Summary of Views on Space Computing •We expect ~5GW space compute capacity by 2028 via ~10k Kessel racks (supply chain ramp 2H27–2028) for SpaceX Space Computing. •Kessel = compact NVL72: 72 Rubin GPUs at ~4,000W TDP each!!!
→ Peak rack power 350–400kW (avg ~280kW for inference)
→ Compressed from conventional 48U into 9U low-and-fat design •Next-gen AI satellite solar likely be upgraded to 350–400kW peak compute (up from AI1’s 210kW solar / 250kW peak) •Hardware: ~60-layer PCB midplane + Amphenol Paladin connectors + PTFE + M9 hybrid (TLPS)
→ Midplane ASP >$20k
→ Single-loop liquid cooling to external radiators •Launch support: 4 operational Starship pads by end-2027 + twice-weekly launches targeted in 2H27 📈 Supply chain: 1) Hon Hai sole assembler; 2) AVC cold plates for the ultra high TDP; 3) Taiflex/EMC for PTFE & M9 CCL; 4) Fositek >50% QD share; 5) Solar Power supply chain: Brave C&H Supply (6538 TT), TSEC (6443 TT) #space# #emc# #nvda# #avc# #PCb# #solar#

Summary of AMD (AMD US) – Earnings Preview
 • Expect inline 3Q results but near-term guidance of ~$17.6b vs. Bloomberg $16.5b, buy side $18b
• bullish on 2027 datacenter: well over 100% YoY growth driven by agentic AI
• EPS revised -5% / +15% for 2026E/2027E on sales of $103b in 2027
• New TP $720 based on 40x 2027E EPS (implies 28x 2028E) Earnings preview 
• 3Q26 rev: $13.3bn (includes ~1 month of MI450)
• 4Q26 rev: $17.6bn (above Bloomberg, below aggressive buy-side) 
• Helios rack shipments started, but volume ramp mostly 1Q27
• Server CPU revenue: +100% YoY in 2027, +50% in 2028
• Higher 2nm wafer allocation + strong Venice yields (CoWoS 97-99% from ASE)
• Venice shipments >5mn units in 2027 → higher blended ASE)
• MI450/455X On schedule, but limited volume by end-2026 (weight/switching constraints)
• Meaningful ramp of racks in 1Q27 → >20k racks expected in 2027
• Multi-GW agreements supporting the ramp #AMD# #AI# #ServerCPU# #MI450# #Helios# #EarningsPreview#

Summary of AI Financing - Overinvestment Amid Widening Quality Divide 
* AI Red Queen Effect keeps CSPs spending hard — but capital is getting more selective • Compute stays tight: MSFT/AMZN/GOOGL all flag supply tightness. 
• U.S. Top-5 CSP capex still strong: +85% / +53% / +20% in 2026-28
-> Balance sheets remain healthy into 2028 (net leverage 0.6x, interest coverage 14x)
• But quality divide is widening fast:
– Hyperscaler 20y+ spreads vs CoreWeave yields 
– Larger players (SPCX, Top CSPs) securing preferential funding
– Expect capex continues, but funding increasingly concentrates with the strongest balance sheets. IG providers (Top-5 CSPs + SPCX) to take 84%/83% of U.S. capacity build in 2026E/27E. #AICapex# #Hyperscalers# #Neocloud# #AIFinancing# #RedQueenEffect#

Summary of Robust VR200 Outlook 
📈NVL72 rack TAM: 73k / 96k in 2026E/2027E • Vera Rubin ramping with 85%+ first-pass yield; further acceleration expected in 1Q27
• 2026E: 67k GB200/GB300 + 6k VR200
• 2027E: VR200 + initial VR300 (Ultra) drive most of the growth (86k units)
• SpaceX set to be largest VR200 buyer in 2027E (20k racks on 30k+ demand / ~10GW)
• Major 6 neoclouds (SpaceX + CoreWeave + tier-2s) to take 19k/34k racks (27%/35% of industry)
• CoreWeave near-term funding risk viewed as manageable Vendor dynamics
• Hon Hai winning bigger SpaceX allocation (model 40-45%) + share gains at Google/Meta
• Pegatron limited; DELL/Wistron face share pressure Recommendations
• Hon Hai (2317 TT) Buy – rising SpaceX + CSP share
 #VR200# #NVL72# #AIServer# #Neocloud# #HonHai# #Lenovo#

