The S&P 500 was near a record high this week, but the top performer was the utilities ETF (XLU). Did money flee into defensives, and will the trend continue next week?
2026-W41
Most of the utilities gain came in a single day
About 75% of XLU's weekly gain of +3.97% came on 10-06 alone ($39.97 to $41.16, +2.98%). That day, Google announced a 20-year nuclear power deal for 890MW with Constellation Energy. Constellation rose more than 12%, followed by Vistra +11% and NRG +7% (Benzinga, Upstox, 2026-10). On the same day, the 10-year yield fell just 4bp, from 5.31% to 5.27%. What lifted utilities this week was AI power demand, not falling yields or haven buying.
Looking at 10-08, the day AI stocks collapsed, shows that the three defensive sectors rose on different days for different reasons.
| ETF | 10-06 (nuclear deal day) | 10-08 (AI selloff day) | Weekly |
|---|---|---|---|
| XLU | +2.98% | -0.19% | +3.97% |
| XLP | +0.94% | +2.11% | +3.60% |
| XLV | -0.17% | -0.39% | +2.79% |
| SMH | -0.22% | -2.84% | -4.32% |
- XLU traded as an AI capex beneficiary. It fell along with AI stocks on 10-08.
- XLP was the real haven. It earned more than half its weekly gain on 10-08 alone, when the 10-year yield also fell from 5.28% to 5.23%, easing rate pressure.
- XLV rose on 10-07 (+1.03%), when yields rose, and on 10-09 (+1.58%), ahead of UNH earnings. It actually fell on the nuclear-deal day and the AI selloff day.
The report grouped the three sectors as a defensive rotation, but in reality three separate flows overlapped in one week.
Utilities face two risks at once
In July, XLU already traded at about 23x earnings, well above its historical average of 17x, with a dividend yield of 2.6%. The same article judged that the 10-year yield of 4.54% at the time also limited multiple expansion (247 Wall St., 2026-07-19). The 10-year yield is now 5.24%. The only reason to buy an asset whose dividend yield is about 2.6 percentage points below Treasuries is growth expectations. Those expectations come from AI power demand such as AEP's 24GW of new load commitments through 2030 and NextEra's roughly 30GW of data-center-related orders (investingLive, 2026-09).
So XLU is now an asset that is rate-sensitive like a bond and also sensitive to AI sentiment like a growth stock. If yields rise or the debate over a slowdown in AI investment grows, it gets hit either way. It came first this week only because neither went badly.
How did defensive-led record highs end in the past?
Past cases do not point in one direction. LPL summarized that large-cap-led bull markets often continued even when breadth was narrow, and that once the concentration cooled, leadership often broadened into value and small caps (LPL Research). Sentimentrader analyzed the 2021 record highs led by utilities and health care and concluded that defensive leadership is not bad news in itself, but that if breadth worsens, the risk of weaker short- and medium-term returns rises (Sentimentrader, 2021-09-02).
"If breadth doesn't confirm the S&P and Nasdaq breakouts, we'd be wary of a pullback." — Dan Wantrobski, Janney (Bloomberg, 2025)
This week the Russell 2000 returned -0.91%, more than 2 percentage points below the S&P 500 (+1.15%). These were record highs that breadth did not confirm.
Conditions for continuation next week
- If XLP and XLV keep leading: AI large-cap selling continues and yields do not rise sharply. That would be a genuine defensive phase.
- If only XLU keeps leading: AI sentiment recovers and yields fall. That is not defense but another face of the AI bull market.
- If all three turn lower: Yields break above the weekly high of 5.36% again. Defensive sectors do not defend against rates either.
What investors should do
- Buying XLU to offset the risk of a large semiconductor or AI large-cap position does not fit. The driver of the 10-06 gain was AI capex, so it wobbles along with AI in a correction. For hedging, XLP, which actually held up on 10-08, is the better fit.
- If you already hold XLU, set a recheck line at a renewed break of 5.36% on the 10-year yield. If either rates or AI sentiment worsens, the high multiple becomes a burden.
- Judge whether the defensive concentration is hardening by whether the Russell 2000 gets back ahead of the S&P 500, not by sector rankings. If small caps follow, the concentration is dissolving. Do not shift weights much on one week of data.