Nvidia's Rubin GPU Delayed by Memory Shortage, Technical Hurdles
Nvidia's Rubin GPU is likely to be late thanks to memory shortage and technical challenges https://t.co/VMHNcFjnzc https://t.co/jjDxxHh9FN
2026 Market Forecast Jumps $200B, Extending ROIC Horizons
It is wild that about two months ago, consensus memory in dollars for 2026 was ~$600B and now people are saying north of $800B... Very challenging to square this with ROIC timelines now, as they will have to lengthen, and...
From $1T in 70 Years to $2T in Five
What no one saw coming about this, since most $$ estimates were on volume not ASP, is just that - this is simply costs going up. It took >70 years to hit $1T and it will go from...
Supply Chains Plan Years Ahead, Not Weeks.
It’s 2026, and I have to wait a month to obtain a basic stripped down Mac Mini. The semiconductor supply chain plans for demand years in advance. It should be able to plan and respond weeks in advance.
Advanced Packaging Forces US Chips Back to Taiwan
Choke Points: Why even the best chips made in the U.S. take a round trip to Taiwan? It’s because of Advanced Packaging facility… World has woken up to the supply chain choke points… Katie Tarasov @KatieTarasov of CNBC explains moe...
TSMC Poised to Dominate Silicon Photonics Like AI Packaging
I'm glad TSMC COUPE is getting more attention of those in the weeds in optics. Our view: "TSMC is increasingly positioned to occupy the same role in silicon photonics that it came to occupy in advanced AI packaging." https://t.co/WMdxP2tF3R
Samsung Predicts Profit Surge From AI-Driven Memory Price Boom
MyPOV: @Samsung forecasts insane operating profit growth thanks to surging memory prices https://t.co/TtaW3vxXCN @SiliconANGLE @Mike_Wheatley “Nobody could have foreseen the ferocity of AI’s memory appetite, but it’s still a great advertisement for…” - @holgermu @constellationr
Export Controls Separate H200 From Vera Rubin Supply Chains
Export Controls: National Security Tool or Industrial Policy Lever? H200 exports do not pose a meaningfully trade off against newer Vera Rubin due to distinct supply chains. Yes, have been saying this for some time, fully debunks this TP.... https://t.co/WJlqCYF9zG
Elon Musk’s $16.5B Deal Rescues Samsung From Chip War
Samsung was getting crushed by TSMC in the chip war. - $3.6B in losses - 7% market share vs TSMC's 67% - About to be overtaken by China's SMIC But Elon Musk just changed everything with one $16.5 BILLION deal... Leaving Big Tech in shock. Here's...
Memory Dominance Determines AI Compute Winners for Years
If the winners and losers in AI compute come down to who gets the most memory in the most ridiculous memory-constrained market we have ever seen.... For the next 3-4 years.... Predicting those winners is pretty easy.
AMD Likens CUDA Battle to Mountain Climb, Doubts RT Rollout
.@AMD being humble, says taking on @NVIDIA’s CUDA is like climbing a mountain… 👇🏽 Then again, why will AMD RT this? 🤔 @dvellante @furrier @Srasgon @dfloyer @Scobleizer @om @karaswisher
AXTI Set for Q2 GaAs Wafer Price Surge
very positive for $AXTI - Tomorrows Digitimes China is reporting Gallium arsenide epitaxial wafers are poised for a Q2 price surge amid large demand. AXTI owns Tongmei Xtal Technology Co., Ltd in China https://t.co/lZIckgtRlu
Self‑Optimizing AI Loop Cuts Cost, Boosts Speed
Awesome. 680% speed improvement. 97% cost reduction. As the hardware/software loop condenses, intelligence is folding back on the physical substrate of its own existence, optimizing the lithography that prints the transistors that run the models that optimize the lithography. We're seeing...
