Our forecasting track record

What changed, in plain words

• Chips and your gadgets: In his final earnings call as Apple's chief executive, Tim Cook described memory-chip pricing as a "hundred year flood" driven by AI data-centre demand. Apple said rising memory costs explain more than 100% of the fall in its profit margin — from 49.3% in March to 48.1% in June, guided down to about 46.5% in September — and that supply limits will cap how many iPhones, Macs and iPads it can make this quarter. Cook noted only three companies supply the world's DRAM memory, and said prices look set to keep rising beyond September. — The AI build-out is now bidding away the same memory chips that go into phones and laptops, and the cost is landing on the manufacturer.
• Korea: The KOSPI closed +17.91% at 6,595.45 — the largest one-day gain in the index's history — with SK Hynix +29.95% (its daily limit) and Samsung Electronics +26.81%, both records. That followed a fall of 10.84% on Tuesday and an unprecedented second consecutive trading halt on Wednesday, when the index touched 5,262.77. Across July the KOSPI still fell 22.19%, its worst month since the 1997 Asian financial crisis. — A crash and a record rebound inside four sessions. The round trip, not the rebound, is the information.
• AI spending: Amazon reported cloud revenue of $42.2bn against a $40.5bn forecast — its fastest growth in eighteen quarters — and raised its 2026 investment budget to about $220bn from $200bn. Its shares rose roughly 9–10% before the open. The day before, Microsoft added about $450bn of market value in a single session, the largest one-day gain by any stock on record. — The bear case that AI spending was about to crack did not fire this week. It accelerated.
• Apple's numbers: Revenue $109.4bn, up 16% and ahead of forecast; earnings of $2.02 a share beat $1.88. But its services arm ($30.7bn) and its China sales ($18.8bn) both fell short of expectations, and its guidance for this quarter — 9% to 11% growth, about $113bn — came in under the $114.9bn analysts expected. The shares fell about 7% after hours. — A strong quarter punished for a soft outlook — the same pattern as Samsung's record quarter earlier in the week.
• Growth and inflation: The US economy grew at an annual pace of 1.5% in the second quarter, down from 2.1% and below the 1.8% expected. The Fed's preferred inflation measure eased to 3.3% a year excluding food and energy, from 3.4% — still well above the 2% target. Slower growth alongside sticky inflation is the combination central banks find hardest to answer. — The economy is cooling while prices stay stubborn — which is why the Fed is split rather than confident.
• Oil: Brent crude traded near $90.25 a barrel, up 1.37%, heading for its biggest monthly rise since March. Shipping through the Strait of Hormuz is recovering after US–Iran strikes earlier in the week briefly pushed the price above $93. — The war premium is still in the oil price, but the physical flows have not broken.
• Trade policy: No new US tariff action this week. The Federal Register formalised the existing measure on Tuesday — additional duties of 10% or 12.5% covering 60 economies, in force since 24 July — and the grace period for goods already in transit has now closed. — The tariff regime is settled for now; nothing new was added.
• On the radar (monitored, never traded): Copper rose 2.49% to $6.44 a pound on Thursday and is up about 44% this year — the raw-material side of the same data-centre build-out. Bitcoin, at about $63,900, is roughly $52,000 below where it stood a year ago. — Money is flowing toward the physical inputs of AI and away from speculative assets — a rotation, not a panic.

Headline

The memory shock crossed a line yesterday. For six weeks the memory-chip story has been a producer story: Samsung and SK Hynix posting record results, their shares falling anyway, the Philadelphia Semiconductor Index entering a technical bear market on 17 July. Apple's fiscal Q3 call on 30 July made it a *consumer* story. Tim Cook — signing off from his last call as chief executive — called memory pricing a "hundred year flood" with exponential increases, and Apple disclosed that memory cost changes explain more than 100% of its sequential gross-margin decline: 49.3% (March, adjusted) → 48.1% (June) → ~46.5% guided for September. Supply, not demand, will cap September-quarter output of iPhone, Mac and iPad. Cook flagged the structural point plainly: the DRAM market has three primary suppliers, and Apple is "evaluating all options" for supply flexibility. That is the world's largest hardware buyer describing itself as a price-taker.

The financing leg of the AI-fragility thesis took a hard disconfirming blow the same night. Amazon posted AWS revenue of $42.2bn against a $40.54bn consensus — its fastest growth in eighteen quarters — said its AI and custom-silicon units each cleared a $25bn annual run rate, and raised 2026 capex guidance to ~$220bn from ~$200bn. Combined with Microsoft's 16% / ~$450bn single-session gain on 29 July (the largest one-day value gain by any stock on record) and Alphabet's earlier raise, the "hyperscalers are about to cut capex" hypothesis is now firmly on the back foot. The honest read: the AI trade's *equity* leg is fragile and violent; its *cash-flow and spending* leg is, so far, not.

Korea is the volatility story and deserves to be read as such. The KOSPI's +17.91% close at 6,595.45 is the largest single-day gain in the index's history. SK Hynix hit its 29.95% limit; Samsung rose 26.81%. But this arrives four sessions after a 10.84% single-day fall (28 July, close 6,023.66) and an unprecedented *second consecutive* circuit breaker on 29 July with the index at 5,262.77 intraday. July closed −22.19% — the worst month since the 1997 Asian financial crisis. Korean regulators have capped single-stock leveraged ETF holdings at 20% of an individual's portfolio, tripled cash-deposit requirements, and are now probing whether foreign high-frequency traders amplified the leveraged-ETF swings. The mechanism matters more than the level: this was a retail-leverage unwind and re-wind, not a repricing of memory fundamentals — Samsung's record quarter produced a flat share price on Thursday and a +27% one on Friday.

