Brent crude crashed 25% in eight days, from $130 to $97, after Iran offered a seven-day deal to reopen the Strait of Hormuz, but tanker war-risk insurance still sits at 40× pre-crisis levels, meaning the physical market hasn't believed the diplomacy yet.
▸
The 10-year Treasury yield hit 5.18%, its highest since 2007, driven not by Fed hike fears but by a structural term premium: Norway's $2.3T sovereign wealth fund is selling $80B of Treasuries, and hyperscalers are issuing billions in debt to fund $586B of AI capex.
▸
Crypto grinded higher without leverage, Bitcoin +42%, Ether +73%, Solana +74% over three months, on $2.4B of spot ETF inflows in five days while perpetual funding stayed flat and open interest fell, a rare "spot-led" rally that suggests institutional accumulation, not speculative froth.
The week that was
Oil's violent round-trip defined the commodity narrative. Brent surged to $130.80 on September 17 after U.S. strikes on Iranian ships in the Strait of Hormuz, then gave back the entire spike, and more, falling to $97.44 by Friday as Iran proposed a diplomatic off-ramp. The tell is in the plumbing: VLCC tanker earnings still print $1.1M per day, and war-risk premiums for Hormuz transits remain at 1-5% of hull value, up from 0.125% pre-crisis.
Over 1 billion barrels transited the Strait in the past year per CENTCOM, yet IMF PortWatch recorded just one commercial transit on September 20 versus a baseline of 85 per day. The market is pricing a diplomatic possibility, not a cleared channel, and until insurance normalizes, the risk premium stays embedded in gasoline and diesel cracks.
That same caution in physical markets echoes in bonds. The 10-year climbed to 5.18% while the 2-year held at 4.87%, leaving a 31-basis-point spread that signals "term premium," not rate-hike expectations. The 10-year breakeven inflation rate sits at just 2.33%, meaning the backup is almost entirely real yield, the highest since 2007. Two structural sellers are absorbing duration: Norway's sovereign wealth fund confirmed its plan to cut Treasury holdings by $80B by early 2027, and the five largest hyperscalers (Microsoft, Google, Amazon, Meta, Oracle) issued heavy debt to fund $586B of trailing-twelve-month capex, up 84% year-over-year.
That capex now consumes 83% of their operating cash flow, leaving aggregate free cash flow down 38% to $121B. Amazon alone issued over $72B in 2026 bonds and guided FY26 capex to roughly $220B. The Fed's policy rate at 3.75-4.00% is effectively decoupled from the long end; the bond market is tightening financial conditions on its own.
Equities split along that fault line. The Nasdaq-100 printed a fresh 1-year high at 30,608 (+3.2% on the week) and the S&P 500 rose 1.2% to 7,743, led by mega-cap tech: Microsoft +4.5%, AMD +12.7%, Meta +12.9% for the week. But the Russell 2000 fell 0.8%, utilities dropped 3.9%, and regional banks lagged, the classic rate-sensitive, low-growth cohort fracturing under 5%+ long yields.
Breadth remains thin: only about 30% of S&P 500 stocks trade above their 50-day average. The Shiller CAPE at 41.5 sits above 99% of history, but the market is pricing an AI productivity narrative that discounts future earnings at a lower effective rate than the 10-year yield implies.
Crypto's move was the quietest surprise. Bitcoin at $84,765 (+4.4% on the week, +42% in three months) cleared its six-month high of $82,139 on rising spot ETF flows, $2.4B net in five days through September 25, while Binance perpetual funding stayed near zero (+0.0028%/8h, roughly 3% annualized) and open interest fell 8.5% in dollar terms since late August.
Ether at $2,713 (+73% in three months) and Solana at $124 (+74%) showed the same pattern: spot accumulation, not leverage. Stablecoin supply grew just 0.8% in 30 days to $312B, confirming this isn't a stablecoin-fueled credit expansion. Fear & Greed sits at 70 ("Greed"), unchanged from a week ago.
The dollar held its ground, reflecting the "dollar smile" in full effect: the U.S. offers both higher growth and the world's reserve currency, so capital flows in whether the outlook is boom or panic. DXY at 101.0 (+0.8% on the week) and the broad trade-weighted index at 119.5; USD/JPY at 157.2 erased the early-September yen rally, and EUR/USD at 1.14 reflects a 150-basis-point policy differential versus the Fed's 4% upper bound. Commodity currencies (AUD 0.703, CAD 1.414) remain pressured by the yield gap.
The big question this week
Can hyperscaler AI capex stay at $145B+ per quarter without breaking the companies' balance sheets, and who actually captures the spend?
As detailed above, the five largest U.S. hyperscalers spent a combined $586 billion on capex in the trailing twelve months to June-August 2026, up 84% year-over-year. That equals 31.6% of their $1.85 trillion in revenue and 82.9% of operating cash flow, leaving just $121B of free cash flow, down 38% from $196B a year ago. The spending is almost entirely for AI infrastructure: GPUs (Nvidia), high-bandwidth memory (SK Hynix, Samsung, Micron), custom ASICs (Broadcom, Marvell), foundry capacity (TSMC), lithography (ASML), and power/cooling (Vertiv, Schneider).
Because capex growth (84% YoY) far outpaces operating cash flow growth (roughly 20-30%), the gap is being filled by debt issuance and cash drawdowns. Amazon's long-term debt doubled to roughly $129B with its 2026 issuance; Microsoft uses long-duration debt alongside operating cash flow; Oracle's capex exceeded operating cash flow by 62%, pushing FCF to -$29B. The mechanism is a classic J-curve: massive upfront investment in long-lived assets before AI revenue scales to cover the cost of capital.
