Our forecasting track record

What changed, in plain words

• Chips: Korea's KOSPI index opened up 1.5% and rose more than 2% to 6,380, then reversed within half an hour and was down 1.66% at 6,153.55 by 09:50 local time. Samsung Electronics fell 2.71% and SK hynix 3.00% — even though US chipmakers had risen the night before (Nvidia +2.93%). The new piece of information behind the nerves is Chinese: ChangXin Memory (CXMT) listed in Shanghai on 27 July, closing its debut up about 470%, and now holds 8% of global memory-chip revenue against 3% a year earlier. But the price war everyone fears has not actually shown up yet — CXMT's own selling prices run only 5–10% below the three incumbents, and Reuters reported on 24 July that it priced a large server memory module *above* the comparable Samsung part while refusing Huawei's request for a discount. — The fear is that cheap Chinese memory chips wreck the industry's prices. So far the fear has moved the share prices; the actual chip prices have not moved with it.
• Shares: Wall Street rallied on Monday as the Middle East risk eased and oil fell. The Nasdaq Composite rose 2.13%, the S&P 500 1.48%, and the Dow Jones average 1.32% — up 693 points to a record close of 53,178. Amazon rose about 4.6% and passed a 3 trillion dollar market value for the first time, the fifth company ever to do so, after cloud revenue of 42.2 billion dollars beat a roughly 40.5 billion dollar forecast. Microsoft (+4.9%), Alphabet (+4.9%), Meta (+6.0%) and Boeing (+7.9%) also rose sharply. — One of the strongest days of the summer, and it came almost entirely from a small number of the very largest companies.
• AI spending: Palantir reported quarterly revenue of 1.935 billion dollars, up 93% on a year earlier, and raised its full-year revenue guidance to 8.15–8.16 billion dollars from 7.65–7.66 billion. Its US commercial business grew 149%. The stock rose about 10% after hours and around 16% in pre-market trading. Taken with Amazon's cloud numbers, this is a second direct answer to the doubt that drove July's selloff — that companies were spending on artificial intelligence faster than they could earn from it. — The two clearest reports this week both say AI demand is still translating into paid revenue, not just spending.
• SpaceX: SpaceX posts its first quarterly results as a public company after tonight's close (webcast 4:30pm New York time), and two days later, on 6 August, its share lock-up starts expiring. Insiders may sell up to 20% of restricted holdings — as much as 911.5 million shares, more than the entire amount currently free to trade. Elon Musk and a selected group stay locked until mid-2027. The company lost 4.28 billion dollars in the first quarter on 4.69 billion dollars of revenue; the shares closed Monday at 114.53, up 5.68%. — The biggest test yet of whether investors will keep funding loss-making AI and space infrastructure at these prices — and it has a date on it.
• Oil: Brent crude fell about 5% to below 84 dollars on Monday after the US called off the strike, then edged back to 84.54 (+0.92%) this morning; US crude was 80.93. Iran denies direct talks with Washington but says Oman-brokered discussions on reopening shipping through the Strait of Hormuz are progressing. Turkey and Iraq extended a pipeline agreement by a year, Kazakhstan resumed pipeline exports, and OPEC+ approved another modest output increase, completing the restoration of the production cuts it made in 2023. Even so, Brent is still 17.4% higher than a month ago. — Cheaper than last week, still much dearer than last month. The month-on-month rise is what eventually reaches the pump.
• Interest rates: The debate is still about a rise, not a cut, after the Fed held at 3.50–3.75% on 29 July with three officials dissenting in favour of an increase. What is striking today is how far the betting venues disagree about the September meeting: interest-rate futures imply roughly 32%, Polymarket about 53%, Kalshi about 57%, and a late-July reading of the CME's FedWatch tool was near 82%. The US 10-year government bond yield was 4.70%. — Nobody has a confident number for what the Fed does next month, and the spread between venues is wide enough that any single quoted figure should be treated with caution.
• Jobs: Weekly claims for unemployment benefit came in at 197,000 for the week to 25 July, below the 200,000 forecast, and the four-week average of 202,750 is 8.9% *better* than the four weeks before it. Unemployment was 4.2% in June. The monthly job-openings survey is published this morning and the July employment report on Friday 7 August. — The job market is the one big gauge that has been quietly improving rather than deteriorating.
• The yen: The Japanese yen has resumed weakening within days of last week's rare joint US–Japan intervention to support it, with the dollar back at 157.81 yen (+0.40%). The dollar index was just under 100 and gold was flat at about 4,055 dollars an ounce; silver rose 2.0% and copper 1.5%. — Two governments spent money to prop the yen up last week and it is sliding again — a sign the pressure on it is structural, not a one-off scare.

Headline — a de-escalation rally led by six companies, and a memory-chip fear that the price data does not yet corroborate

Monday's session was a clean risk-appetite reversal. With the US strike on Iran called off, Brent fell roughly 5% to below 84 dollars and the whole inflation-and-rates chain relaxed with it: Treasuries rebounded, yields eased across the curve, and the equity indices closed higher — Nasdaq Composite +2.13%, S&P 500 +1.48%, Dow +1.32% (+693 points) to a record 53,178 (EVT-0294). Note the record claim carefully: Trading Economics describes the S&P as *nearing* its record rather than setting one, while a lower-provenance outlet asserted a first-ever close above 7,600. We report the level and the percentage and decline to assert the record for the S&P (see fact-check).