Summary of Unimicron ( – Initiate BUY, TP NT$1,820
77% EPS CAGR amid AI ABF supercycle • AI ABF substrate TAM set to hit US$13bn / $18bn in 2027E/28E (+100%/41% YoY)
• Unimicron poised to capture >30% share
• ABF capacity +18%/35% in 2026E/27E → ABF revenue +66%/90% to NT$105bn/199bn
• Supply shortage expected to persist into 2029 (18%/16% shortage in 27/28)
• Product mix shift (high-end ABF + mSAP + LPDDR) drives GPM to 28.8% / 41.7% / 44.6%
• EPS: NT$28.6 / 60.8 / 89.3 (vs consensus 25/55/80)
• TP based on 30x 2027E P/E #Unimicron# #ABF# #AI# #Semiconductors# #TaiwanTech#

For design, GOOG team is the bottleneck in my opinion. For substrate supply, Ibiden gives the full support to Intel, while ShinKo yields are slightly ahead. All the substrate makers should feel happy about supporting Intel (one can double check). The mass delivery is far away but yield improvement is on track.

Summary of ASE Holdings (3711 TT) - Tailwinds from Pricing and LEAP (Leading Edge Advanced Package) 🚀 Raise TP to NT$843 (from NT$754), based on 25x 2027E EPS * Packaging bottleneck broadens: Supply tightness extends beyond AI into analog and mature-node chips, with lead times increasing >30%. * Aggressive price hikes: Expect +10–80% for FC/Wirebond/QFN and +10–20% for CoWoS in 3Q–4Q26. * LEAP growth accelerates: Forecast LEAP revenue of US$4bn/8bn in 2026/27, above guidance of US$3.6bn/7.2bn. * AMD as a key driver: Higher server CPU wafer allocation, * CoWoS adoption for Venice/Verano (and all future AMD server CPUs) , and limited competition support growth. * 4Q26 preview: Expect ATM revenue +9% QoQ, EMS -5% QoQ and gross margin reaching 23%, supported by pricing and richer LEAP mix. #ASE# #Semiconductors# #AdvancedPackaging# #CoWoS# #AMD# #AI#

Our U.S. Tour and OCP Global Summit 2026 Takeaways ☕️OCP – System-level inflection
• Decisive shift toward full-stack co-design across hyperscalers
• Nvidia extended networking roadmap new USFP pluggable standard (co-developed with Cisco). Enables 12.8 Tb/s modules
• Showcased PTFE for NVLink Board (inline with our PCB Spec view for 2H27 on Oct 5)
• Microsoft emphasized AI scaling as a system-engineering challenge
• MSFT Unveiled multi-GW Maia 300 deployments from 2027
• Meta maintained pragmatic stance on optics, piloting near-packaged solutions in scale-out domains and Retaining copper for high-density scale-up fabrics ☕️OCP – Optical scale-up and power integration
• Broadcom and OpenAI advanced OCI-MSA multi-rack optical scale-up, highlighted hybrid topologies: high-bandwidth copper for tensor parallelism + optical links for expert parallelism
• GEN2 and GEN3 timelines: 400G in 2027, 800G in early 2030s 
• Google shifted focus to power, positioning datacenters as grid-integrated resources
• 1 GW demand-response capacity with U.S. utilities
• LVDC roadmap: from ±400 VDC Mt Diablo racks to native 800 VDC distribution
• Power engineering now dictates deployment velocity as much as silicon performance
• AMD reinforced CPU-dense trajectory of agentic AI
• Announced Helion CPU racks reaching 32,768 EPYC cores
 🚌 U.S. Tour — Supply Chain and Compute Economics
• No near- to mid-term power-supply risk, Supported by gas-based generation and remote locations 
• Liquid-cooling infrastructure is the bottleneck
• Rack pricing continues to rise (GB300 US$5.5–6.5m; VR200 US$8m), reflecting strong compute demand
• Some believe Inference economics challenging: GPU utilization 20–30%, memory and cooling costs rising
• For agentic AI, model shifts CPU-to-accelerator ratios from 1:4–1:8 to 1:1 → doubling CPU TAM #OCP# #AI# #MSFT# #GOOG# #NVDA# #VR200#

TSMC (2330, TSM) 3Q26 Summary - Growth intact, profitability less so ☕️ Earnings lukewarm, growth high, margins softer. • 3Q: Rev +16% QoQ / +41% YoY, GM 66.4% (slight miss), EPS NT$29.4
• 4Q guide: Rev $52.3–54.3bn (+14% QoQ / +58% YoY), GM 65–67% softer
• 2027: +39% YoY rev on strong AI demand
• Capex elevated; N2 capacity more than doubles in 2027, A14 ramps faster
• Faster A14 important for mobile customers facing recent SoC challenges in 2nm, but adds margin pressure to TSM
• Any Terafab collab would require TSMC to operate the fab (therefore -> unlikely to happen given Musk’s comments)
• Margins expected largely flattish through 2027–28 on dilution + overseas expansion 💰Slight EPS lift, TP nudged to NT$2,999 (20x 2027E). Hold Rating ! #TSM# #AI# #semi#

