# AI capex — X 热门讨论 (2026-09-30 20:48 UTC)
## @Abbycadabby87 (Abby) · 09-30 19:55 · ♥63 ↻3 💬0 The memory bears are looking at this market completely backward. Nomura’s latest projections show global data center memory demand scaling from $107B in 2025 to $454B in 2026—eventually topping $1.4 Trillion by 2030.
Data center capex is shifting hard into memory, with AI and general DC storage rising to command over 56% of total industry revenue. As AI inference explodes and HBM/eSSD demand eats up raw wafer capacity, supply constraints will force an unprecedented, prolonged supercycle.
$MU,$SNDK, Samsung, and SK Hynix are the core infrastructure tollbooths of the decade. Time for the market to rerate them. > 引用 @RealNickMugalli: Heading into $MU's print, the focus is on a high conviction beat and raise scenario. F4Q revenue, GM, and EPS should easily clear Street consensus ($51.4B / 86.2% / $31.73) backed by ~20% sequential ASP hikes across both DRAM and NAND and record datacenter revenue topping $5B.
The bigger upside driver lies in F1Q27 guidance, where mgmt is set to raise the bar as CY27 shapes up to be even tighter than CY26. With HBM4 ramping 2x faster than prior generations and forward bit production under SCAs scaling toward 50%, earnings visibility and floor margins are structurally higher. World Trade Securities reiterates overweight with a $2,180 price target ahead of today’s earnings report!
Full post: https://t.co/DvV2XBy5gC https://x.com/Abbycadabby87/status/2105386054018531677
## @wallstengine (Wall St Engine) · 09-30 20:11 · ♥51 ↻6 💬2 MICRON $MU Q4’26 MORE ER DETAIL:
Q4 Segments Revenue: 🔹 Cloud: $16.28B (Est. $15.14B) 🟢; +258% YoY 🔹 Core Data Center: $18.00B; +1,042% YoY 🔹 Mobile: $13.11B (Est. $12.95B) 🟢; +249% YoY 🔹 Auto: $6.82B (Est. $4.73B) 🟢; +376% YoY
CapEx & Cash Flow: 🔹 CapEx (net): $10.77B 🔹 Adj. FCF: $33.20B (Est. $19.91B) 🟢
Fiscal 2026 Highlights: 🔹 Revenue: $133.19B vs $37.38B prior year 🔹 Non-GAAP EPS: $75.52
Updates: 🔸 Began sampling the industry's first 512GB ultra dense high-capacity DDR5 RDIMM modules capable of speeds up to 9,200 MT/s
🔸 Server LPDDR SOCAMM portfolio revenue more than doubled sequentially in the fourth quarter of 2026
🔸 7600 PCIe Gen 5 and 9650 PCIe Gen 6 SSD high-performance products are now shipping to market-leading customers for KV-cache applications
🔸 Now shipping 6-channel 1-gamma LPDDR5X for flagship mobile devices
🔸 Secured design wins with every Tier 1 OEM customer for AI workstation product https://x.com/wallstengine/status/2105390182476521491
## @stock_arabic (وارن) · 09-30 19:01 · ♥52 ↻2 💬2 أكبر من أن يسمح له بالفشل - Too Big to Fail
- بحسب قسم ادارة الاصول في بنك جي بي مورجان !
- من المتوقع أن تضخ شركات الـHyperscalers قرابة ال 800 مليار دولار خلال عام 2026 في بند الإنفاق الرأسمالي (CapEx) !
- وبحسب التقديرات، أضافت الاستثمارات المرتبطة بالذكاء الاصطناعي نحو 0.47 نقطة مئوية إلى نمو الناتج المحلي الأمريكي البالغ 2.1% ، أي أن ال AI ساهم تقريبا بـ خُمس نمو الناتج المحلي الأمريكي !!!!!!!!!!!
- قطاع الذكاء الاصطناعي واسع وممتد من شركات أشباه الموصلات والحوسبة السحابية، إلى الطاقة والشبكات وأنظمة التبريد لكن رغم اتساع هالمنظومة، تبقى شركات الـHyperscalers هي العمود الفقري الحقيقي اللي يعتمد عليه هالقطاع بشكل كلي !!!!