The composition matters more than the move. Amazon (+4.6%) crossed 3 trillion dollars in market value for the first time on the back of 42.2 billion dollars of quarterly cloud revenue against a ~40.5 billion dollar consensus (EVT-0295); Microsoft, Alphabet and Meta each added 4.9–6.0%. Palantir then raised full-year revenue guidance by roughly 500 million dollars on 93% growth, with US commercial revenue up 149% (EVT-0296). Read as a set with last week's Amazon and Microsoft capex guides, the July "AI capex has outrun AI revenue" thesis has now taken three direct hits in a week — the demand leg of the AI-buildout chain is confirming, not cracking.

The supply leg is where today's genuinely new information sits. CXMT's Shanghai debut on 27 July (+470% on the day, briefly the mainland's most valuable company) triggered a 13–14% one-session fall in Samsung and SK hynix, and the Korean tape is still unstable: this morning the KOSPI gave up a +2% open to trade −1.66%, Samsung −2.71%, SK hynix −3.00% (EVT-0297), and Seoul brokers cut more price targets than they raised for the first time in fifteen months (EVT-0306). Counterpoint puts CXMT at 8% of Q1 global DRAM revenue versus 3% a year earlier, fourth behind Samsung 38%, SK hynix 29% and Micron 22%, with planned capacity additions of 100,000 wafers per month in 2027 and again in 2028. And yet the mechanism the fear depends on — CXMT dumping cheap supply — is not visible in prices: SemiAnalysis measures CXMT's average selling prices at only 5–10% below the incumbents, and Reuters reported on 24 July that CXMT priced 64GB DDR5 server modules *above* the comparable Samsung product and rejected Huawei's discount request (EVT-0298). The estimated technology gap is three to four years, Samsung and SK hynix are sampling HBM4E while CXMT is reportedly still stabilising 8-high HBM3 at sub-standard yields, and CXMT does not appear in Counterpoint's Q1 high-bandwidth-memory rankings at all. The honest statement is therefore narrow: a credible new entrant has emerged in commodity memory, capacity is coming, and the market has repriced the *risk* of a glut — but the glut itself has not yet appeared in contract prices. That distinction is now a logged, dated prediction (P-0064) rather than an assertion.

The financing leg carries the week's discrete event. SpaceX reports Q2 after the close tonight — its first as a listed company, confirmed by its own investor release — and on 6 August up to 911.5 million insider shares become sellable, more than the current free float, with a further 455.8 million possible under certain conditions (EVT-0299). The reported dollar value varies between about 116 and 123 billion depending on the price used, which is itself a reason to quote the share count instead. Against a first-quarter loss of 4.28 billion dollars on 4.69 billion dollars of revenue, this is the cleanest scheduled test available of whether the market will keep funding loss-making AI-and-infrastructure capacity at current valuations; logged as P-0065. Caterpillar, a builders-and-equipment proxy on our radar, beat at 8.17 dollars per share, and Super Micro reports today (EVT-0307).

Underneath all of it the systemic gauges are quiet. The deterministic credit-and-curve gauge read calm: the 2s10s Treasury slope was +0.45 percentage points on 3 August with no re-steepening event, and high-yield spreads were 2.84 percentage points on 30 July, well under the 4.00 watch threshold, with 21-day movement of only +0.10 (EVT-0301). The labour gauge also read calm — Sahm-rule gap 0.067 percentage points, claims 197,000, four-week average 8.9% better than the prior four weeks (EVT-0302). Bitcoin at 63,694 and Ether at 1,864 stayed flat while equities made records, which continues to corroborate the capital-rotation thesis rather than a common risk-appetite move (EVT-0304). Candour on our own instrument: the headline global gauge below reads Watch for what is now roughly its thirty-fifth consecutive session, driven by localised energy, chip and currency flags rather than any systemic breach — the Reviewer has formally flagged this as alert fatigue (proposal PR-0031, pending human decision), and readers should treat today's Watch as a sector description, not a systemic warning.

Readings

Fact-check log

Standing theses — where the evidence moved

• #1 AI-credit fragility — mixed, and squarely in play. Confirming: SpaceX carries a 4.28 billion dollar quarterly loss on 4.69 billion dollars of revenue into its first public earnings report tonight, with 911.5 million insider shares — more than the entire current free float — becoming sellable on 6 August (EVT-0299). Disconfirming: high-yield spreads are 2.84 percentage points with no widening flag (EVT-0301), and the demand side keeps printing paid revenue rather than promises (Amazon EVT-0295, Palantir EVT-0296). The financing channel remains the thesis's live test, and it now has a date. — The worry is that AI infrastructure is funded on optimism. Credit markets say no stress; the SpaceX unlock is the nearest thing to a scheduled stress test.
  For your money If funding for loss-making infrastructure tightened, the companies that have driven index gains would be hit first — which reaches ordinary savers through index funds and pension holdings rather than through anything they bought deliberately. This isn't a reason to make a big move; keep your money spread across different kinds of assets and avoid borrowing against a single story.

