Skip to content
Sign in
The Markets Desk — Week 41 · Thu 8 Oct 2026
Beats / 99 priced constituents

hardware

An equal-weight index of the priced names on this beat, rebuilt every Friday from the constituents that were actually in it.

326.6+4.5%base 100 · to 2026-10-02

This week

What happened

Events on hardware
DateTypeWhat happenedSource
2026-10-05editorial noteMargin figure revised: 52% → 53.7%Margin moved from 52% (deep-dive-2026-07-16.md) to 53.7% (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteRevenue figure revised: $1.27 → €2,761.7MRevenue moved from $1.27 (deep-dive-2026-07-16.md) to €2,761.7M (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteVerdict changed: Q2 beat + a ~€7.5B FY26 guidance raise since the 20-F (now €43–45B net sales, 54–56% GM) confirms the June dossier's "AI demand broadening, not peaking" read — but the Q2 beat's engi…Before (deep-dive-2026-07-16.md): Q2 beat + a ~€7.5B FY26 guidance raise since the 20-F (now €43–45B net sales, 54–56% GM) confirms the June dossier's "AI demand broadening, not peaking" read — but the Q2 beat's engine was Installed-Base/service upside (+€300M mgmt-cited), not new systems, so it confirms today's revenue durability more than it confirms the forward EUV order book. Thesis direction unchanged (BULLISH structurally / WATCHING on entry); nothing here moves conviction off the June read. After (deep-dive-2026-10-05.md): The earnings base has re-rated far faster than the stock. FY2026 guidance went from €34–39B to €43–45B, which puts the base FY26 EPS near €39.6 against €29 in the June dossier. On house EPS, ASML trades at ~42x FY26 and ~33x FY27, down from ~57x FY26 in June (€1,656.4 ÷ €29.0), even though the share price is about the same. The new risks are an unprecedented back-end-loaded Q4 (implied €12.9–15.9B), a down-payment float that shrank in H1, and the first credible report of Chinese immersion-DUV production. The Q3 print on 14 Oct is a test of whether the Q4 load and 2027 coverage are real. Structurally still BULLISH; still WATCHING on entry.dossier
2026-10-05editorial noteCapex figure revised: $52bn → $52.0bnCapex moved from $52bn (deep-dive-2026-07-06.md) to $52.0bn (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteMargin figure revised: 67.9% → 79.5%Margin moved from 67.9% (deep-dive-2026-07-06.md) to 79.5% (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteRevenue figure revised: 10% → $82.55bnRevenue moved from 10% (deep-dive-2026-07-06.md) to $82.55bn (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteVerdict changed: The purest listed play on the commodity-DRAM shortage — a levered cyclical mid-cycle, not a structural compounder; the AI-memory squeeze is real and the balance sheet is now fortress…Before (deep-dive-2026-07-06.md): The purest listed play on the commodity-DRAM shortage — a levered cyclical mid-cycle, not a structural compounder; the AI-memory squeeze is real and the balance sheet is now fortress-grade, but the moat is the cycle, the node is a generation behind, and CXMT is the clock ticking on the far side of the peak. After (deep-dive-2026-10-05.md): The shortage outran our last dossier. Nanya will earn roughly 2x our July FY26 EPS base (~NT$75 vs NT$38–42), it has ~NT$200bn net cash, and it trades at ~5.3–5.6x annualized Q3 earnings (5.3x at our 85% gross-margin base, 5.6x at the Street's ~80%). The question is no longer whether the shortage is real. It is when the 2028 greenfield wave (Nanya's own Fab 5A included) and CXMT's G5 capacity end it. Treat it as a peak-earnings cash machine whose price already assumes the turn, not as a compounder.dossier
2026-10-05editorial noteMargin figure revised: $9.8M → 14.6%Margin moved from $9.8M (deep-dive-2026-07-10.md) to 14.6% (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteRevenue figure revised: 88.6% → $505MRevenue moved from 88.6% (deep-dive-2026-07-10.md) to $505M (deep-dive-2026-10-05.md).dossier
2026-10-05editorial noteVerdict changed: The Q3 print settled the prior dossier's open question — Advanced Computing (the "AI factory" leg) grew just 3.8% and its operating income collapsed 84% while Integrated Memory (DRAM…Before (deep-dive-2026-07-10.md): The Q3 print settled the prior dossier's open question — Advanced Computing (the "AI factory" leg) grew just 3.8% and its operating income collapsed 84% while Integrated Memory (DRAM/Flash) exploded +111% and carried 89% of segment operating income; the bears were right on the MECHANICS and wrong on the TRADE, because the memory super-cycle is extending into 2027 (not rolling over) and the stock re-rated from $67 to $81 as management successfully re-labelled the memory spike as "AI-driven"; the new tell is the balance sheet — record EPS but operating cash flow swung to −$75M in the quarter on a $639M AR+inventory build, the classic late-cycle memory-distributor signature. After (deep-dive-2026-10-05.md): The story since July is the capital structure, not the P&L — PENG refinanced into $750M of 0% converts and crystallized ~8.7M shares from its deep-in-the-money 2029/2030 notes, its memory supplier SK hynix agreed to buy SK Telecom's preferred stake, the CFO left, and the stock round-tripped from $81 to $44 and back to ~$61; the memory tailwind is still blowing (TrendForce sees DRAM +10–15% and NAND +15–20% QoQ in 4Q26) but decelerating from Q2's +58–75%, so Tuesday's print matters less for Q4 (Street $0.77 / $519M) than for the formal FY27 guide against the ~30% framework and for whether operating cash flow turns back positive.dossier
2026-09-25editorial noteVerdict changed: The circularity flag stopped being a footnote and became the tape — a reported $250B backstop of OpenAI's data-centre debt (plus ~$350B of GPU-purchase financing under discussion) tr…Before (deep-dive-2026-08-10.md): The circularity flag stopped being a footnote and became the tape — a reported $250B backstop of OpenAI's data-centre debt (plus ~$350B of GPU-purchase financing under discussion) triggered a $1.3T semi selloff, cost NVIDIA the world's-largest-company title, and drove its 5-year CDS to the largest one-day widening on record. Yet the stock is UP ~10% since the last pass at an unchanged ~25x, because consensus EPS rose as fast as the price. The franchise is intact and the moat widened again (Vera CPU opens a claimed $200B TAM); what changed is that NVIDIA escalated from taking equity in its customers to guaranteeing their debt. BULLISH / MEDIUM / 1Y held — but the reflexive tail is now a credit tail, not just a mark-to-market one. After (deep-dive-2026-09-25.md): The print was a blowout and the guarantee turned out real — both at once. Q2 revenue $96.2B (+106%) beat the guide by 5.7%, Q3 is guided to $108B, and management gave its first-ever year-ahead number, ~70% revenue growth for FY28, which lifted FY28 consensus EPS from ~$12.82 to ~$15.68. Yet the stock is flat since the last pass ($223.96 → $223.03) because the multiple fell from ~25x to under 17x forward. The reason is in the filings. NVIDIA signed a residual value guarantee of up to $105B on OpenAI's Ohio leases, buys back $36B of capacity from the AI clouds it sells to, extended payment terms so receivables jumped $22B and operating cash flow halved, and borrowed $25B. The press figure of $250B was wrong. The primary one is smaller, delayed to FY2029, and indemnified, but it is now on the record. BULLISH / MEDIUM / 1Y held. The MarketCall's "market multiple" leg no longer holds: at ~14x FY28 the market is discounting financed earnings, not paying for them.dossier
2026-09-25editorial noteVerdict changed: The toll-booth on chip yield — a ~58%-share process-control near-monopoly whose revenue scales with chip *complexity*, not wafer volume, so it compounds even through capex air-pocket…Before (deep-dive-2026-06-17.md): The toll-booth on chip yield — a ~58%-share process-control near-monopoly whose revenue scales with chip *complexity*, not wafer volume, so it compounds even through capex air-pockets; the only real debate is the price you pay, not the quality of the business. After (deep-dive-2026-09-25.md): The business got better and the stock got cheaper. Backlog is up 60% to $12.6B, FY27 is guided to accelerate, and the shares are ~21% below the June dossier price. The June dossier said the problem was the entry price; that problem is now smaller. The risks to watch are memory-cost margin drag running into 2027 and export-control headlines.dossier
2026-09-25editorial noteVerdict changed: The number-two AI compute franchise is finally real (DC $16.6B FY25, MI450/OpenAI 6GW signed) — but at ~58x forward EPS with 320M warrant shares of dilution overhang and an analyst P…Before (deep-dive-2026-06-17.md): The number-two AI compute franchise is finally real (DC $16.6B FY25, MI450/OpenAI 6GW signed) — but at ~58x forward EPS with 320M warrant shares of dilution overhang and an analyst PT below spot, the stock has already priced the win. WATCHING, not chasing. After (deep-dive-2026-09-25.md): The business ran past the June bull case — Data Center +107%, a $13B Q3 guide, and management now guiding Data Center revenue to more than double in 2027 — while the stock rose only ~18%, so the multiple compressed rather than expanded. The June "priced for perfection" charge is weaker than it was; what replaced it is a larger, more concentrated, more debt-and-prepayment-funded bet on one ramp (Helios/MI450) that has not shipped a rack yet.dossier

