortgage Trust absorbed $800 million in impairments while Kimco posted record occupancy. American pay growth trailed inflation for a second straight quarter, and the 2027 Social Security adjustment landed at 3.6% to immediate complaints that it understates what retirees actually spend. Airlines responded to the same environment by chasing fares rather than load factors. Churchill Downs and VICI both chose buybacks over construction. China offered a version of the same divergence: industrial profits rose 14.4%, but the gain concentrated in AI-related manufacturing rather than the broader base, and Citic Securities fell on soft profit, adding a fundamental crack to a regulator-driven bear market. Brazil was the week's political shock. Lula won a fourth term in a knife-edge runoff and markets sold the result rather than the uncertainty — PagSeguro crashed, the broader index fell, and Flavio Bolsonaro's refusal to accept the outcome added transition risk to an already unresolved fight over congress and debt sustainability. The FT's framing was that the political center of gravity shifted rightward even as Lula took the presidency, which is the kind of outcome that keeps a risk premium attached to Brazilian assets well beyond inauguration. A persistent sub-theme deserves naming: beating estimates stopped being sufficient. Nokia beat and earned an upgrade only to Reduce. HSBC's profit rose 32% and its ADR still missed. Robinhood posted a record $1.4 billion quarter and fell 7%. Stryker and Illumina both beat and both sold off. EQT announced a $50 billion merger and declined on the day. Guidance, spending discipline and the height of the bar now matter more than the print itself. What carries into next week: whether Alphabet's and Amazon's results mark the start of a broader reckoning with AI capex that Microsoft's cloud number simply hasn't reached yet; whether the Fed's 10-2 split becomes its new normal and what the dissenters actually wanted; whether the ECB delivers its signaled December hike; and whether Anthropic's roadshow holds a near-$2 trillion valuation once the S-1's warning signs get priced rather than read through. Oil and gold remain hostage to an Iran situation pointing in two directions at once — Khamenei signaling it is "time for a deal" while prices rose on the prospect of fresh US military action. And uranium above its 2007 peak with Cameco down 42% from its record remains the cleanest reminder available that a strong commodity price and a re-rated equity are different things entirely.