Comparison window: W45 (2026.10.30–11.06) vs W46 (2026.11.06–11.13). Risk levels are unchanged at 7 red / 4 yellow / 2 green, but the risk form rotated into something that is unusual and worth naming: input-cost pressure falling while financing-cost pressure rises, with memory spot falling and memory contracts rising at the same time. Sixteen categories moved materially.
| Category / Metric | Last week (W45) | This week (W46) | Direction | What changed |
|---|---|---|---|---|
| 🆕 Macro — rates and CPI Cost of money up |
Dec hike odds ~66% | Dec hike odds rising again | 🆕 Hike back on the table | The cost-of-money side of this week's divergence. October CPI came in at +0.4% month on month against +0.3% expected, and core at +0.3% against +0.2% expected — both 0.1pp above consensus. A Fed governor stated on Nov 10 that he still leans toward a 25bp December hike and could consider 50bp, and the Cleveland Fed president called for a restrictive stance. There is no November meeting; the next FOMC is December 8-9 and carries the projections. The market repriced accordingly: 2-year Treasury +37bp to 4.88%, 10-year +29bp to 5.15%, 30-year +21bp to 5.47%, with the 2s10s at 27bp, and the dollar index at 103.08, a 19-month high. Implication: the local-currency benefit of a firming yuan is now competing with a firmer dollar and higher dollar funding cost — neither of which a procurement team can negotiate away |
| 🆕 Energy — Brent / WTI Cost of input down |
Brent 107.39 / WTI 100.06 | Brent 91.03 (-15.23%) | ↓ Premium nearly gone | EIA official Wednesday fixing: Brent USD 91.03/bbl, -15.23% for the week, with WTI -9.65% to USD 90.40 — and the Brent-WTI spread collapsing from USD 7.33 to USD 0.63 (cross-check: 91.03 − 90.40 = 0.63), below the bottom of its normal 3-8 band. Roughly 93% of the war premium has been squeezed out. But the driver is political and it reversed this week: on Nov 13 Iran reiterated that the Strait stays closed until its seven conditions are met and explicitly denied any talks with the US. Implication: with crude back where it was before the conflict, the market is about to test what it actually believes about physical supply — and a premium that left in a week can be re-injected in a week. Keep war-risk and reroute clauses in place |
| Memory spot — the whole complex 1 up, 6 down |
1 up, 5 down | 1 up, 6 down | ↓ Broad pullback | Nov 12 session: one tier up, six down. DDR5 16Gb 60.233 (-0.72%, single session flat, 30-day +2.26%, YTD +447.8%) — a second consecutive weekly fall with the decline narrowing from -1.36%, and still a second week above USD 60. DDR5 16Gb eTT 30.45 (+0.66%, the only gainer). DDR4 16Gb 79.750 (-2.36%), now 7.0% below its Sep 18 range high of 85.750. DDR4 8Gb 44.714 (-2.87%), breaking below the USD 45 level for a second consecutive week. DDR4 16Gb eTT 13.075 (-2.97%) and DDR4 8Gb eTT 5.63 (-3.26%) — the low-capacity white-label tier fell hardest, which is the direct signal that low-cost substitution economics are narrowing fastest where demand is weakest. DDR3 4Gb 13.777 (-1.39%, essentially flat on a single-session basis), with the legacy tier moving far less than the mainstream tiers. And note the generational spread widened rather than narrowed - from USD 14.46 last week to about USD 15.52 now, as DDR5 held while DDR4 8Gb fell, which is the opposite of the convergence story the previous two weeks carried. Module side: no new session, the Oct 26 quote of USD 2,165.00 is carried forward |
| Storage contracts Still rising |
4Q26 +13-18% | 4Q26 DRAM/NAND +10%-20% | ↑ Contracts still rising | This is the week's central contradiction, and it is not noise: spot fell 1 up / 6 down while contract prices kept rising. The 4Q26 band moved to DRAM/NAND +10%-20%, with PC DRAM specifically maintained at +13-18% and November contracts expected above October. One major's November NAND contract price rose 50% — far above expectations — while another controller maker expects NAND prices flat to slightly higher through 4Q and into 1Q, forecasting extreme tightness after the lunar new year; two other memory makers have issued DRAM shortage warnings. At the channel, a new round of increases started verbally: leading makers have told distributors of an intention to raise enterprise-grade particle prices, with no formal letter issued yet but channel prices already moving — concentrated in DRAM contracts and enterprise SSD, not a broad consumer-NAND move. Procurement read: budget and lock on the contract side, because the spot tape is telling you about positioning, not about your next invoice. |
| 🆕 Cycle-top signal First from an OEM |
No vendor signal | Cycle seen ending June 2027 | 🆕 Anchor | The only cycle-top signal this week came from an original equipment maker, and it deserves more weight than a brokerage view. The chairman of a listed memory maker said memory price increases have reached their limit and this up-cycle is expected to end in June 2027. In the same window, PC memory configuration regressed year on year for the first time and DDR4's share overtook DDR5 back to 54.5%, and one bank raised its 2030 global memory market estimate to USD 2 trillion. Implication: this is the anchor for long-horizon planning — a mid-2027 peak implies that 2027 capacity commitments should be sized for a peak, not for a plateau, and that any 2028 demand assumption built on a rising curve needs revisiting. It also means the squeeze has a shelf life, and procurement should know when it ends |