Further update on PTFE Adoption: NVDA Showcased PTFE at OCP 🤖 Nvidia showcased its NVLink PTFE board prototype at OCP Global Summit 2026, naming 7 suppliers (in the chart) Key takeaways:
• PTFE hybrid remains the preferred solution for VR300 NVL576 switch boards — a balance of performance vs manufacturing 
• Long-term potential extends beyond switch boards (IEC standard for PTFE-filled CCL already published; testing underway for compute boards)
• Near-term impact on conventional resins limit (only 8–12 of 32 layers use PTFE), but pure PTFE solutions could rise
• Favor suppliers with proven PTFE MP capabilities: Taiflex, leading Chinese CCL, and the leading Chinese PTFE PCB maker Positive on PTFE-exposed names with real technical edge. #PTFE# #PCB# #RUBINULTRA# #NVDA#

ASIC Update: TPU v9 Timeline & v10 Opportunities • We push out TPU v9 Humufish expectations by one quarter
Core A slipped ~3 weeks → MTK Core B/I/O mods. Sampling still Nov; expect tape-out in Jan 2027, MP tape-out 2Q27, shipments 1Q28. • Constructive on EMIB-T: root causes of substrate yield (45-50%) identified → potential re-acceleration. • Google still undersupplies its own models. Gemini 4 trained on ~200k TPU v7, but heavy external allocation (esp. Anthropic) in 1H26. More internal in 2H26 helps, yet compute gap widens — Anthropic + OpenAI each +5–10 GW in 2027 vs few GW for Google. • Broadcom unchanged: 460k CoWoS wafers in 2027, shipments late 2026.
MTK Zebrafish volumes higher into 2027–28, partially offsetting v9 timing. • TPU v10 (4Q28/1Q29): Icefish (MTK + GUC 3DIC on A14) the only confirmed mainstream SKU. Merope uncertain; Broadcom faces AMD+Marvell competition. EMIB-T momentum rising for v10 • Near-term headwinds for MTK; mid/long-term still like MRVL, MTK (100% v9 + potential new DC customers), and Intel (EMIB-T). #MTK# #TPU# #Broadcom# #INTC# #MRVL#

$TSM earnings preview: AI growth, Terafab reset & N2/A14 • Raise TP to NT$2,960, based on 20x 2027E P/E; Hold • Expect 3Q26 revenue +15.2% QoQ and 67.8% GM; forecast 4Q +12% QoQ and 67.2% GM, supported by Rubin/N2. • Forecast 2027 revenue +37% YoY, driven by AI, CPUs, networking and modest price hikes. • Expect 2026 capex guidance unchanged; estimate $83bn/$95bn in 2027/28. • Terafab expectations reset: Musk says they will build and run the fab themselves, tempering hopes for a major TSMC partnership and is inline with our view • Expect Apple to adopt A14 in 2028, accelerating its typical three-year node cadence (consumer feedback on recent 2nm mobile SoCs lukewarm) #TSM# #TSMC# #AI#

Summary of Thoughts on PTFE vs. HC ✍️ PTFE hybrid still the most likely PCB solution for Nvidia’s VR300 NVL576 switch board • We est. HC hybrid (HC + ABF/M9) recently passed SI testing and is ~3 months behind PTFE hybrid in the qualification — has not yet entered PCB MP testing • Final material decision expected by end-26. We think Leading Chinese PCB player has already achieved meaningful yields on the PTFE hybrid • Remain constructive on PTFE for high-end/high-speed AI PCBs; any HC adoption would likely be transitional #PCB# #PTFE#

Summary of UMC 2303 TT note on market worries vs. our view re-rating • UMC share price weak on worries over “no price hike in 2H26”
• We see 2H26 hikes are for selective customers (e.g. MTK 8”). → The broad-based hikes to be in 1Q27 (10-30%). ⬆️ Raise TP to NT$215 : Meaningful re-rating path into 2H26–2028 ☕️ 2H26–2027: Profitability optimization on Higher UTR + price hikes + improving mix
• 12” UTR → ~93% by end-2026 (Novatek/Samsung LSI 22/28nm + Apple Mac)
• 8” UTR → ~90% by end-2026, full loading by 2H27
• Pricing momentum extends into 1H27 (10–30% across nodes) 2028 ⬆️new growth phase from 2028
• Asset-light FinFET 12nm model , with Apple DDIC likely be the key driver 
• Optical full-solution model (SiPh/TFLN + interposer/advanced packaging) - TFLN EPS contribution of NT&0.7/NT$1.6 in 27/28. 💰 EPS lifted and target multiple to 25x 2027E EPS. Solid FCF + dividend yields intact. #UMC# #Optical# #TFLN#

