Today, Friday September 11, the August CPI print lands at 8:30 a.m. ET — the last major inflation data point before the Fed votes on September 16. And it lands on a market that is more concentrated than at any point in the modern record, more leveraged into a single trade (AI infrastructure), and more sensitive to the cost of money than at any time since the dot-com peak. This post puts real numbers on all three facts.

“Today, Friday September 11, the August CPI print lands at 8:30 a.m.”

The Odds Have Tripled in Five Weeks

Rate-hike probability for the September 16 FOMC meeting has climbed almost relentlessly since early August: roughly 44% on August 7, 58% after August's blowout jobs report (+162k payrolls versus ~53k consensus), about 61% after Fed Chair Kevin Warsh's hawkish Jackson Hole speech — where he first said inflation must return to 2% "at sufficient speed" — and around 70% after Thursday's August PPI report plus an oil spike above $100 a barrel (Brent ~$107) on Strait of Hormuz and Red Sea disruptions.

CME FedWatch, September 2026, compiled from public reporting. The consensus CPI expectation is +3.4% year-over-year headline, +0.2% month-over-month core; the Cleveland Fed nowcast sits at 3.38%. Scenario analysis circulating on the Street: a print at 3.5% or hotter pushes hike odds to ~85%+; below 3.2% collapses them toward 10–25%.

The macro backdrop is the worst of both worlds for the AI trade. The 10-year Treasury yield has pushed through 4.9%, the 30-year is at its highest in more than 19 years, and one-year inflation swaps have jumped nearly 100 basis points since the start of August to ~2.70%. AI infrastructure capex — Goldman Sachs pegs global AI-related investment at ~$1 trillion in 2026 alone — is increasingly financed with corporate debt. A hiking cycle doesn't just raise the discount rate in every DCF; it directly raises the cost of building the data centers the whole thesis rests on.

The Week in Sector Data: Everything Is Down Except Two Things

The ChartRow large-cap universe (500 US names, $83.0T combined market cap, close of September 10) tells the story of the week in one line: breadth is broken. Only 151 of 500 stocks are up over five sessions; only 160 were green on Thursday. The median stock fell -1.5% on the week.

Only two sectors finished the week green: Energy (+0.4% median), riding the oil shock, and Technology (+1.4% median) — and tech's resilience is worth unpacking, because it is narrow. Healthcare (-4.5%), Consumer Discretionary (-3.7%) and Materials (-3.3%) — the sectors most exposed to consumer demand and commodity costs — led the damage. Copper and oilfield names were Thursday's worst performers (SCCO -7.2%, BKR -6.7%, FCX -6.6%) even as the index-level damage stayed modest.

Zoom out to year-to-date and the picture inverts in an uncomfortable way: the sector that is supposed to be the inflation hedge for the last four years has quietly become 2026's best performer, while the consumer — the economy's engine — is the year's loser.

ChartRow data, medians of 500 US large caps, close 2026-09-10. Energy's +43.7% median YTD is the sound of a supply-shock regime. Consumer Discretionary's -6.2% median is the sound of a consumer that stopped keeping up with inflation three years ago. Communication Services is red YTD too (-2.2% median) despite hosting two of the seven most valuable companies on earth — a reminder that "owning AI winners" and "owning tech" were never the same trade.

$29.5 Trillion in Ten Names: The Concentration Math

Here is the number that should frame every portfolio conversation this fall: the ten largest US companies — Nvidia ($5.26T), Apple ($4.77T), Alphabet ($4.07T), Microsoft ($3.66T), Amazon ($2.72T), TSMC ($2.22T), SpaceX ($2.02T), Broadcom ($1.72T), Meta ($1.64T), Tesla ($1.44T) — now account for 35.5% of the entire US large-cap market ($29.5T of $83.0T) on a full market-cap basis.

On the S&P 500's float-weighted basis the top 10 sit near 38–40% of the index, per RBC and Capital Group research — versus ~26% at the 2000 dot-com peak and ~30% in the 1963 Nifty Fifty era, per Morgan Stanley's 145-year study. The French data library version is starker: the largest decile of US stocks hit 78.5% of total US market value in November 2025, surpassing the previous record of 73.8% set in June 1932.

The uncomfortable comparison is not valuation — today's leaders trade around 34x versus ~43x in 2000, and unlike the 2000 cohort they print enormous free cash flow. The uncomfortable comparison is weight. RBC notes the top 10 carried ~41% of index weight at end-2025 but only ~32% of expected earnings, and the cap-weighted S&P has beaten the equal-weighted S&P by ~32% over the last three years — the largest three-year spread on record, exceeding even the late-1990s.

What the History Books Say About Concentration Extremes

GW&K Investment Management's century-long study offers the soberest framing available: after every concentration peak, the pendulum swung back — but never on schedule.

  • After the 1932 peak: small caps beat large caps by ~49 points in the following year, ~15 points annually over five years.
  • After the Nifty Fifty peak (December 1972): small caps compounded 6.5 points/year ahead of large caps for a decade.
  • After the dot-com peak (March 2000): a ~6%/year size premium and ~17%/year value premium as leadership broadened into the 2000–2006 small-cap rally.

There are early signs of broadening in 2026: small caps were up 22.6% YTD through June versus ~10% for the S&P 500, helped by tax incentives for capital spending. But the last five weeks have been a reminder that the regime can re-tighten fast: when the market's question collapses to "will the Fed hike Wednesday?", flows retreat to the megacaps' balance sheets, and the top-heavy index decides the market's direction for everyone.

Three Takeaways for AI-Era Operators

  • Today's CPI print is a genuine regime marker, not noise. Core inflation staying at 0.2% m/m keeps the hold case alive and probably relieves the rate overhang into October. A hot core (0.3%+) means the first hike since 2023 lands Wednesday — and the AI infrastructure names with the highest multiple expansion have the most discount-rate sensitivity.
  • Watch the debt, not just the capex headlines. Goldman's $1T-of-AI-investment number is increasingly funded in bond markets. Rising yields are a direct tax on the buildout — the second-order effect almost nobody put in their 2026 AI models in January.
  • Concentration is a portfolio decision, not an index default. If you own an S&P 500 index fund, you own a ~40% bet on ten companies, most of them levered to a single technology thesis. History says that bet eventually mean-reverts in relative terms; it also says timing the reversion cost more money than holding the concentration through three previous peaks. Equal-weight and small-cap sleeves are hedged bets on the pendulum — expensive for now, cheap in a decade.

The September 16 decision is three trading days away. Whatever CPI says this morning, the structural story is already clear: the market is concentrated at century-extreme levels, the AI buildout is debt-financed at scale, and the cost of money is moving the wrong way. Position accordingly.

---

Data: ChartRow 500-stock US large-cap universe (close 2026-09-10); CME FedWatch; Cleveland Fed inflation nowcast; GW&K Investment Management, RBC Wealth Management, Capital Group and Morgan Stanley concentration studies. Charts generated from live screener data via Plotly. This post is for information only and is not investment advice.