- الارتفاع الكبير في أسهم شركات الذكاء الاصطناعي عزز مفهوم (Wealth Effect) فلما ترتفع قيمة الأسهم والمحافظ الاستثمارية، الأسر الأمريكية صارت تحس انها اكثر ثراء وهذا يخليها أكثر استعدادا للصرف والاستهلاك , وهذي نقطة مهمة لأن الاستهلاك الشخصي يمثل قرابة 70% من الاقتصاد الأمريكي !!!!
- أي تباطؤ في حجم الانفاق الرأس مالي من قبل الـHyperscalers ، أثره ما راح يقتصر عليهم فقط !
-فالهايبر سكيلرز اليوم هم نسخة عام 2026 للبنوك في عام 2008 ! https://x.com/stock_arabic/status/2105372392352522336
## @Kaizen_Investor (KaizenInvestor) · 09-30 20:08 · ♥31 ↻2 💬3 $MU stays flat after a massive double beat. EPS rose 1,003% YoY and Revenue 379% YoY.
Some early takeaways
HBM Yield Progress & Margin Accretion Non-GAAP gross margin expanded to 87.0% in FQ4 (with guidance holding near 86.25% for FQ1-27) and Cloud Memory gross margins held firm at 83%, confirming that high-density TSV stacking yields have matured rapidly enough to convert AI premium pricing directly into record profitability.
Wafer Cannibalization & Capacity Squeeze The disproportionate cleanroom and wafer consumption required for advanced packaging has successfully starved broader supply, creating structural tightness that propelled Core Data Center gross margins to 90% and operating margins to 85%.
Pricing Durability Across Node Transitions Rather than succumbing to spot-market softening, conventional memory pricing displayed remarkable contract power; Mobile and Client revenue surged to $13.11B with 90% gross margins as 1-gamma LPDDR5X shipments and high-density DDR5 RDIMM qualifications accelerated.
Capex Discipline & Balance Sheet De-risking While FY26 net capex expanded to $27.37B ($10.77B in FQ4), capacity additions remain strictly anchored to demand via multi-year Strategic Customer Agreements which is evidenced by $12.75B in customer contract deposits and $73.48B in total cash and investments.
Customer Inventory Digestion Pacing Channel overhang concerns in legacy end-markets were decisively erased, with broad-based sequential expansion across Mobile, PC, and Automotive driving FQ1-27 revenue guidance to $61.5B (±$1.5B) and demonstrating immediate absorption of edge AI silicon. > 引用 @Kaizen_Investor: Some of my thoughts on the upcoming $MU earnings this Wednesday.
Top line expectation - $51,500 Bottom line - 31.6 EPS
There are 5 other structural drivers I'm watching closely
HBM yield progress & margin accretion The complex 12- and 16-high TSV stacking makes High-Bandwidth Memory exceptionally difficult to manufacture at scale. Because these dies command a massive premium, incremental yield improvements flow entirely to the bottom line. If yields scale efficiently, watch for forward EPS estimates to revise sharply higher.
Wafer cannibalization from packaging constraints Producing a single HBM wafer consumes roughly 3x the cleanroom capacity of standard DRAM due to die size and backend complexity. Surging AI demand is actively starving the conventional memory market of supply. Tracking this trade-off dictates whether the broader DRAM cycle stays structurally tight or loosens prematurely.
Pricing durability across conventional transitions While HBM dominates the narrative, traditional DDR5 and LPDDR5x still fund baseline operations. Amid concerns over softening spot prices, the durability of contract pricing will prove whether the HBM-induced supply squeeze is actually strong enough to override sluggish macro demand in the PC and smartphone segments.
Capex discipline vs. cleanroom limits The market inherently fears memory upcycles ending in undisciplined overbuilding. Capital expenditures must be scrutinized to ensure they are tethered strictly to multi-year, non-cancelable agreements rather than speculative capacity. Physical cleanroom construction bottlenecks act as a natural governor on supply growth, making long-term dynamics more predictable.
Customer inventory digestion pacing Beyond the AI data center, legacy markets are navigating uneven demand. If mobile and PC OEMs accumulated excess memory inventory to front-run earlier price hikes, an extended digestion period will stall new orders. The pace at which these specific channels clear dictates when baseline volume leverage returns.
The technical setup will react to the headline, but these underlying fundamentals dictate the multi-quarter trend. https://x.com/Kaizen_Investor/status/2105389462834266179