Trim ASML Position Amid Stretched Valuation Despite Strong Demand
Macro: robust EUV demand supports ASML. Key: €100m buyback, Q4 rev beat, large SK hynix order. Risk: stretched valuation despite 113% YTD gain. Insight: trim position to lock gains. — Viktor Kopylov, PhD, CFA More insights: t.me/si14Kopylov
Memory and Storage Converge Into a Single Layer
New report on @DiligenceStack. Companion to the memory deep dive as co-optomization is a memory and storage story. Storage Wars: When Memory and Storage Collapse Into One Layer https://t.co/ozDZpHYNi3
JPM Predicts AVGO Valuation Exceeds $120B Post‑earnings
After $AVGO earnings, I said ~120B was the likely number for 27. Now, looks like higher is likely. This from JPM. https://t.co/OvxZg1VUWr
Intel Partners with Musk on Terafab AI Chips
Intel joins Musk's Terafab AI chip project to power humanoid, data center goals “One facility would power cars and humanoid robots, while the other would be designed for AI data centers in space, Musk had said.” https://t.co/tCGZgjWZ72 @intel https://t.co/M5L1Q6Njr6
Interface Issues Appear in Advanced Packages, Not Originate There
In advanced packages, the interface is where problems show up, but rarely where they begin. https://t.co/Tefs3KJnTc #semiconductor #advancedpackaging #semiconductortest #interface
Intel Poised to Surge if Musk's Terafab Materializes
Intel $INTC would jump up more if the market would realize what exactly @elonmusk is about to pull off with Terafab (no I don't invest, just observing, ignore me) https://t.co/Y5iJPsX0w7
Terfab Ranks Among Top Three WFE Spenders for 1GW Compute
Putting Terfab into perspective. Based on stated goals: 1GW of compute requires about 50kwpm of logic, 140kwpm of DRAM, ~30kwpm of NAND, and ~35 kwpm of CoWoS (advanced packaging). That translates to about $30bn of WFE per 1GW produced...
Intel Partners with Musk's Terafab, Signaling Its Decline
Poor, old Intel is reduced to working with Musk. How the once-mighty do fall. Intel says it’s joining Elon Musk’s ‘Terafab’ project https://t.co/kXVUDQWDBf
China Claims 99% Cost Cut for Military‑grade Infrared Chips
China reports 99% cost cut on military-grade infrared chips Researchers have developed a new way to make high-end infrared chips that could slash their cost and boost smartphone cameras, self-driving cars https://t.co/t52LLpN3xK via @scmpnews
Congress Targets Rare‑Earth Magnets, Threatening US‑China Ties
#MATCHAct very real danger to bilateral relationship, + for US firms facing REE/magnet-related issues. Beijing would respond harshly if some version moves forward. From Coordination to Constraint: Congress Eyes a Memory Squeeze on China—and the World https://t.co/y1sG4ZDOZo https://t.co/oZ3fvm8npl
AI Compute Profits Flow Upstream,
economics of compute: - OEM sells chip + components at 80% margin - neocloud 2-4x multiple expansion running chip - inference co 5-10x multiple expansion selling tokens - consumer gets subsidized inference so net loss for AI co (or shift cost to consumers) compression...
Photonics Emerges as AI’s Next Data Bottleneck
From Copper to Light: The Critical Bottleneck Reshaping AI Infrastructure 🧵 Photonics is becoming one of the most important AI bottlenecks. Compute is scaling fast, but moving data between GPUs is now the harder problem. Copper hits power and bandwidth limits....
Anthropic Secures TPU Capacity, Signaling Compute Diversification
.@Anthropic inks deal with @Google, @Broadcom for TPU capacity https://t.co/FCwUyLMN3c >> Congrats. Compute restrictions are real and supplier diversity is critical. Inflection point for Google TPUs going beyond @GoogleCloud. #NextGenApps by @LDignan
Questioning Chip's Role Replacing Marvell Backend for TPU
Wasn’t this the chip that supposedly was going to replace $MRVL back end design for TPU?
AMD‑Arm 2014 Opteron A1120 Packs Eight Cortex
I finally got one. @AMD x @Arm in 2014. Opteron A1120, aka AMD Seattle Up to 8x Cortex-A57, 4MB L2, 8MB L3 128-bit DDR3/4-1866 PCIe 3.0 x8 and 2x10GbE https://t.co/w08sMceEPB
Broadcom Expands AI Chip Deal with Google, Anthropic
Broadcom to Supply AI Chips to Google, Computing Capacity to Anthropic in Expanded Collaboration https://t.co/yHCXipodGy
Apple’s Chips Reshape Tech, Windows Remains Biggest Win
Apple's chips are the core of a new landscape, but its biggest win is Windows https://t.co/4r2S3X5l3x https://t.co/cZkkMSXv99
Broadcom Secures Expanded AI Chip Deals, Shares Surge
Only good news for $AVGO I will post my scorecard soon. Broadcom agrees to expanded chip deals with Google, Anthropic - Broadcom said it agreed to produce future versions of Google's artificial intelligence chips, - The company also announced an expanded deal with Anthropic,...
Semiconductors Demand Trillion‑scale Cleanliness Beyond Human Vision
#Technology #Thread #Semiconductor #Manufacturing #Trillion The Semiconductor Visualizing Trillion: 1/ - I Was Once Asked How Small A Contaminant Has To Be To Ruin A Chip. - My Answer: "Smaller Than A Speck Of Dust And A Million Times More Invisible." - That Is...