For markets, the megacap basket's forward P/E of 22.2 embeds an assumption that capex peaks soon and FCF inflects positive. If the roughly $145B quarterly run-rate persists, FCF yield compresses further and duration risk rises with 10-year yields at 5.18%. For a normal person's money: the AI build-out is pulling forward years of semiconductor, memory, and power-equipment earnings, benefiting NVDA, MU, AVGO, TSMC, ASML, and VRT, but it also means the largest S&P 500 components are self-funding less and issuing more debt, increasing systemic leverage.
What would change the view: Any of MSFT, GOOGL, AMZN, or META guiding next-quarter capex below roughly $145B combined; Nvidia data-center revenue growth dropping sustainably below 50% YoY (currently +128%); 10-year yields breaking above 5.5%; or hyperscaler aggregate debt-to-EBITDA rising above 2.5× (currently roughly 1.8×).
The week ahead
Tuesday, September 29, RBA rate decision (00:30 ET) + Australian CPI (21:30 ET). The Reserve Bank of Australia is forecast to hike 25bp to 4.60% after trimmed-mean CPI ran at 0.5% m/m in August. A surprise hold would drop AUD/USD (currently 0.703) and signal the global tightening cycle is stalling. Why it matters: Australia is a bellwether for China-linked commodity demand; its rate path feeds directly into Asian FX and resource equities.
Wednesday, September 30, U.S. Core PCE m/m (08:30 ET, forecast 0.3% vs 0.2% prior) + Final Q2 GDP (08:30 ET, forecast 1.5%). Core PCE is the Fed's preferred inflation gauge; a 0.3% print annualizes to 3.6%, well above the 2% target and consistent with the September SEP median of 3.4% for 2026. GDP at 1.5% would confirm the "solid but not booming" growth the FOMC described. Why it matters: A hot PCE print locks in the "higher for longer" narrative, keeping mortgage rates near 7% and pressuring rate-sensitive sectors.
Thursday, October 1, Nike earnings (after hours, EPS est. $0.44) + Accenture earnings (pre-market, EPS est. $3.19). Nike is a consumer-discretionary canary; its guidance on China demand and inventory will signal whether the global soft-landing story holds. Accenture's consulting bookings are a real-time read on enterprise IT spend, including AI implementation budgets.
Friday, October 2, U.S. Jobs Report (08:30 ET). Non-farm payrolls forecast at 98K (vs 162K prior), unemployment steady at 4.1%, average hourly earnings +0.3% m/m. The Fed has explicitly tied further hikes to labor-market resilience; a sub-100K print would be the first clear crack in the "strong jobs" pillar. Why it matters: Your mortgage rate, car loan, and savings yield all hinge on whether the Fed sees enough labor softness to pause.
Worth knowing
Your brain is the biggest risk
Resolving Friday's prediction: The BOE's Governor Bailey spoke on September 25, and the pound barely moved. GBP/USD closed at 1.325, up just 0.3% on the day and down 1% on the week. The mechanism: Bailey reiterated that policy will remain "restrictive for sufficiently long," which was already priced into the 150bp rate differential versus the Fed. When a central banker says exactly what markets expect, the currency doesn't budge, the move happens on surprises, not confirmations.
Now the lesson. You lose more money to your own psychology than to bad analysis. Four biases do the heavy lifting:
Loss aversion, the pain of a $1,000 loss feels roughly 2.5× stronger than the pleasure of a $1,000 gain. So you hold losing positions too long (hoping to "get back to even") and sell winners too early (locking in the good feeling). Remember the yield-price seesaw from our bond lesson: when yields rise, bond prices fall. Loss aversion makes you freeze instead of rebalancing.
Recency bias, you overweight what just happened. After three months of crypto grinding higher (BTC +42%, ETH +73%), it feels inevitable. After three days of oil crashing 25%, it feels permanent. Neither is true; both are just the latest page in a longer distribution.
Herding, you feel safer when everyone agrees. The Nasdaq-100 at a fresh high with 30% breadth feels safe because "tech is winning." But the crowd is often most confident at the turn, see the VIX at 14.87, near its low for the year.
Overconfidence, you mistake a bull market for brains. The hyperscaler capex story ($586B, 83% of cash flow) looks brilliant if AI revenue scales on schedule. But the same managers who built the cloud are now leveraging their balance sheets to bet the company on a technology whose monetization timeline is uncertain. That's not genius; it's a concentrated bet.
Simple guardrails: (1) Write down your exit rule before you enter, "I sell if the thesis breaks, not if the price drops." (2) Rebalance on a calendar, not a feeling, quarterly forces you to sell winners and buy losers, the opposite of what your instincts want. (3) Size positions so no single idea can wreck your year, if you're right, you still win; if you're wrong, you survive.
Concept 71 of 83 in the Fair Value course.
Tomorrow's setup: The RBA decides on rates Tuesday (forecast 4.60% vs 4.35% prior). Which asset is most exposed to a surprise, AUD/USD, Australian 3-year yields, or ASX 200 financials, which way does it move on a hawkish shock, and through what mechanism? We'll have the answer in tomorrow's edition.
Not financial advice. This brief is for informational and educational purposes only and does not constitute investment advice or a recommendation to buy or sell any security.
Data: macro indicators per FRED® (Federal Reserve Bank of St. Louis); not endorsed or certified by the Federal Reserve Bank of St. Louis. Energy data per the U.S. Energy Information Administration (EIA). Auction data per U.S. Treasury Fiscal Data. Filings per SEC EDGAR. Market prices per Yahoo Finance. Earnings calendar per Financial Modeling Prep.