| MLCC lead times Hardening |
Distributor listings 61 weeks | Seven high-cap types at 61 weeks | ↑ Lengthening | The constraint moved from the general market into named parts. One major now quotes 61 weeks on seven high-cap MLCC types, and a 1005-package 47μF high-cap part has been out of stock for three months. Two other majors are running near full utilization with order-to-shipment ratios rising, and AI high-cap MLCC demand is expected to grow roughly eight-fold by 2027. The industry-wide gap is now expected to persist into the first half of 2027. Procurement implication: high-cap MLCC has moved from a price problem to an allocation problem with a named-part list — ask for allocation on specific part numbers, not for a quote on a category |
| 🆕 CCL — a plant fire New supply disruption |
CCL in execution | ~20% of MEGTRON capacity lost | 🆕 New disruption | A physical supply event rather than a price event. On Nov 4 an explosion and fire at one Japanese major's Taiwanese plant removed roughly 20% of global MEGTRON-series capacity — into a laminate chain that was already short. Expectation of shortage and quotes moved up together, while one major's November increase (M6-and-below boards +15%, prepreg +25%) remains in execution verification with books scheduled to January 2027. Implication: high-speed laminate and conventional laminate are affected differently, and the split matters — the affected capacity sits in the high-speed grade the AI chain needs, which is the opposite of what a buyer would want. Re-check the laminate supplier list on any platform using high-speed material. |
| E-fabric — spot checks Field check |
7628 at RMB 14.6-14.65/m | Pickup prices above RMB 15/m | ↑ Still rising | The Nov 3 assessment is now landing in the real market, and the field check is worse than the published number. A Nov 12 market survey found some makers' pickup prices already above RMB 15/m, more than RMB 10/m above the start of the year, with inventory at large makers down to roughly one week and small makers close to none. The published assessment itself was unchanged at 7628 at RMB 14.6-14.65/m and G75 e-yarn at RMB 28,750-28,850/t, now in the fourth week of execution verification. The market reading: no clear sign of a pullback, still rising but with a narrower range. Implication: published assessments are a floor, not a quote — ask suppliers for pickup price, not assessment price |
| 🆕 Cross-year price letters Window open |
One 2027 letter expected | Two 2027 letters landed | 🆕 Two letters | Two of the three large cross-year negotiations landed inside this window, and both landed the same way — with no magnitude and no part list. One maker issued its third letter of the year on Nov 2, effective Jan 1, 2027, citing raw material, energy, labour, logistics and upstream supplier costs; it is the only one of the three that disclosed neither a percentage nor a part list, with prices set by customer manager through customer account management. The other issued its fourth letter of the year on Nov 4, effective Jan 1, 2027, raising part of the product line with no list and no magnitude, communicated per customer. A third maker's 2027 annual negotiation is under way, with channel guidance of a full-category increase in 2027 Q1 and roughly 40% of Q1 and 50% of Q2 capacity directed to AI — a channel caliber, not a published company figure. Implication: the only remaining lever is timing. Ordering inside November is the single window that avoids both letters. |
| Metals — the direction reversed Futures |
Sn 342,760 · Ni 122,380 · Cu 104,700 | Sn 355,440 · Cu 105,380 · Ni 122,460 | ↗ Tin leads, aluminum lags | Completely reversed from last week, and the dispersion widened. Shanghai tin RMB 355,440/t (+3.70%) was the strongest industrial metal, with LME tin 46,855 (+1.06%) and LME zinc 3,692 (-0.91%) — the external tin move confirms the domestic one, which is the signature of genuine supply rigidity rather than a positioning bounce. Aluminum was the biggest decliner at RMB 22,150/t (-4.03%), even as A00 spot was marginally higher at RMB 23,130, a futures-spot divergence worth noting. Copper RMB 105,380 (+0.65%) with 1# spot at RMB 109,480 and a premium of RMB 490. Zinc RMB 25,285 (-0.34%), nickel RMB 122,460 (+0.07%) with London inventory rising, lead RMB 15,760 (-1.56%). Implication: the solder-cost relief that last week's tin decline opened has closed — tin is the strongest metal again, and the nickel-powder input for MLCC electrodes has stopped falling too |
| Precious metals Broke down |
London Ag 51.78 / Au 4,088.12 | London Ag 50.444 / Au 3,923.51 | ↓ Broken down | Last week's rebound reversed, for a macro reason. Shanghai gold RMB 842.94/g (-3.11%), closing at RMB 839, a new year-to-date low, and Shanghai silver RMB 12,099/kg (-2.50%), closing at RMB 11,985, down 4.08% on the day — the heaviest single-day precious-metal move in the window. Year to date, gold is -14.1% against a base of 981.22 and silver -32.9% against a base of 18,037. The driver is the dollar at 103.08 and a December hike being priced again, not anything about components. Procurement read: the silver-bearing negotiation window that reopened last week has closed again, so revert silver-bearing parts to on-demand buying. And note the structural point has not changed: silver price is not the long-term variable, silver consumption is |
