Long Yields Hit a 19-Year High; CoreWeave Default Odds Near a Coin Flip
US Stocks · Options · News · Views
Soft Data, but the Long End Is the Pressure
Tuesday’s tape, August 18, sold off on volume, with Nasdaq leading. Semiconductors gave back last week’s rebound. Software names were mixed inside the group and still held up better than semis; the two sleeves keep taking turns, and neither has completed a breakout. Industrials and materials lagged. Energy, health care, and staples were the relative bid.
The data was soft. It was not the driver. Import and export prices came in below expectations, extending July’s run of tame inflation prints. Housing starts fell 12%, also below forecast. Building permits, the more forward-looking print, rose 5% and beat. Industrial and manufacturing production were roughly in line. The Atlanta Fed’s Q3 GDP nowcast was cut from 4.3% to 4.0%. The odds of the Fed holding in September remain around 65%. What actually pressed sentiment was the long end of the Treasury curve.
Home Depot Holds Guidance; Treat It as Defense
HD’s fiscal Q2: adjusted EPS $4.92 versus $4.73 expected and $4.68 a year ago. Net sales $47.86 billion, up 5.7% year over year, versus $47.33 billion expected. Comparable sales +1.7% versus +0.9% expected. Guidance was left unchanged: full-year revenue growth 2.5% to 4.5%, comps flat to +2%. Management said tariff refunds helped offset higher fuel costs.
The beat did not rewrite the multiple. On a January 2027 fiscal year, the forward range is about 258 to 431 with a 344 median. The stock around 337 sits near that median. January 2028 is about 281 to 469 with a 375 median, so next year still has a little room. The 328 to 336 zone near the tape is fine. A more conservative entry still wants a lower overlap of technicals and fundamentals.
The worst of the earnings trough is likely behind it. The print is stabilizing at a bottom, not accelerating. This year’s growth is weak; next year gets marked up some, but that is still about a year and a half out, with the cycle and rates still moving. Growth is not fast. The stock is in a defensive phase. A larger move needs a break of 366. It has chopped there for nearly a quarter. What is missing is a clear raise in guidance.
Data-Center Default Insurance Starts to Reprice
Long-term yields jumped. The 30-year printed as high as 5.318 on Monday, a roughly 19-year high and the highest since 2007. Japan’s 10-year is at a high not seen since 1996. Global long ends are elevated. If this is a trend, the pressure transmits through three channels: growth-stock valuations, AI financing costs, and crowding-out of the rest of the economy.
The valuation channel is the 10-year, not the 30-year. DCF models use a 10-year factor, so a higher 10-year compresses tech multiples. The financing channel is more direct. Higher long-term borrowing costs make debt-funded hyperscale CapEx less attractive. Treasuries already pay. To raise more money, issuers have to offer more. If they cannot, capital can just sit in government paper. Annual AI CapEx may already be heading through $1 trillion. 2027 is expected to grow off 2026, but the increment is wide — $1.1 trillion versus $1.5 trillion versus $1.7 trillion is a different cycle. Mega-cap free cash flow plus private credit cannot cover that scale, so the gap is bonds, and some equity issuance. Year-to-date AI debt financing is already $489 billion. At the end of 2025 the expected ceiling was $322 billion. That is about $160 billion extra, with four months of 2026 still left.
If too much capital piles into data centers, other parts of the economy get crowded out. Credit-default swaps move with that stress. Data-center CDS had eased, then started bouncing again, though not back to the peak. A higher CDS is more expensive insurance on those bonds, and a higher implied default risk.
The main names sit roughly here: CRWV (CoreWeave) 728.82; ORCL 203.94; SPCX (SpaceX) 161.37; META and NVDA both near 80, almost the same; then GOOG, AMZN, and MSFT. Converted into five-year cumulative default probability, that is about 45% to 48% for CoreWeave — close to a coin flip; 15% to 16% for Oracle, about one in six; about one in eight for SpaceX; about one in fifteen for Meta and Nvidia; about one in twenty-two for Amazon and Google; and about one in twenty-eight for Microsoft, which has barely borrowed and still has unused free cash flow. Microsoft, Google, Amazon, and Meta are the ones funding the complex. If they keep earning, the chain can still turn. If the long end keeps rising, this risk set moves higher.
SPY Loses 768; SOXX Snaps Back Under 532
SPY lost 768. That was a small support with a little more than a week of trade and light volume in between, so it does not count for much. Next is 756, then 745 to 756 — the prior breakout zone that was used to look for new highs, and an aggressive dip area. A break of 729 is what turns this into a downtrend and a larger pullback. On fundamentals, Q2 still points to upward revisions in 2026 S&P forward EPS, and Street 12-month index targets are more likely to be lifted than cut. The offset is yields. Treasury is unlikely to tolerate a sustained spike for long; similar expectation management showed up around March and April. Watch whether they step in. That is a risk, not an intervention that has already happened.
QQQ is still 705 to 724. Lose 705 and the tape goes weak, with a harder short-term bounce. Lose 693 and it is a breakdown, with 665 and below in view. Neither level is gone yet. This is just the map.
SOXX remains capped in 550 to 580. Until it clears, this is still a bounce. It dropped nearly 5% today and went back under 532, very fast. Buying a breakout into that resistance looks fine when it works and is a local top when it fails. The gap-down leaves no exit. The two lower levels have not changed. First question is whether 532 can put in a real stop-decline.
SMH is 557 to 581. After this rebound and sideways stretch, it is tracking SOXX again and the key levels rhyme. A break of 535 would turn it down, and that is still a little distance away. How to use these zones: if the fundamental damage is not as bad as the tape implies, a high-volume stop-decline is a reload or the next swing low. If the level fails after you buy it, the swing plan is broken — stop out.
META dropped again. A $1.4 trillion lawsuit will not be collected at face value; that would be a bankruptcy event the market will not get. Some blood, a settlement, and a lot of time is the more likely path. 540 is a stronger support and sits near the last decline low. Closer to 500, anyone still constructive on a six-month horizon has to decide whether the risk is worth taking. MSFT is still a light-volume range into 487. Today’s candle wants to look like a stop-decline, but the volume is thin and the pattern is weak. Wait. IGV barely moved, still around 101, with only a few components diverging.