Analytics & Product StatusBird PositionBird Sensa Wellness Investing Contact
Back to Investing

Chips Lost 11.4% and Bonds Rallied. The AI Trade Is Now a Credit Trade.

Hot CPI, hot PPI, and the long end still fell 15 basis points while the semiconductor index lost 11.4%. The repricing happened in credit spreads, not in rates.

Chips Lost 11.4% and Bonds Rallied. The AI Trade Is Now a Credit Trade.

The AI trade did not break on a growth number. It broke on a credit spread. The Philadelphia semiconductor index fell 11.4% last week, its worst since April 2025, across eight straight losing sessions and its longest streak since 2008. The Nasdaq Composite closed at 26,140.03, down 4.5% and its worst week since June. The S&P 500 finished at 7,659.44, down 1.95%. The Dow closed at 52,002.41, up 0.7%, carried by staples and health care while technology lost between 5% and 6%.

Then the part that should not fit. September CPI ran hot on Wednesday: headline up 0.6% on the month and 3.7% on the year, core up 0.3% and 2.6%, both a tenth above consensus. Thursday's PPI was worse at 6.3% against 5.3% expected. Retail sales rose 0.8% and claims printed 199,000. The long end rallied anyway. The 10-year closed at 5.085%, down 15 basis points after spiking to 5.33% on the print, and the 30-year fell 17 to 5.43%. Hot inflation, falling yields, collapsing chips: that has one explanation, and it is not the Fed.

Scorecard: October 12

Seven rows. One right, four wrong, two split. The worst week this column has put on the board.

One line from last Monday earned credit and a correction at once. The column wrote that AI capex was migrating onto credit, and that credit reprices with the long end. The first half was the whole story. The second half was backwards: credit repriced while the long end rallied, because money leaving CoreWeave paper went into Treasuries rather than out of duration.

The Repricing Happened in Credit, Not in Rates

CoreWeave is the instrument that matters now. Its five-year credit default swaps blew through 1,000 basis points and its bonds traded to yields near 15%, levels that price real default risk, against roughly $51 billion of debt and leases. Oracle's five-year CDS hit a record near 288 basis points with its rating one step above junk. Barclays has AI-linked risky corporate spreads wider by more than 50% in six months.

None of that is a demand problem, which is why calling it a bubble popping is the lazy read. TSMC reported $46.72 billion of revenue on Thursday, up 41.2% and above the top of its guided range. ASML printed a record 11.66 billion euros and said demand may outrun supply through 2028. Both sit inside a sector that just fell 11.4%. The question changed rather than the answer. This trade is now priced off who funds the buildout and at what spread, and the levered neoclouds set that price for everybody.

CXMT Did Not Ship a Chip. It Shipped a Doubt.

The memory leg fell for its own reason. Over the prior weekend CXMT, China's state-backed DRAM maker, said it had made significant progress on an architecture that packs more data into less space, then followed it by targeting server memory. Micron fell 5.63% Friday to $878.84 against $1,029.00 a week earlier, a 14.6% week. SanDisk and SK Hynix each dropped more than 6%.

Goldman Sachs says the market is overestimating how fast CXMT closes that gap, and the manufacturing objection is the serious one: an architecture announcement is not a commercially viable yield. Contract pricing on server DRAM is still up on the order of 130% across two quarters. Micron ran up nearly 230% this year and SanDisk nearly 500%. A stock that has tripled does not need a demand shock to fall 15%, it needs a reason to doubt how long its pricing power lasts, and CXMT supplied one for free.

Hot Prices, Cold Hike Odds

Warsh praised a synchronous global tightening from Bangkok on Friday and signaled nothing, leaving October 28 near 20% and December near 92%. Take the September minutes at their word and that is coherent: this is a supply-side price level driven by energy and AI investment rather than wages. Brent settled at $109.86 with mines reported in the Hormuz traffic lanes. A central bank cannot hike away a mined shipping lane. It can hike into one, and in December it probably will.

This Week

Key Levels for the Week

S&P 500, on a closing basisHolds 7,575 and reclaims 7,765 by Friday
Philadelphia semiconductor indexSnaps the eight-session streak, closes the week higher
MicronHolds $850 on closes
CoreWeave five-year CDSBack inside 1,000 basis points
10-year yield, weekly closeBetween 5.00% and 5.25%
Netflix Tuesday, Meta and Tesla WednesdayNetflix above $0.82, Meta revenue above $63 billion, Tesla at or above $0.45
BrentWeekly close below $112, absent a confirmed Hormuz closure

My Take

Three weeks ago the risk here was breadth. Two weeks ago it was inflation. Last week it was neither. It was the discovery that a very large AI buildout has been financed in the credit market, and that the marginal lender has started asking for more. Nothing in the demand data broke. What broke was the price of the money, and that reprices every multiple built on a 2029 cash flow.

Playbook: watch CoreWeave's CDS rather than Nvidia's chart, because a move back inside 1,000 basis points ends this drawdown and a wider one extends it; keep Nvidia at BUY with the $268 target at $214.13 while owning that the rating has cost money for two months, since a 6.6% week on no company news is multiple compression rather than a demand signal; stay with memory but respect the level, holding Micron above $850 on closes and treating the CXMT claim as a yield question until somebody shows a wafer; own Meta and Tesla into Wednesday for the capex guide rather than the quarter, and fade any Meta rally that arrives with 2027 capex guided down; take the profit on the 10-year bought between 5.25% and 5.35% and do not chase 5.085%; and treat staples and health care as the hedge they just proved to be rather than new leadership.

Disclaimer: This is not financial advice. All analysis is for informational and educational purposes only. Do your own research before making any investment decisions.