YieldTerminal Pro · October 18, 2026
A note from YieldTerminal Pro
The Qs lost 5.52% this week. CoreWeave and Nebius each lost about a quarter of their value. But the selected Meta, Alphabet and Amazon bonds in our weekly study had firmer last-trade prices.
That split is worth keeping intact. A sharp decline in an AI stock is not the same observation as a deterioration in every AI borrower's debt. And a firmer Amazon bond cannot tell us that CoreWeave's financing has become safer.
Next week brings Netflix, Tesla and Meta earnings. The question running through this issue is what remains after the spending: cash for shareholders, capacity that still needs to earn its return, and chart levels that need a proper reclaim.
Chart of the week: QQQ's gaps are no longer untouched support
QQQ finished Friday at $709.80. The September 17 gap at $711.88–$713.32 was fully traded on Friday; the September 21 gap at $721.73–$727.81 filled on Thursday. Those areas remain dated references, but we should stop describing them as open gaps waiting to be filled.
Below Friday's close, the first references are Friday's $707.23 intraday low, the September 1 low at $704.66, and the September 16 low at $700.00. These are session extremes, not closing lows.
The first recovery question is whether price can regain those filled September areas. Beyond them, the 50-day average at $726.77 and 10-week average at $727.14 sit close together. The 20-day average at $742.52 and the surviving October 13 gap at $744.20–$744.85 are higher repair references.
A daily retest means the test and a daily close back above the relevant level. A touch alone does not establish it. Nearby daily and weekly averages also use different samples; their proximity is not two independent guarantees of support.
Watch the Sunday QQQ deep dive for all five charts, including the broader weekly trend. Friday remained above the 200-day average at $674.91. Near-term weakness and that longer trend are separate facts.
Where things stand
SPY closed Friday at $763.23, down 1.97% from October 9. QQQ closed at $709.80, down 5.52%. Both finished below their completed 20- and 50-day averages, while remaining above their 200-day averages. For SPY those averages were $771.18, $768.50 and $725.01 respectively. These are closing-price changes, excluding dividends.
The latest official 10-year Treasury observation used in our published debt study is 5.10% on October 15, versus 5.24% on October 9. That is an earlier endpoint than Friday's stock closes and corporate-bond observations. We have not calculated a credit-spread change from those mismatched dates and maturities.
The week's watchlist movers
Up: FICO +15.48%, ACN +9.82%, LMT +8.27%, SBUX +7.46%, INTU +7.40%.
Down: AAOI -31.07%, SMTC -27.00%, DOCN -26.78%, MXL -25.83%, COHR -25.48%.
Measured from October 9 to October 16 closes, excluding dividends, across all 201 qualifying US-listed stocks and depositary receipts in the October 16 watchlist with source-reported market capitalization strictly above $5 billion. All 201 had both closes. From 275 watchlist symbols, 20 were not returned by the US scanner, 10 were non-stock listings and 44 were at or below the cap threshold.
This is a retrospective board of our qualifying watchlist, not the entire market or a fixed-membership portfolio. Last week used a $10 billion threshold; this edition uses the October 16 daily screen's $5 billion rule. Membership and the cap rule differ, so the two leaderboards are not a like-for-like measure of market breadth.
AI debt: keep the issuer and the security straight
All eight stocks in our unchanged AI watch group fell. CoreWeave lost 25.21%, Nebius 25.15%, IREN 23.30% and Oracle 9.31%. Microsoft and Meta declined roughly 1%; Alphabet and Amazon also finished lower.
The eight selected bonds are a narrower basket: four Meta notes, two Alphabet and two Amazon. Each had an October 16 last reported trade price above its October 9 observation, by 0.28–1.68 points per $100 face value, with lower reported yields.
Those are differences between trade snapshots, not weekly bond total returns or official closing marks. Coupons, trade sizes and execution terms are not captured by simply subtracting two prices. The basket contains no CoreWeave, Nebius or IREN debt.
