The Tape Repriced Compute, Not the Economy
Market regime
The 10/14 session was the most one-sided risk-off print of this sequence. The average US stock fell -1.68% (median -1.08%) on 780 advancers to 2,214 decliners, and 371 names lost 5% or more against just 16 winners — versus 167 and 33 the prior session. Average RSI slid to 43.9 and only 788 of 3,016 stocks still hold their 50-day. Critically, this was not a small-cap liquidity event: 87 of the 896 stocks above $10B fell 5%+, and the damage was worst where balance sheets are least flexible.
Yet the index-level proxies are barely touched, because this was a factor unwind, not a growth scare. Mega-caps ($500B+) lost only -1.45% while micro-caps fell -2.23% — but inside the mega-cap cohort the dispersion is extreme: Information Technology mega-caps -3.63%, Health Care mega-caps +2.39%. Breadth has now deteriorated for four straight sessions (191, 167, then 371 decliners down 5%+) while TA state counts remain split almost evenly (1,065 uptrends vs 1,190 downtrends). That is a market bleeding from the edges of one crowded trade.
Mega forces
1. The cost of AI capital is the new risk factor. CoreWeave fell -15.5% on 3.6x volume with its bond yields near 15%; Nebius -10.1%, Oracle -8.8%, Dell -11.3%, Ciena -10.2%, Coherent and Lumentum -9%+, Flex/Jabil/Celestica -7.5% to -9.6%, Astera Labs and Credo -7.7%/-9.6%, Bloom Energy -9.2%, Vertiv -6.1%, GE Vernova -6.2%. SK hynix ADRs -6.1% with Micron -4.1%, AMD -5.7%, ASML -3.9%. The common thread is not end-demand — it is financing: every name levered to or dependent on hyperscaler capex is being repriced against a 10-year at 4.63% and a 10Y-2Y spread widening to +0.51. The market is charging a discount rate to the AI build-out, and financing-dependent names pay first.
2. Real assets and cash-flow certainty are the destination. Health Care mega-caps rallied +2.39% (UNH +4.7%, MRK +3.8%, ABBV +3.3%, BMY +4.6%, LLY +3.6%), Consumer Staples +2.25% (WMT +3.5%, PM +3.2%, CL +3.2%), XLV +2.5% and XLP +2.2% as ETFs against a -2.37% average thematic ETF. The rotation is toward businesses that fund themselves, not businesses that rent capital.
3. Brazil remains the cleanest non-AI momentum. EWZ +12.8% over one month and +52.5% over a year on the Bolsonaro first-round shock, with EWZS small-caps +23.2%. Money leaving US AI infrastructure is being recycled into a political-repricing trade, not into safety alone.
What's working — and what to avoid
Energy refining is the strongest momentum block that is not AI: MPC +18% in a month (RSI 80, +24.8% above its 50-day), VLO +15.8%, PSX +12.2%, DINO +13.8%. The rally has turned parabolic in the near term, but it is backed by earnings power rather than narrative. Health care leadership is broad and orderly — XLV +4.2% in a month, and 133 large-cap names already oversold has been worked off by price rather than by continued selling.
Avoid, or at minimum stop adding to: anything funding growth with debt. Solar (-18% 1M), industrial storage (-17%), quantum computing (-22%), neoclouds (-17%), Bitcoin miners (-21.5%) and crypto platforms (-16%) are all rate-and-liquidity duration trades now. Gold/silver miners (-13.7%) and silver (-12.7%) broke even with the real-rate move, so the usual hedge did not work — a sign this is a liquidity preference, not an inflation repricing.
Strategy
Do not buy the AI complex simply because it is down; the sector still sports 200 names in uptrends and was the source of the crowding, so the unwind has room. Instead, barbell: own cash-generative health care, staples and refiners as the funding trade, and keep a watchlist of storage/lithography names (NetApp, Everpure, Micron, ASML) for the point when volume exhaustion replaces price exhaustion. Watch CoreWeave's bond yields and the 10-year as the trigger variables — if credit spreads on AI-linked issuers stabilize, the highest-quality compute names drop back into the leadership bucket. Until then, the risk is that 371 decliners down 5%+ becomes the new baseline rather than the outlier.
Put this into practice tonight
Ask Relaxfolio in plain English and get a researched answer in minutes.
Get started