| Freight — container Spot |
SCFI 3,687.35 (carried) | SCFI 3,572.33 (-1.46%) | ↓ Both prints now available | Both prints are now in hand, and they point the same way. The Nov 6 issue published at 3,633.45 and was backfilled; the Nov 13 issue printed 3,572.33, down 1.46% on the week, ending the rise that had run since the post-holiday thaw. CCFI 1,919.27 (-0.28%). Drewry's World Container Index reported USD 4,305 per 40ft on Nov 12 (-2%). Europe forward futures closed at 2,545 (+3.94%). Lane structure is split: one major has held at USD 3,000 per FEU for consecutive weeks and its November increases have largely not landed, while a second carrier has issued a December increase letter targeting USD 4,600 per FEU, roughly double the current level. Implication: the December contract negotiation window opens in late November and the two carriers are pointing in opposite directions |
| 🆕 Red Sea — capacity returns Structural change |
~37% via Suez | ~63% via Suez | ↗ Rerouting easing | This is the most consequential logistics change of the cycle and it cuts against last week's assumption. One major carrier restarted its China–Red Sea service on Nov 12 with a first call at Qingdao, and all four major alliances have now passed the resumption threshold; by one carrier-tracking estimate, about 63% of Far East–Europe capacity is now routing via Suez (around 70% on the Mediterranean service and 58% on North Europe). Direct routing cuts 7-10 days of voyage and releases effective capacity, which pressures Asia-Europe spot rates against carrier price increases. Implication: the "rerouting cost is rising" assumption used in previous weeks needs to be revised downward — while war-risk, which does not depend on route share, must be kept. |
| 🆕 US West rail land-side New |
Not tracked | ~14,500 boxes stranded | 🆕 New risk | A land-side constraint that has not been in previous editions: roughly 14,500 import rail containers are stranded on the US West coast — about 9,500 at Los Angeles and 5,063 at Long Beach — waiting more than nine days for rail cars, and the rail labour contract expires on Nov 20, with the union saying it will support any strike picket. Implication: this needs to enter December scheduling, not just this month's freight budget. A separate bridge closure from Nov 7 to March 2028 adds a drayage detour |
| 🆕 Leadership & credit 2027 capex continuity |
— | CEO change; credit doubling | 🆕 Watch 2027 | Two governance and credit events that do not show up in the price of any component, and both matter for 2027. First, the founder of the largest GPU supplier stepped down as president and chief executive, effective Nov 11 and announced Nov 12, confirmed by an SEC filing and a company release; the finance chief is interim chief executive and a new chair was appointed. The market's immediate reaction was the opposite of alarm — the S&P 500 rose 2.51% that week, which is precisely the cognitive bias to guard against: key-person risk does not show up in the index on the day; it shows up in 2027 capital-expenditure continuity. Second, a large AI cloud operator's Q3 filing shows recourse debt at USD 41.6bn, up 100.7% in nine months, at effective rates of 7-15%, with USD 9.2bn due within twelve months and a nine-month net loss of USD 2.1bn. Implication: AI credit-chain risk has moved from valuation to refinancing, which is a question about 2027 order continuity for high-layer boards and high-speed laminate. Watch for wording changes in capex guidance, not index moves |
| Trade policy Buffer |
Port fee due to expire | Suspended to Jan 2027 | ↗ Two-month buffer | A narrow but real relief, and it should be described accurately rather than as détente. The US trade representative extended the suspension of 178 China tariff exclusions, and of the port fee and quay-crane surcharges on China-linked vessels, to Jan 9, 2027; on the Chinese side, the suspension of the 24% additional tariffs was continued for two months from Nov 10 13:01 with the 10% baseline retained, and the special vessel port fee suspended to Jan 10, 2027. The two measures correspond to each other and buy the export chain roughly two months of buffer. That is "avoiding the negative," not a broad relaxation — and one country's trade-remedy measures were extended again in the same window. Implication: use the two-month buffer to pull forward shipments rather than to rebuild inventory. |
| USD / RMB FX |
6.7168 (11/06 parity) | 6.7152 (11/13 parity) | ↗ Small appreciation | Nov 13 central parity 6.7152, a further 16bp appreciation against 6.7168 — a very small move, and it is being driven by a dollar index at a 19-month high rather than by domestic strength. Implication: the local-currency buffer on dollar-priced imports keeps narrowing, and with the dollar rising on rate expectations the two effects now work against each other. Keep FX-linkage clauses on import quotes |
Color block = supply / price tightness. Bold outline = moved this week. Red = rising / high risk, green = falling / loose supply-demand (China market convention). Overall rating unchanged at 7 red / 4 yellow / 2 green; the risk forms changed on memory, energy, precious metals, macro and freight.
