The AI Repricing Was a Multiple Event. Next Week Tests Spending.

The week's own market wrap described the mechanism plainly: a crypto shock that drove violent moves in levered equity proxies until Bitcoin recovered above $70,000 late in the week, and software and AI leadership cracking on a "good isn't good enough" standard. Alongside that, benchmarks rebounded from a poor start to February with the Dow reaching new all-time highs and the bounce spread across sectors — while Amazon, Alphabet and Meta struggled. That combination is not a growth scare. It is consistent with a shift in what investors will pay for a given stream of AI-linked earnings, accompanied by leadership broadening away from the names that had carried it.

The factor evidence points the same way. One sell-side-independent view published this week downgraded the iShares MSCI USA Momentum Factor ETF to hold, citing over 40% US tech weight, deteriorating technicals and a P/E above 25x against broadening market leadership. That is a valuation-and-positioning argument, not a demand argument. Nothing in the week's news flow showed a hyperscaler or enterprise buyer reducing an AI commitment; if anything the opposite, with Goldman Sachs disclosing a partnership with Anthropic to deploy AI agents into accounting and client onboarding. Last week repriced what investors will pay for AI earnings; this week is the first test of whether the spending that produces those earnings has changed at all.

Why Applied Materials is the load-bearing print

Three of the week's reports sit on the same supply chain, and they are not interchangeable. Applied Materials (February 12, consensus EPS $2.21) sells the tools that fabs buy before capacity exists, which makes its order commentary the earliest point in the chain where a change in 2026-27 investment intent would appear. Arista Networks (February 12, consensus EPS $0.758) sits closer to the customer, where hyperscaler network spend converts into shipments. Cisco (February 11, consensus EPS $1.02) reads enterprise rather than hyperscaler demand. NextEra reports February 13, on the power side of the same build-out. The signal hierarchy runs upstream: Arista and Cisco can beat on a backlog that was already booked, while Applied's forward order language is harder to satisfy with work already in hand. That is the number with the most revision risk attached to it.

What makes this more than a preview is that disclosed positioning in the window leans the same direction the calendar is about to test. Among the largest sampled institutional rows — a single filer's January 22 disclosure, so directional colour rather than a market-wide census — the reductions sit in the attention names and index-like compounders, with Alphabet cut 3.1% and Microsoft 1.21%, while the additions sit in silicon and hard infrastructure: Broadcom +0.73%, Amazon +0.77% and Brookfield Infrastructure +5.7%. Separately, among the largest congressional disclosures published in the window, a Senate filing shows a purchase of Eaton — electrical equipment sitting directly in the data-centre power chain — alongside disclosed sales of Lowe's and TJX, while a House filing shows a purchase of IREN. Congressional disclosures across the full window numbered 868 trades; these are the largest sampled rows, and they show positioning, not intent.

The strongest competing interpretation is that last week had nothing to do with AI at all. Bitcoin's break lower tightened risk appetite across levered proxies before recovering, and a broad sector rebound with the Dow at record highs is equally consistent with a liquidity-and-rotation episode in which the AI complex was simply the most crowded thing to sell. That reading deserves weight, and it would also explain the momentum factor's deterioration. This publication favours the capex-verification framing for one reason: the disclosed adds in the sampled data cluster in the physical layer — silicon, listed infrastructure, electrical equipment — while the reductions cluster in the attention names, and the week ahead puts precisely that physical layer's own order books on the tape. When positioning and the calendar point at the same seam, the seam is where the information is.

The second-order problem if orders soften

The macro calendar is the week's second story

Wednesday's labour data will get the headlines and should get less of the attention. Consensus looks for January non-farm payrolls of 70,000 against a prior 48,000, with the unemployment rate held at 4.4% — an estimate set that implies consensus reads December's weak print as a hiring pause rather than labour-market deterioration, since a genuine deterioration would be difficult to reconcile with an unchanged jobless rate. Both payroll figures describe slow hiring, so a modest miss confirms an already-soft picture rather than revising it; the unemployment rate is the variable that would force a rethink if it moves. The one high-impact release where consensus explicitly forecasts deterioration is existing home sales on February 12, at 4.18 million against 4.27 million prior — a reminder that the rate-sensitive part of the real economy is still being marked down while the capex story is marked up. Insider cluster data produced no buy or sell clusters in the window, and provided no directional signal.