The systemic gauges did not move. `market_stress_gauge.py` reads calm on both components: the 2s10s slope at +0.45pp (30 July) with no re-steepening crossing, and high-yield spreads at 2.87pp (29 July), having widened just 0.12pp over 21 observations. VIX peaked at 20.66 on 29 July — the month's high, and still an ordinary number — and eased to roughly 18 by Thursday. This is the shape of a *localised* dislocation: extreme in semiconductors and in Korea, invisible in credit. That contrast is the day's genuine signal, and it is why the rollup below should be read as a sector flag rather than a systemic warning.

Macro adds a second, slower problem. Q2 GDP grew 1.5% annualised, down from 2.1% and under the 1.8% consensus. June core PCE eased to 3.3% y/y from a three-year high of 3.4%, with headline PCE at 3.7% helped by a 5.9% fall in energy goods and services (gasoline −9.2%) — a fall that Brent's move back to $90 is already unwinding. Cooling growth with inflation stuck near 3.3% is why the FOMC split 9–3 on 29 July, with Hammack, Kashkari and Logan dissenting *for a hike*. September pricing has repriced sharply toward a rise, though the precise probability could not be verified — see the fact-check.

Readings

Fact-check log

Standing theses — where the evidence moved

• #1 AI-credit fragility — DISCONFIRMED this week, materially. Amazon raised 2026 capex to ~$220bn from ~$200bn with AWS at its fastest growth in eighteen quarters; Microsoft's spending translated into 43% Azure growth and a record ~$450bn single-session value gain; Alphabet raised its guide on 22 July. The thesis's central mechanism — that the spending is credit-funded and about to crack — is not visible in this evidence. High-yield spreads at 2.87pp corroborate. The thesis is not dead (the funding *mix* is the untested part), but it has now been wrong-footed three quarters running and should be scored down. — The 'AI spending is about to collapse' case lost again this week.
  For your money This matters because a great deal of ordinary retirement money now sits in the handful of companies doing this spending. The evidence says the spending is being funded out of enormous operating cash flows rather than borrowing, which is the healthier of the two. The exposure is still concentrated, so the sensible framing is awareness of how much of what you own is the same few names — not a change of course.

  What to watch Matters more if the next round of results shows this spending being funded by new borrowing, or if cloud revenue growth slows while the budgets keep rising. Matters less while cash from operations comfortably covers it.
• #4 Compute as a strategic resource — CONFIRMED, and it now has a consumer price. Apple, the world's largest hardware buyer, publicly described itself as constrained by a three-supplier DRAM market and guided its gross margin down ~160 basis points on memory costs alone. This is the clearest evidence yet that compute inputs are being allocated by price to the highest bidder, and that the highest bidder is not the consumer-device industry. — Memory chips are now scarce enough that even Apple cannot buy its way to the front of the queue.
  For your money The practical consequence is that devices with a lot of storage get more expensive, or come with less storage at the same price, over the next few quarters. It is a slow, visible cost — not a shock. Nothing about it calls for a change in how savings are held; it is a reason to expect certain purchases to cost more than they used to.

  What to watch Matters more if other major device makers repeat the warning on their own results calls. Matters less if new memory capacity comes online or if the three suppliers start competing on price again.
• #7 Equity concentration / breadth — CONFIRMED, uncomfortably. A single stock (Microsoft) added ~$450bn in one session and carried three indices with it; Korea's entire market moved 18% on two chipmakers' shares. When index-level outcomes are decided by a handful of names, index-level calm can coexist with severe underlying instability. That is precisely the configuration on display this week. — The market's direction is being set by a very small number of companies.
  For your money Most ordinary index funds are concentrated in the same few names, so 'diversified' does less work than it used to. This is not a reason to sell anything, and there is no allocation prescription here — it is a reason to actually look at what your funds hold rather than assuming breadth you may not have.

  What to watch Matters more if index gains keep depending on one or two stocks while the average company goes nowhere. Matters less if participation broadens and equal-weighted measures keep pace with headline indices.
• #5 Sovereign/rates stress — WATCH. Japan is suspected of intervening again to defend the yen, with the dollar falling as much as 3.3% against it on Thursday. US 10-year yields sit near 4.65% with the FOMC split 9–3. Neither is a stress event; both are strain indicators worth keeping under observation. — Currency intervention and a split central bank are signs of strain in the system, not of breakage.
• #6 Energy & critical commodities — WATCH. Brent's move back to ~$90 unwinds the 9.2% petrol decline that flattered June's inflation reading. Copper at $6.44/lb is up ~44% this year on data-centre and grid demand, though that figure is single-source today. — Both the fuel and the metal are being pushed up, for different reasons.
• #3 Crypto → AI/compute rotation — CONFIRMED, continuing. Bitcoin sits ~45% below its level a year ago and did not participate in this week's equity rebound, while AI infrastructure absorbs capital at a $220bn annual run rate from one company alone. — Speculative money has moved from crypto to AI infrastructure, and has not come back.

Research only — not financial advice. The practice portfolio is for learning only; this is a read-only risk briefing meant to help explain risk, never tell anyone what to buy or sell.
Project snapshot — changes waiting for review: 8 · tracked forecasts open: 43 (checking now: 11) · track record so far: 0.137908 · practice portfolio updated: 2026-07-30.
Past briefs & dashboards: market-sentinel-lyart.vercel.app/reports