Dated ahead

Dated catalysts ahead
DateCompanyPeriodFirmness
2026-10-08TSMC TSM2026-M10expected
2026-10-15TSMC TSM2026-Q3expected
2026-10-22Texas Instruments TXN2026-Q3expected
2026-10-23Amphenol APH2026-Q3expected
2026-10-23Intel INTC2026-Q3expected
2026-10-23MaxLinear MXL2026-Q3expected
2026-10-27SK Hynix 000660.KS2026-Q3expected
2026-10-28Amkor Technology AMKR2026-Q3expected
2026-10-28Cadence Design Systems CDNS2026-Q3expected
2026-10-28NXP Semiconductors NXPI2026-Q3expected
2026-10-28Qorvo QRVO2026-Q3expected
2026-10-28Rambus RMBS2026-Q3expected
2026-10-28Ultra Clean Holdings UCTT2026-Q3expected
2026-10-29FormFactor FORM2026-Q3expected
2026-10-29Impinj PI2026-Q3expected
2026-10-29KLA Corporation KLAC2026-Q3expected
2026-10-29Lam Research LRCX2026-Q3expected
2026-10-29Teradyne TER2026-Q3expected
2026-10-30Allegro MicroSystems ALGM2026-Q3expected
2026-11-02Monolithic Power Systems MPWR2026-Q3expected