| Segment | 1H 2026 | Current (Nov 13) | 2027 Outlook | |
|---|---|---|---|---|
| Memory DRAM / DDR5 | High | Red alert | Red alert | Spot 1 up / 6 down; contracts +10-20%, NAND contract +50% |
| NAND / eMMC | High | Red alert | Red alert | NAND contract +50%; enterprise SSD channel rising verbally |
| MLCC Ceramic Capacitors | Red alert | Red alert | Red alert | 7 high-cap types at 61 wks; one 47μF part out 3 months |
| Tantalum Capacitors | High | Red alert | Red alert | Nov 1 round +20-30% in execution; longest lead time 84 weeks |
| Aluminum-Electrolytic / Film | Yellow alert | Red alert | High | Aluminum -4.03% but the constraint is lead time, not price |
| MCU Microcontrollers | High | Red alert | Red alert | Two 2027 letters effective Jan 1 with no magnitude or part list |
| Analog / Power Management | High | Red alert | Red alert | 2027 annual negotiation open; notified per part number |
| Power Devices | High | Red alert | Red alert | Taiwan +10-15% unannounced, 5th week — last window |
| PCB / CCL | Yellow alert | Red alert | Red alert | ~20% of MEGTRON capacity lost to fire; fabric field prices above RMB 15/m |
| Copper / Base Metals | High | High | High | Direction reversed: tin +3.70% leads, aluminum -4.03% lags; TC -$241 |
| Freight / War-risk | Yellow alert | High | High | ~63% back via Suez; war-risk unchanged; US West rail 14,500 boxes stranded |
| Connectors / RF | Yellow alert | Elevated | Elevated | Certification still 12-18 months; no new disclosure |
| Precious Metals | Watch | Watch | Watch | Silver -32.9% YTD; window closed again; consumption is the variable |
| Macro / Rates / Credit | Watch | High | Elevated | CPI beat, Dec hike priced, DXY 103.08; AI credit moving to refinancing |
The Nov 12 session produced one gainer and six decliners, with the legacy DDR4 tiers falling hardest and DDR4 8Gb breaking below USD 45. Yet contract prices kept rising — one NAND contract up 50%, a new round starting verbally at the channel, and two large cross-year letters landing effective Jan 1, 2027. Both halves are true, and the divergence between them is more informative than either half alone.
Memory spot is falling and memory contracts are rising. That combination has one cause: HBM, advanced packaging and AI-grade components are absorbing capacity under multi-year contractual reservation, and the marginal tier loses. Spot reflects positioning in the listed names; contracts reflect allocation. This matrix tracks where AI orders are being absorbed and what is displaced as a result — a standing weekly tracker, because it is the mechanism behind the whole price complex rather than a one-week event.
Two of the week's largest moves were not price moves at all. A plant fire on Nov 4 removed about 20% of one maker's MEGTRON capacity, and a field check on Nov 12 found glass fabric pickup prices already above RMB 15/m with large makers down to roughly one week of inventory. Meanwhile the metals tape reversed direction completely: tin led at +3.70% — closing the solder-cost relief that last week's decline opened — while aluminum fell 4.03% and copper edged up 0.65% with concentrate TC still at a near-decade low.