Andreessen Predicts Years‑long Chip Shortage, Investors Weigh Timing
Marc Andressen says there's going to be chip shortages for years to come. Tempted to buy 2377 TT but how do you time it https://t.co/mmAb7yVsW1
Broadcom to Build Custom TPUs for Google AI Racks
Broadcom $AVGO just said it has entered into a long term agreement with Google $GOOGL "for Broadcom to develop and supply custom Tensor Processing Units (“TPUs”) for Google’s future generations of TPUs and a Supply Assurance Agreement for Broadcom to...
Predicting XPUs: Once Mocked, Now Inevitable GPU‑only World
I remember so many people thinking I was nuts to suggest that years ago that XPUs would be successful and that it would be a GPU-only world.
China’s CXMT IPO Casts Doubt on Memory Upcycle
China’s CXMT IPO clouds memory upcycle outlook Rising supply risks threaten pricing power of Samsung, SK hynix, Micron Clouds? Shortages everywhere you look.... https://t.co/VfzNQDNgtZ
Packaging Shift Lets Amazon, Google Embrace Intel Foundry
Pat’s Law of Something: All major vendors and suppliers are always in talks and testing. I do think this will move forward, though. Packaging is an easier commitment than wafers as it doesn’t force a redesign. Not easy but just...
New Model Updates Cost‑Per‑Transistor Economics for AI Era
Working on building a new cost per transistor model with updated economics since it's been a few years since I have seen one. So many more dynamics are needed than the ones from a few years ago, pre AI...
AI Auto-Optimizes CUDA, Outperforms PyTorch and Torch.compile
People are using AI to break CUDA’s moat. Given any pytorch model, it profiles it, ranks bottlenecks by amdahl's law, writes triton or CUDA C++ replacements, and runs 300+ experiments overnight with no human in the loop. - 5.29x over pytorch eager...
Understanding the Different Types of Semiconductor Products
#Technology #Newsletter #Semiconductor #Manufacturing #Product NLOG-298 | Semiconductor And Beyond Newsletter | The Semiconductor Product Types: https://newsletter.chetanpatil.in/p/semiconductor-and-beyond-newsletter-298/
SiC Devices Power Next‑Gen Medium‑Voltage Solid‑State Transformers
Drew Baglino: "Medium voltage is the next frontier." "The latest SiC devices are more than a better switch. They are the foundation for solid-state transformers, or SSTs." #alwaysbecharging ⚡️⚡️⚡️ https://t.co/6ck5EnX9Sb
From Silicon Chips to Integrated Systems: Industry Shift
#Technology #Blog #Semiconductor #Manufacturing #System BLOG-321 | The Semiconductor Shift From Silicon To System: https://www.chetanpatil.in/the-semiconductor-shift-from-silicon-to-system/
Memory Supply Chain Faces Growing Constraints, Market Unaware
The memory point is not the only interesting comment here, but we have had a range of supply chain chats on memory, and I don't think the market fully grasps what's coming. We are stacking constraints at this point. This...
Tech Giants Wield Supply Chains to Crush Competition
Not just Apple making it hard on competition via supply chain power, but NVIDIA and Broadcom also. I outlined all this in my Master of the Supply Chain report. https://t.co/9ZCdGgQZVs
Memory Market Focus Shifts to Duration Over Demand
New report in the @DiligenceStack. Memory market update and sharing the framework for our tracking model of the cycle's durability. The Next Debate in Memory Is Duration, Not Demand https://t.co/gB1OsgGpBW
MediaTek Pushes 400G Copper Limits; Optical Next
Good discussion here on copper/optical with @mediatek. They are pushing the limits with NPC (near package copper) and CPC (co-packaged copper) at 400G per lane, with distinct efficiencies in those solutions. But past that, most likely optical NPO/CPO.
MediaTek's Flexible IP Strategy Positions It for Customer‑Owned Tooling
Being flexible, customer friendly, and willing to work with as little or as much customer owned IP the customer wants is an approach @mediatek does seem well positioned for. If you believe customer owned tooling becomes a thing. They seem competitive....
Mediatek's Custom XPU Gains Edge via TSMC Ties
This may be the money slide for @mediatek on custom XPU solutions. One area I do believe they have a distinct advantage is their privileged relationship with TSMC. https://t.co/2dIFSrVRMz
MediaTek Unveils XPU Within Broader Custom Solution Stack
The XPU is one part of what is now becoming a custom solutions approach (not just the compute ASIC) but @mediatek showing what they have been investing in across the custom solution stack. Hadn’t seen many of these details before....
Diverse Supply Chains Boost Customer Confidence, MediaTek Shows
In a supply chain world of constraints having a diverse supply chain and key partners gives customers higher confidence in execution. @mediatek showing that depth as an advantage for their datacenter efforts. https://t.co/OMAXc4kkPE