SemiAnalysis Weekly

Ep. 018 - Stop Saying Half of 2026 US Datacenter Capacity Is Canceled (Datacenter, Energy) | Jeremie Eliahou Ontiveros, Reyk Knuhtsen, Ellie Holbrook, Jordan Nanos

Media reports claiming that half of projected 2026 US data center capacity is canceled misinterpret early-stage project volatility as systemic failure. In reality, the industry is experiencing a strategic pivot toward behind-the-meter power generation to bypass grid interconnection bottlenecks, with forecasts projectin...

Ep. 017 - DeepSeek V4 and Huawei Ascend NPU Performance (InferenceX) | Kimbo Chen, Cam Quilici, Bryan Shan, Jordan Nanos

DeepSeek V4’s transition to a 1-million context length architecture relies on aggressive innovations in sparse attention and Mega MoE, which reduce KV cache memory requirements by approximately 100x compared to standard models. Achieving day-zero inference performance on new hardware requires complex engineering, speci...

Ep. 016 - What Unitree's Evolution Means For Robotics (Robotics) | Jordan Nanos, Reyk Knuhtsen, Niko Ciminelli

Unitree is rapidly scaling the humanoid robotics market by applying the same economies of scale and supply chain integration that propelled China’s dominance in drones and consumer electronics. By utilizing the mature manufacturing ecosystem in Shenzhen, the company produces affordable, iterative hardware that allows r...

Ep. 015 - DG Matrix Explains 800V DC vs Legacy AC Distribution (Datacenter, Energy) | Jordan Nanos, Jeremie Eliahou Ontiveros, Nicolas Bontigui, Haroon Inam

The transition to 800V DC architecture in datacenters is essential for supporting the rapidly increasing power demands of AI compute racks, which are scaling toward one-megawatt densities. By raising voltage, operators significantly reduce current and copper requirements, improving efficiency and lowering infrastructur...

Ep. 014 - Finding Miscompiles For Fun, Not Profit (AI Infrastructure) | Justin Lebar & Jordan Nanos

Compiler engineer Justin Lebar details his process for identifying critical bugs in NVIDIA’s PTX compiler and LLVM backends using a combination of traditional fuzzing and LLM-assisted code analysis. By leveraging AI to generate random programs and directly inspect source code, he uncovered significant vulnerabilities, ...

Ep. 013 - AWS Margins Jump 10% While Azure and GCP Flatline (Tokenomics) | Jordan Nanos, Jeremie Eliahou Ontiveros, Joey Brookhart, Crystal Huang

Anthropic’s rapid growth and the successful implementation of the "token-as-a-service" model through AWS Bedrock are significantly boosting Amazon’s operating margins, contrasting with the declining or stagnant margins of peers like Microsoft and Google. Unlike the capital-intensive, bare-metal infrastructure-as-a-serv...

Ep. 012 - Cerebras IPO: 1,100 Tokens Per Second vs GPU's 500 (AI Cloud TCO) | Jordan Nanos, Howie , Myron Xie

Cerebras’s wafer-scale engine technology is redefining AI inference by prioritizing raw speed through massive on-chip SRAM, effectively bypassing the data movement bottlenecks inherent in traditional GPU clusters. By stitching 84 die radicals together, the architecture delivers exceptional performance for low-arithmeti...

Ep. 011 - GPT 5.5 vs Claude 4.7: OpenAI's Comeback From the Brink (Tokenomics) | Jordan Nanos, Dylan Patel, Doug O'Laughlin, Max Kan

Frontier AI models are currently locked in a tight competition where raw benchmark scores no longer reliably predict real-world utility. While GPT 5.5 has restored OpenAI’s standing against Anthropic’s Claude 4.7, users increasingly prioritize agentic workflows and CLI-based orchestration over marginal improvements in ...