  What to watch Matters more if insiders sell heavily into the unlock and the shares fall sharply, or if borrowing costs for lower-quality companies start rising. Matters less if the unlock is absorbed quietly and credit markets stay open on current terms.
• #3 Crypto-to-compute capital rotation — confirming again. Bitcoin at 63,694 and Ether at 1,864.76 were roughly flat while US equities closed at or near records (EVT-0304). A genuine broad rush into risk would normally lift both; the persistent divergence is more consistent with money rotating out of crypto and into the AI complex. — Crypto keeps sitting out equity rallies, which looks like money moving between assets rather than new money arriving.
  For your money It means the strength in shares is not a general wave of enthusiasm lifting everything, so a portfolio holding both may not be as spread out as it looks — the same underlying appetite is being reshuffled, not multiplied. This isn't a reason to make a big move; it is a reason to check whether your holdings genuinely differ from one another.

  What to watch Matters more if crypto keeps falling while equity indices set highs. Matters less if the two start moving up together again, which would suggest fresh money rather than rotation.
• #4 Compute as a strategic resource — materially reframed by CXMT. For the first time the memory oligopoly's pricing power has a credible state-backed challenger: 8% of Q1 global memory revenue against 3% a year earlier, fourth behind Samsung 38%, SK hynix 29% and Micron 22%, with 100,000 wafers per month of new Shanghai capacity planned for 2027 and again for 2028 (EVT-0298). But the technology gap is estimated at three to four years, CXMT is absent from Counterpoint's Q1 high-bandwidth-memory rankings entirely, SK hynix rather than CXMT won Xiaomi's next-generation low-power memory business, and CXMT's prices sit at or above the incumbents'. Verdict: a real long-run structural threat in commodity memory; not yet a pricing event. Escalated as P-0064. — China now has a serious memory-chip company. It is not yet undercutting anyone on price, and it is years behind on the advanced chips that matter for AI.
• #6 Energy & critical commodities — two-sided. Disconfirming the supply-shock case: the immediate war risk came out of the price (Brent −5% Monday), OPEC+ completed restoring its 2023 cuts, Turkey–Iraq pipeline capacity was extended and Kazakh exports resumed (EVT-0300). Confirming: Brent is still 17.4% above a month ago, copper rose 1.5% and silver 2.0%, and Hormuz shipping is not confirmed restored. — Supply news improved this week; the month-long price rise has not been undone.
• #7 Equity concentration & breadth — confirming. The Dow's record close was built on Boeing +7.9%, Microsoft +4.9% and Amazon +4.6%; the Nasdaq's +2.13% came from the same handful of names, with Amazon becoming the fifth company ever worth 3 trillion dollars (EVT-0294, EVT-0295). Index highs achieved by a narrowing set of leaders is exactly the pattern this thesis tracks. Note the honest limit: we have no same-day participation measurement, only the composition of the largest movers — the weekly discovery screen carries the breadth reading and it last read calm. — Records driven by six companies tell you less about the wider economy than records driven by six hundred.
  For your money Concentration means an ordinary index fund is quietly more exposed to a few technology companies than most savers realise, so a setback in that group would show up in savings that were never deliberately pointed at it. This isn't a reason to make a big move; it is a reason to look at what your funds actually hold and keep money spread across genuinely different assets.

  What to watch Matters more if new index highs keep coming while most listed companies fail to participate. Matters less if gains broaden across industries and equally-weighted measures keep pace with the headline index.
• #9 Labour-market deterioration — disconfirming. The deterministic gauge reads calm on both components: Sahm-rule gap 0.067 percentage points against a 0.30 watch threshold, and the four-week claims average of 202,750 is 8.9% *better* than the prior four weeks (EVT-0302). Friday's July employment report is the next real test; logged as P-0068. — The recession early-warning measures are moving the right way. This is the clearest good news in today's brief.
  For your money A steady job market is the single biggest protection most households have, because income security matters more to savings than any market move. This isn't a reason to make a big move, but it is the part of the picture that supports keeping to a plan rather than reacting to headlines.

  What to watch Matters more if Friday's report shows unemployment rising off its low or hiring turning negative. Matters less if unemployment holds and weekly claims keep falling.
• #2 parallel payment rails, #5 sovereign and government-debt stress, #8 the consulting and IT-services value chain — no material new evidence today. Adjacent to #5, Japan's 10-year government bond yield at 2.86% alongside a yen that is weakening again despite official intervention (EVT-0303) is worth continued monitoring, but nothing published today moves the score. — Three of the nine standing theses had no news today. Recording that is part of the discipline.

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: 11 · tracked forecasts open: 55 (checking now: 13) · track record so far: 0.137908 · practice portfolio updated: 2026-07-31.
Past briefs & dashboards: market-sentinel-lyart.vercel.app/reports