Summary of MTK (2454) Update - 100% EMIB, 100% Share in v9 •Street-high TPU estimates → raised TP to NT$6,650 (from NT$6,420) •MTK fell 7% yesterday on noise around potential v8t delay, EMIB/substrate supply issues & v9 share-loss fears — we take the opposite view •v8t/Zebrafish remains intact; demand well above 2027 CoWoS allocation of 180k, suggesting extended life cycle vol. •EMIB-T / substrate yields & supply tracking better than expected, at ~50% currently. Yield rate in every stage in report •EMIB Substrate delivery in small-volume ~100k mid-2027 → mass-volume by 4Q27 💯 100% EMIB-T for v9 (Humufish, Triggerfish). on-track substrate yields + suppliers dedicated support. Also no reason for TSM to be a back-up or 2nd source. 💯 100% market share in v9 (Humufish for training, Triggerfish for inferencing). We expect no other v9/2nm projects during the time period •Life-cycle volume for v9 expected >5m (vs bears at >2m) •2nd ASIC project (likely SpaceX ASIC, not AI6) more visibility expected in 4Q26 •Datacenter ASIC forecast raised to NT$2.8bn / $18.6bn / NT$75.7bn •Near-term watch: engineering tape-out in Nov (passing bar not high; chipset testing at TSMC largely on track) •For v10: still expect MTK to win major SKU Icefish (AMD+MRVL or AVGO competing for Merope) #TPU# #MTK# #AVGO# #AMD# #MRVL# #v9# #intel# #intc#

Takeaways - CPO Accelerates, Yet Stock Implications Vary: •Nvidia CPO scale-out progressing faster than expected. Spectrum-X CPO switches in mass production thanks to improving yields at key component suppliers. •Nvidia CPO scale out switch shipment forecast: 15k / 16k / 30k / 100k for 4Q26 / 1Q27 / 2Q27 / 2027. •Scale Up OE: We revised estimates to 6.5m / 42m in 2027E / 2028E on intact NVL576 but slower T4 (likely in 1Q28) •VRU NVL576 CPO (QM5 + 4 OEs) and NPO (QM5 + 4 OEs with sockets) progressing in parallel, though we retain full adoption of NPO for the cycle. •Feynman scale up CPO in design likely up to 6.4T OEs for QM6 packages. Likes •Lumentum – CPO acceleration supports valuation; CPO/NPO revenue contribution seen at 12%/26% in CY27/CY28. •Broadcom (3163 TT) & Bizlink (3665 TT) – Beneficiaries of CPO’s higher content vs NPO. •Semtech (SMTC) – TIA/Driver exposure to NPO; tight supply. Dislike •Largan (3008 TT) – EPS contribution only NT$20/NT$97 in CY27/CY28; implied valuation of ~20x CY28 not cheap + market share assumption risks due to current supplier’s yield improvement and growing number of competitors #NVDA# #CPO# #NPO# #scale# #LITE#

View summary- Akamai–Anthropic deal: 💰 $11.6bn 7-year accelerating CPU agreement announced 24 Sep (potential +$9bn expansion). Anthropic to use Akamai’s distributed AI infra & software for growing CPU workloads. 
🤖 Vera CPUs expected: Nvidia previously flagged Akamai’s Vera deployment plans. Vera targets agentic workloads (code, orchestration, data processing) with high-bandwidth LPDDR5X. 
• Positive read-throughs:
– Server CPU & DRAM demand (agentic AI needs more compute + memory)
– eSSD / NAND (context, vector DB, KV-cache growth; better cost/power vs DRAM/HDD)
– ABF substrates
– Lenovo (major server OEM for Akamai)
• Our Server CPU TAM: 2030 server CPU TAM of $215bn 
• Stocks & Positive Implications :
– CPU: NVDA, AMD, INTC
– Memory: SK Hynix, Micron (DRAM/HBM); Samsung, SK Hynix, MU + SanDisk (NAND)
– Advanced packaging: ASE (Buy), TSMC (Hold), Amkor
– OEM: Lenovo (Buy) #anthropic# #NVDA# #CPU# #DRAM#