| Node | This week (Nov 13 close) |
|---|---|
| Glass fabric & e-yarn | The field check is worse than the published assessment. The Nov 3 monthly assessment moved 7628 heavy cloth to RMB 14.6-14.65/m (+1.5/m, about +11.2%), G75 e-yarn to RMB 28,750-28,850/t (+3,000/t, about +11.2%), 2116 thin cloth to RMB 20.2 (+2.5), 1080 thin cloth to RMB 20.75-20.8 (+2.5-2.6), E225 yarn to RMB 71,500 (+10,500-12,500) and D450 fine yarn to RMB 147,500 (+27,750). Year to date: 7628 +275%+ against a Sep 2025 base of RMB 3.9/m, G75 +121%, thin cloth +190%+, with 1080 E-glass up about +300% at one producer. A Nov 12 field survey found some makers' pickup prices already above RMB 15/m, more than RMB 10/m above the start of the year, with inventory at large makers down to roughly one week and small makers close to none. The market reading: no clear sign of a pullback, still rising but with a narrower range. Implication: published assessments are a floor, not a quote — ask for pickup price. |
| Laminate (CCL / prepreg) | A physical supply event landed in the middle of an execution period. On Nov 4 an explosion and fire at a Japanese major's Taiwanese plant removed roughly 20% of global MEGTRON-series capacity — into a laminate chain that was already short, so shortage expectation and quotes rose together. Meanwhile one major's November increase (M6-and-below boards +15%, prepreg +25%) and a second major's eighth increase of the year (2116-and-below thin cloth board material and prepreg, +15%) are both in execution verification, with orders scheduled to January 2027 and the supply gap unchanged. Implication: high-speed and conventional grades are affected differently, and the split runs the wrong way for AI work — the affected capacity is the high-speed grade. Re-check the laminate supplier list on any platform using high-speed material. A separate forward item: one Japanese major has been reported to raise some conventional-grade board material by +200% with lead times extending from Jan 1, 2027 — applicable only to certain conventional glass-epoxy grades and not to be extrapolated to high-speed grades, whose separate announcement covers roughly +20% |
| Copper foil | SHFE copper RMB 105,380/t (+0.65%) with 1# spot at RMB 109,480 and a premium of RMB 490. The foil processing-fee concession window has not substantively reopened, because copper concentrate TC remains at a near-decade low of USD -241/dmt and cannot be used as a basis for pressing price. High-end HVLP foil processing fees hold at RMB 170,000-200,000/t against RMB 26,500/t for ordinary 18-micron HTE and RMB 24,000/t for 6-micron battery foil — a 6-7x generational gap that did not narrow; HVLP foil demand is forecast up more than 260% in 2026 with high-end supply short by close to 50%, and the HVLP4 gap about 1,500 tonnes in 2026 and 2,500 tonnes in 2027 |
| Tin / solder & plating | SHFE tin RMB 355,440/t (+3.70%) — the strongest industrial metal of the week, with 1# spot at RMB 360,250 (+5,000) and LME tin 46,855 (+1.06%); LME copper 13,493.50 (-0.82%) and LME zinc 3,692 (-0.91%) - London tin rose against a weaker London copper, the cleanest confirmation available that the tin move is supply rigidity rather than a general commodity bid. The external market moving with the domestic one is the signature of genuine supply rigidity rather than a positioning bounce, and the low social-inventory, easy-up-hard-down structure is unchanged. Procurement: the solder-cost relief that last week's 5.02% decline opened has closed — tin-based solder cost is rising at the margin again, and the negotiation window that last week's price fall provided is gone. Buy on demand in batches; do not chase the rise, because the supply structure says it will not run away |
| Nickel / MLCC electrodes | SHFE nickel RMB 122,460/t (+0.07%) — essentially flat, with London inventory rising and pressuring any rebound, and 1# spot at RMB 124,100 (+450). Implication: the electrode-cost relief that two weeks of nickel weakness provided is over, though nickel powder is priced on conversion cost so pass-through still takes one to two quarters |
| Aluminum | SHFE aluminum RMB 22,150/t (-4.03%) — the biggest industrial-metal decline, even as A00 spot was marginally higher at RMB 23,130, a futures-spot divergence worth noting; alumina followed lower at RMB 2,640. The aluminum foil cost relief at the margin is real but it is not the constraint — the constraint on aluminum electrolytic capacitors is lead time, and the 400-450V industrial grades remain the binding item |
| AI high-speed CCL (M8/M9) | One institution's forecast: AI-server CCL demand moves from 42 million sheets in 2026 to 131 million in 2028, a 77% CAGR, with average pricing rising from USD 165 to 282 to 362, roughly a 48% CAGR. Another puts the 2026 high-speed CCL market at about USD 8bn. The gap is around 70%, lead times have stretched from 2 weeks to 8-12 weeks, and leading orders are scheduled into 2027. On the platform: PCB value per card has doubled from about USD 400 to over USD 800, with 40-78 layers and CCL moving from M6 to M8 |
| PPE resin (hard supply constraint) | ~70% of global PPE resin is produced in Saudi Arabia, where only about 65% of capacity has been restored and a maintenance outage is scheduled inside November — the largest single materials risk for Q4. And the geopolitical assumption behind it has reversed this week: with Iran reiterating that the Strait stays closed and denying talks, the shipping and insurance cost of moving Saudi product is no longer a tail risk being discounted. Electronic-grade PPE is up +261% year to date |
Because the binding constraint upstream is glass and foil capacity, not metal. The November assessment moved 7628 up 11.2% and D450 fine yarn up RMB 27,750/t in the same week that aluminum fell 4.03% and copper rose 0.65%. A plant fire then removed about a fifth of one maker's high-speed laminate capacity. Metals are priced on an exchange against macro expectations; fabric is priced against how much capacity exists and how quickly furnaces can be relit — and a furnace, once lit, must run 8-10 years, 2026 is a concentrated cold-repair year, and laminate expansion runs 18-24 months with about 12 months of certification. That is why the fabric field check shows pickup prices above the published assessment, and why it will stay that way.
On the metals side the week's dispersion was itself the signal. Tin rose 3.70% with London confirming at +1.06% — external confirmation of domestic tightness — while aluminum fell 4.03% against a marginally firmer spot, and copper added only 0.65% with concentrate TC still at USD -241/dmt. A broadly falling complex would have signalled cost relief; what actually happened was one metal with a supply problem getting more expensive while another with an inventory problem got cheaper. For a buyer, that means the cost-down opportunity is narrower and more specific than a headline "metals are down" would suggest — and the specific one, solder, has just closed.