Ep. 010 - How Much Do GPUs Really Cost, and Where Does the Value Go? (AI Cloud TCO) | Jordan Nanos, Dan Nishball, Kang Wen Cheang, Zane Fong

Total cost of ownership for GPU clusters extends beyond initial purchase prices, requiring a focus on "good put"—the measure of actual useful work performed by a cluster. High-performance computing environments face significant performance degradation from hardware failures, such as GPU link flaps and memory errors, wh...

Ep. 009 - Using Open Source Data To Drive Investment Decisions (ChipBook) | Chaim Eisenberg, Simi Sherman, Jordan Nanos

Semiconductor investment strategies rely on transforming messy, distributed open-source data into actionable intelligence to validate theses and monitor industry inflections. By tracking granular supply chain metrics—such as regional wafer fabrication equipment (WFE) imports, DRAM inventory levels, and photo mask write...

Ep. 008 Claude Code Psychosis: How SemiAnalysis Is Token Mogging Meta | Dan Nishball, Sam Harshe, Jordan Nanos

Agentic workflows powered by Claude Code are fundamentally transforming research productivity by automating complex, multi-step tasks like financial modeling, conference synthesis, and supply chain analysis. By structuring "swarms" of specialized agents—such as research directors, model builders, and event summarizers—...

Ep. 007 - The 3 Choke Points Killing the AI Boom (Core Research) | Nick Doyle, Nigel Chiang, Konrad Wang, Jordan Nanos

The SemiAnalysis Weekly podcast explores current choke points in the AI infrastructure supply chain. The discussion highlights constraints at nearly every layer, from TSMC's N3 node being sold out through 2027 to shortages in INP-based lasers used in optical transceivers and PCB substrates. The panelists note that whil...

Ep. 006 - The AI Silicon Shortage Explained (AI Supply Chain & Fabs) | Sravan Kundojjala, Ivan Chiam, Jordan Nanos

The podcast analyzes the AI silicon shortage impacting the tech industry, highlighting wafer fab capacity constraints at TSMC and memory supply issues. Demand for AI accelerators is outpacing supply, squeezing consumer electronics. TSMC's role as "kingmaker" in allocating resources and prioritizing stable, high-volume ...

Ep. 005 - Measuring AI's Impact On The Market (Memory, Tokenomics, Macro) | Ray Wang, Malcolm Splitter, Joey Brookhart, Jordan Nanos

The podcast explores the impact of memory constraints on AI development, the potential for AI in various sectors, and the evolving economic models needed to measure AI's true value. It highlights how increased memory costs are driving up hyperscaler budgets, drawing parallels to the oil crisis in terms of elasticity of...

Ep. 004 - The Impact of AI Datacenters On Consumer Power Costs (Datacenter, Energy) | Jeremie Eliahou Ontiveros, Jordan Nanos, Doug O'Laughlin

The SemiAnalysis Weekly podcast explores the impact of AI data centers on consumer electricity prices, contrasting the PJM and ERCOT utility models. The discussion highlights that AI data centers' effect on electricity prices is significantly influenced by market design, particularly the forecasting accuracy and real-t...

Ep. 003 - Deep Dive on NVIDIA Vera Rubin VR NVL72 (AI Supply Chain) | Jordan Nanos, Myron Xie, Copper Wei (Wega), Howie

The Seminalysis Weekly podcast delves into NVIDIA's upcoming Vera Rubin GPU, successor to Blackwell, focusing on its architecture and performance enhancements. The discussion highlights Rubin's extreme co-design, featuring two compute dies, eight HBM4 stacks, and desegregated IO chiplets. A key improvement is the adapt...

Ep. 002 - InferenceX 2.0 Release (Technical Staff) | Cam Quilici, Bryan Shan, Doug O'Laughlin, Jordan Nanos

InferenceX provides a transparent, iterative benchmarking platform for AI inference, tracking the performance of frontier hardware like NVIDIA’s GB200/B300 and AMD’s MI355. By implementing advanced techniques such as disaggregated prefill, expert parallelism, and multi-token prediction (MTP), the project demonstrates t...

Ep. 001 - Claude Code, Memory Mania, CPUs are Back | Jordan Nanos, Doug O'Laughlin, Myron Xie