Fabless account for 80%+ of TSMC revenue and there is a decent but not strong correlation between fabless inventory days and TSMC q/q revenue growth. For every 10 days of days of fabless inventory decrease, TSMC revenue increases by 1.2%. TSMC outperformed/underperformed fabless

2025 AI Diffusion Export Controls Microsoft Regulatory Capture Oracle Tears Impacts Quantified Model Restrictions Malaysia's Stranded Capacity India Brazil Middle East Concerns Sovereign AI Killed And Much Much More semianalysis.com/2025/01/15/202…

SemiAnalysis code is the basis of AMD's training benchmarks 🫡

In this clip from #YAVP episode #289, Jeremie Eliahou Ontiveros from SemiAnalysis discusses the demand from AI for data centers and electricity on the heels of DeepSeek.

Check out this awesome episode by Yet Another Value Empire and Jeremie Eliahou Ontiveros youtube.com/watch?v=-H4Gak…

Fire

On Friday, there was a report stating that Microsoft was canceling datacenter leases for several hundred of MWs – alluding to a high risk of overbuild/oversupply. We discussed this in December in our Datacenter Model where we track every individual DC. Let's break it down cause

America Is Missing The New Labor Economy Robotics Part 1 China's Dominance Playbook General Purpose Robotics Is The Holy Grail Robotic Systems Breakdown Supply Chain Hardships The West Is Positioned Backward and Covering Their Eyes semianalysis.com/2025/03/11/ame…

Tariff Armageddon? GPU Loopholes Mexico Supply Chain Shift Wafer Fab Equipment Optical Module Pricing UPS, Generators, Transformers, Switchgear, Power Distribution Equipment, Cooling, Pumps OEM & ODM Supply Chain Softbank Impact Nvidia Balance Sheet Usage semianalysis.com/2025/04/10/tar…

Just dropped my third Power Bytes note introducing a new metric I developed: Data Center Reserve Margin (DCRM). It’s designed to quantify how much additional AI/data center load a region can realistically absorb before grid reliability breaks down. (1/4) 🧵 Full analysis + case

Meta Superintelligence Leadership Compute, Talent, and Data Massive AI Datacenter Cluster "The Tent" Llama Behemoth 4 Failure Post-Mortem Talent Wars, AI Data, Zucc Founder Mode Meta Will Never Pay Taxes Again semianalysis.com/2025/07/11/met… Analyst Credit: Dylan Patel, Jeremie Eliahou Ontiveros, Wei

Going live on TBPN at 12:45P PT to talk about Meta, Oracle and big AI clusters!!

We spoke with Jeremie Eliahou Ontiveros (Analyst @SemiAnalysis) about the US data center race: "Since mid-2023, there's a frantic search for power happening in the US and worldwide." "There's ~500 gigawatts of data center load requests in the US which is a figure close to the nation's actual

Last week, Christopher Mims 🤌 asked us if NVIDIA’s claims for 30x inference performance with Blackwell were exaggerated. We said yes, 30x improvement is gen-on-gen, doesn’t consider perf/$, and only applies in specific scenarios (large MOE’s). So, Chris put me on the front page of wsj dot

Robotics Levels of Autonomy Single-Purpose Robots Automating Hundreds of Jobs Pick And Place with low Autonomy is Expensive General Purpose Autonomy Navigating And Inspecting Large Sites Targeting Low-Skill Labor In Early Pilots with Promise Autonomy Capable Of Any Task in

Microsoft datacenter strategy – still in⏸️mode 🧵 Jeremie Eliahou Ontiveros Four months ago we flagged Microsoft’s pause on more than 1.5 GW of self-build capacity — on top of the leasing freeze we called out in December. Microsoft’s latest 10-K backs us up: committed datacenter construction

Thanks for the invite Andrew Walker !! Looking forward to our next convo!!

Amazon and Anthropic clearly aren't falling behind on GenAI :)

Too often, AI and energy conversations happen in separate rooms. That gap inspired Megha Jain and I to launch the Watt & Why Series ⚡ Excited to kick this off and big thanks to Crusoe and SemiAnalysis for co-hosting the first session. A space that brings everyone

ClusterMAX 2.0 is here! In this update we rate 84 neoclouds, up from 26 Our market view tracks 209 providers, up from 169 in March and 124 last October We spoke to over 140 end users as part of the research And we wrote over 43,000 words about our experience This is more than