On demand the bifurcation that keeps every PCB sub-tier out of the green zone persists and is now quantified better than before: the 2026 global PCB market growth forecast has been revised up to +18.8%, about USD 101.95bn, with boards above 18 layers up 79.0% and China above 18 layers up 93.5%; AI-server high-layer and HDI are the increment; and in China, 1H26 disclosed PCB expansion investment exceeded RMB 70bn, essentially all of it AI-directed, with zero traditional consumer-line expansion. One capacity figure worth holding on to: equipment lead times run 10-24 months, so capacity ordered in 2027 forms effective supply in 2028 at the earliest — the 2027 gap does not close on this capital cycle. And with the November fire removing high-speed laminate capacity on top of it, the gap in the specific grade the AI chain needs is now larger than the headline picture implies.
Year-on-year growth is refreshed from disclosed monthly and quarterly data each week and labeled with its basis. Actual and forecast calibers are never mixed without saying so, and a change of measurement basis is always stated.
| Application | Demand | Visibility | YoY | What changed this week |
|---|---|---|---|---|
| AI servers / compute | Extreme | Into 2027 | +31% TrendForce 2026 global AI server shipments (>2m units) |
Two governance and credit events land this week, and both point at 2027 rather than at orders. First, the founder of the largest GPU supplier stepped down as president and chief executive effective Nov 11, confirmed by an SEC filing and a company release — and the S&P 500 rose 2.51% that week, which is the point: key-person risk does not show up in the index on the day; it shows up in 2027 capital-expenditure continuity. Second, a large AI cloud operator's Q3 filing shows recourse debt of USD 41.6bn, up 100.7% in nine months, at 7-15% effective rates, with USD 9.2bn due within twelve months and a nine-month net loss of USD 2.1bn — AI credit-chain risk has moved from valuation to refinancing. On the demand side nothing changed: original-equipment capacity remains sold out to 2029 with more than half of 2027-2028 capacity under long-term agreement. Action: watch for wording changes in capex guidance, not index moves. |
| Consumer electronics (phone / PC) | Contracting | 6-12 months | -22% IDC 3Q26 global phones, released Oct 14 |
No new readings this week. Maintained: 4Q26 major notebook ODM order volumes forecast down 10-15% year on year, and PC-maker DRAM inventory stretched to 7-10 weeks. One structural signal worth adding to that: PC memory configuration regressed year on year for the first time, with DDR4's share back over 54.5% — which means the downgrade is running through the mix, not just the unit count. Consumer-line material should be bought on demand, not on a forward view |
| Automotive electronics | Falling (structural divergence) | 3-6 months | -21% Nov 1-8 passenger retail, released Nov 12 — partial month |
A second consecutive negative reading, on a new caliber. For Nov 1-8, national passenger-vehicle retail was 293,000 units, down 21% year on year and up 2% on the day, with new-energy retail at 219,000, down 7%, and new-energy retail penetration at 74.9%; daily average retail over the period was 37,000 units. The full-month October reading was -25%, so the short-window figure is marginally less bad but confirms the direction: automotive electronics demand has now been negative for two consecutive periods and the purchasing basis must be marked down again. But the structural cause of the squeeze is unchanged — AI continues to take the allocation from vehicle-grade parts, and automotive memory certification runs 2-3 years, so vehicle-grade parts cannot be re-sourced inside a quarter. Guidance: negotiate on annual framework prices and do not place expediting orders. |
| Industrial control | Firm | 2-4 months | +18.9% September industrial robot output |
No new readings. European energy costs remain high — German and French industrial power above EUR 120/MWh — but this week's 15.23% fall in Brent marginally eases input-cost pressure for European industrial customers. Industrial MCU lead times of 72-80 weeks, booked into 2028, are unchanged |
| Photovoltaics | Weak | 1-3 months | -36% to -49% CPIA 2026 full-year forecast, 160-200GW — forecast caliber |
No new readings this week, and last week's expectation-driven rebound did not extend. Two cost-side reversals matter more than the demand side: silver fell back below USD 51 (London 50.444) after reclaiming USD 50, so conductive silver paste cost relief reverses again; and polysilicon closed at 35,360, -2.62% on the day, with inventory at 370,000 tonnes. Implication: the last week's silver-bearing concession is gone — revert to on-demand buying and track consumption, not price, because 2026 photovoltaic silver consumption is down 20% on silver-coated copper and copper-plating substitution |
| Energy storage | Recovering | 1-3 months | +42% CNESA September newly installed power — tracked separately from solar |
No new concentrated procurement data. Cost side: lithium carbonate futures at RMB 92,380, up 0.72% on a wide intraday range driven by short covering, while battery-grade spot fell 1,150 on Nov 5 — the loose supply picture is unchanged and the green alert is retained. The bigger issue for storage PCS remains aluminum electrolytic lead time at 40 weeks, not the lithium price |
| Telecom base stations | Steady (capacity-driven) | 3-6 months | +11.7% 5.254m 5G base stations end-Sep vs 4.705m a year earlier |
No new tender announcements; the domestic carriers' 4Q26 tendering cycle that began Oct 1 is maintained. The structural watch remains: communication-grade MLCC used in optical modules has started to show shortages — and with high-cap MLCC now at 61 weeks and 47μF parts out of stock, the base-station and optical-module chain now competes for the same constrained capacity as the AI-server chain |
Standing section, rewritten each week for the changes. Pull-in strategy runs parallel to cost-down / substitution paths.