Read the full AI Debt weekly update, including the exact securities and source dates. The separate CoreWeave stock-versus-bond review examines the common shares and July 2032 unsecured note. Its coupon, reported market yield and a hypothetical bond-price model are different things.
What happened to the levels?
The September gaps discussed in the chart work were traded through: the September 21 gap filled on October 15 and the September 17 gap on October 16. A later bounce would not make either gap untouched again. They now belong among the historical references a recovery would need to reclaim.
That is a chart outcome, not a scored trading return. An entry, exit and executable rule would be required to make that a strategy result.
The Daily Levels archive keeps the dated charts and their original notes. A later outcome belongs beside the original evidence, whether the level held or failed.
Seven company reviews from the week
Each review retains its own publication-date price basis. The earlier snapshots below are not being presented as fresh Sunday quotes.
- SanDisk: customer payments, joint-venture adjustments and the cash behind the memory boom.
- Nebius: rapid growth, customer advances and the bill for building AI-cloud capacity.
- GlobalFoundries: why a $21 billion project-investment headline is not immediate company revenue or profit.
- TSMC: strong earnings, the N2 ramp and cash left after manufacturing investment.
- Netflix: profitable growth, quarterly cash timing and the October 20 earnings test.
- Diamondback: Friday's energy strength against lower oil realizations and the next cash-flow test.
- Travelers: core EPS grew faster than total core income; repurchases and reserve releases need their context. GAAP and adjusted book value tell different investment-portfolio stories.
Comparisons worth opening
TSMC versus ASML: strong semiconductor demand meets different business models, currencies, cash timing and capacity commitments. A foundry and an equipment supplier are not interchangeable exposures.
Alphabet versus Meta: growth is strong, but infrastructure payments absorb much of the cash. Lease treatment and unusual earnings items matter before comparing free cash flow or P/E.
Qualcomm versus Arm: the same mixed legal verdict drew favorable statements from both companies. The investment comparison still turns on cash generation, growth, stock compensation and the price paid for them.
The week ahead: October 19–23
Our 50-company earnings preview is ready, with a daily chart for every name. It is a selected watchlist, not the whole market calendar.
Netflix reports Tuesday, October 20 after the close. Tesla and Meta follow Wednesday, October 21 after the close. For Netflix, watch the relationship between operating profit and cash generation. For Tesla, separate deliveries, pricing and the margins earned. For Meta, advertising growth and infrastructure spending belong in the same discussion.
Equipment, analog chips and storage bring different questions: Lam Research and Texas Instruments remain vendor-estimated Wednesday dates with release sessions unconfirmed in our calendar; Western Digital is confirmed for Thursday after the close. Aerospace and defense add delivery execution, program margins and cash conversion. Backlog is not current revenue.
Across the slate, 40 reporting dates and 39 release sessions are company-confirmed; 11 release sessions remain unconfirmed. Estimates are labelled. A confirmed conference-call date does not automatically confirm when financial results will be released, and no upcoming result is treated as already reported.
New this week: Test the Theory
The first Test the Theory video asks whether waiting for a breakout to retest actually improves the entry.
We tested 94 completed setups in NVIDIA, Tesla and AMD since January 2020. For this particular rule and sample, waiting for daily confirmation did not improve the average result across all original opportunities. Looking only at entered trades gives a different impression because it leaves out the setups that never retested.
These are event returns, not portfolio returns. Three selected surviving stocks, related signals and idealized first-touch fills limit the conclusion. It is a test worth examining, not a new instruction to buy every breakout.
The research archive, Daily Levels and earnings calendar remain open to everyone. The morning and after-close videos now connect those dated scenarios with the actual outcomes.
Market research only, not a recommendation to buy or sell. Unless specifically labelled otherwise, market context uses completed regular-session prices through Friday, October 16, 2026. Stock price changes exclude dividends; the Treasury endpoint and corporate-bond trade basis are stated separately.