| Segment — this week's move | Procurement / pull-in strategy (this month) | Cost-down & alternatives |
|---|---|---|
| MLCC — part-specific allocation7 high-cap types at 61 weeks; one 47μF part out 3 months | The constraint has stopped being a category and become a list of part numbers. With seven high-cap types at 61 weeks, one 47μF part out of stock for three months, and AI high-cap demand expected to grow roughly eight-fold by 2027, a category-level quote no longer describes what you can actually get. Action: compile the high-cap part numbers on your bill of materials and request allocation on each of them separately, in writing, with a stated delivery window — and do it before the allocation round closes, because two other majors are already at near-full utilization. Ask for pickup price rather than published assessment on any glass-fabric-dependent line: the field check already shows real prices above the published number with large makers down to about a week of stock. | Generic 0402/0603 and consumer grades remain decoupled from AI high-cap, and consumer and mid-cap spot fell about 5% in early October before stabilizing — so there is genuine slack below the AI tier. Rationalize part numbers to improve allocation leverage. And because one leading maker has signalled a mid-2027 peak for this cycle, 2027 capacity requests should be sized for a peak rather than a plateau. |
| Cross-year 2027 price lettersTwo letters landed, neither with magnitude or part list — both effective Jan 1 | Timing is the only lever left, and it closes this month. Two makers have now issued letters effective Jan 1, 2027 — one on Nov 2 for its third round of the year, one on Nov 4 for its fourth — and neither disclosed a percentage or a part list, with prices set per customer through account management. A third vendor's 2027 annual negotiation is open, and channel guidance points to a full-category increase in 2027 Q1 with roughly 40% of Q1 and 50% of Q2 capacity directed to AI. Action: order inside November for anything you need in 2027. That is the one window that avoids both letters, and with 35-40 week lead times on analog, 1Q27 material has to be ordered this quarter in any case. Then request per-part-number price notifications with stated quote validity — a blanket percentage applied across a portfolio will be wrong on part of it. | Do not extrapolate a single vendor's letter across the portfolio; three vendors are negotiating on different bases and with different disclosure. Where a part has an alternative vendor already qualified, the letter is a negotiating asset rather than a cost. Where it does not, the only defence is ordering early and asking for a written price per part number with a validity period. |
| Memory — contracts over spotSpot 1 up / 6 down while NAND contracts rose 50% | This is the week to be explicit about which half of the market you are buying from. Spot fell across the board — with the low-capacity white-label tiers falling hardest and DDR4 8Gb below USD 45 — while contract prices kept rising: the 4Q26 band at DRAM/NAND +10%-20%, PC DRAM +13-18%, one NAND contract +50%, and a new enterprise-grade particle increase round already moving at the channel before any letter was issued. Budget and lock against contracts; treat the spot tape as positioning in the listed AI complex, which is where the Oct 5-equivalent moves came from. And note where the substitution economics are going: the low-capacity white-label tier fell hardest, so any cost-down plan built on low-capacity white-label substitution needs re-testing now, not later. | Use the weak legacy tier to rebalance mix rather than to build a cost-down case: DDR3 4Gb was essentially flat while mainstream tiers fell, which makes legacy parts the stable holding. If a design uses a low-capacity part that was a substitution candidate, the white-label discount has narrowed fastest there — re-quote before committing. And with a vendor signalling a mid-2027 cycle peak, avoid buying long-dated commitments beyond that horizon. |
| PCB / CCL — a supply event~20% of one maker's MEGTRON capacity lost; fabric field prices above RMB 15/m | Two things to do that have nothing to do with price. First, re-check the laminate supplier list on any platform using high-speed material — a plant fire on Nov 4 removed about 20% of global MEGTRON-series capacity, and the affected grade is the high-speed one the AI chain needs, which is the opposite of what a buyer would want. Second, stop treating the published fabric assessment as your price. A Nov 12 field check found pickup prices already above RMB 15/m, with large makers at about one week of inventory and small makers close to none, so ask for pickup price and for the current lead time in weeks, per grade. And revise the tin assumption: tin rose 3.70% and is the strongest metal of the week, so the solder-cost relief that last week's 5.02% fall opened has closed — do not budget a further reduction, and do carry the aluminium relief as the only one that survived, and even that is not the binding constraint. | Keep glass fabric, resin, foil and base metal as four separate cost line items — this week they moved in four different directions, and the one that tightened (tin, foil, laminate) is the one that matters for your BOM. Rational material downgrade; layer reduction; panel optimization; qualify a second laminate supplier on high-speed grades specifically, not generically. A second producer's loom capacity is already pre-booked, so qualification slots are worth taking early. |
| Hormuz — premium out, risk inSpread to USD 0.63, but Iran denied talks on Nov 13 | Two things moved apart this week and only one of them is good news. The crude premium is essentially gone — Brent -15.23%, the Brent-WTI spread at USD 0.63, below the bottom of its normal band, roughly 93% of the war premium squeezed out. And on Nov 13 Iran reiterated that the Strait stays closed until its seven conditions are met and denied any talks with the US, reversing last week's diplomatic-restart narrative. Action: keep war-risk and reroute clauses in place, and do not treat the lower crude price as permission to reduce them. What has genuinely improved is the route: about 63% of Far East–Europe capacity is back via Suez and direct routing saves 7-10 days. So revise the reroute-cost assumption down and the insurance assumption not at all — they are different exposures and this week they moved in opposite directions. | With 7-10 days of voyage recovered, in-transit inventory carrying cost and pipeline stock fall — worth revisiting delivery-date commitments and buffer-stock sizing on Europe lines. Air freight is another story: global air rates rose about 6.6% to a 2026 high, and a bridge closure from Nov 7 to March 2028 adds a drayage detour, so do not assume the cheaper ocean route fully substitutes. Hold air-freight contingency for lanes where the war-risk rate has stepped up. |
| US West land-side~14,500 boxes stranded; rail contract expires Nov 20 | A new constraint that is not in most risk registers. Roughly 14,500 import rail containers are stranded on the US West coast — about 9,500 at Los Angeles and 5,063 at Long Beach — waiting more than nine days for rail cars, with the rail labour contract expiring on Nov 20 and the union saying it will support any strike picket. Action: put this into December scheduling, not just this month's freight budget — if a strike lands, December US-line arrival cost and lead time both need re-estimating, and the same week brings the Panama Canal's 2027 dry-season slot auctions on Nov 17 and Nov 24, which allocate December-to-January US-line capacity. | For US-bound product, where the schedule allows, moving volume to the west coast through ocean with rail delay accepted is cheaper than air at current rate levels. For US-line shipments that cannot slip, book rail capacity and rail equipment earlier than usual in December, and treat the Nov 17/24 canal slot auctions as a cost input rather than a logistics afterthought. |
| Taiwan power / MCU+10-15% planned, unannounced for a fifth week | Fifth consecutive week, and the last remaining pricing window. The Taiwan plan for a 10-15% increase on spot products remains "planned, not announced" as of Nov 13, and one Chinese maker's Q3 commentary puts channel inventory at 11 weeks with roughly 80% of orders on lead times beyond 16 weeks. Complete the new-order negotiation with the Taiwan makers this week and secure written magnitude and effective date — if it slips past the midterms, you negotiate from a weaker position with a 30-50+ week lead time behind you. Note also that the Section 232 refined-copper decision is now 46 days overdue and still suspended rather than cancelled, so copper-intensive lines carry two open questions at once. | Cross-brand pin-to-pin on MOSFET / IGBT / SiC; package arbitrage; qualify second sources on long-lead SiC now rather than after the announcement. One maker's January 1, 2027 repricing belongs in 2027 budgeting rather than in this cycle's scramble — and note it is attributed to structural costs in materials, energy, advanced packaging and logistics, all four of which are currently moving against you. |
| Macro — the cost of moneyCPI beat, Dec hike priced, 2Y +37bp, dollar 103.08 | The side of this week's divergence you cannot negotiate away. October CPI came in 0.1pp above consensus on both headline and core, a Fed governor publicly leaned toward a December hike of 25bp with 50bp on the table, and there is no November meeting before the December 8-9 FOMC with projections. The 2-year yield rose 37bp to 4.88% and the dollar index hit a 19-month high at 103.08. For a buyer this means three things: the local-currency benefit of a firming yuan is shrinking (parity 6.7152, only 16bp of appreciation, and dollar-driven rather than domestic); working capital costs are rising on the same window in which component prices are also rising; and any financing-linked or index-linked component clause now carries a different risk than it did a month ago. Action: fix currency and rate exposure on 2027 commitments now, while the clause is still negotiable, rather than after the December projections. | Prefer fixed-price or capped-adjustment clauses over index-linked ones for anything extending into 2027, because the index reference itself is what moved this week. For USD-denominated imports, buy forward where the local-currency gain is small. And for precious-metal content, note the window has closed again: silver is -32.9% year to date and the buyer has no leverage left there — revert to on-demand and change specifications instead. |
Attention allocation: act immediately in the top-right, monitor continuously in the bottom-left.