Bloomberg said half of 2026 US data center capacity is delayed. The SemiAnalysis Data Center, Energy, and Industrials team pulled the underlying report and found a broken denominator. Amazon alone built 4GW in 2025 and is adding 5GW plus in 2026. CoreWeave adds a gigawatt, all under construction. Jeremie Eliahou Ontiveros (@JeremieEO), Reyk Knuhtsen (@robotknower), and Ellie Holbrook join Jordan Nanos (@JordanNanos) as they walk through why the number is wrong and what the real forecast shows. 

The team covers behind the meter power generation reaching 40GW by 2028, Oracle's New Mexico problem, and the gas turbine supply chain that is running toward peak. They break down the three types of data center delays, how OEMs are responding, and where solar, batteries, and nuclear fit. 

00:00 Intro

00:51 The "half of capacity is canceled" myth

04:43 Why early stage projects get canceled

08:06 Three types of data center delays

08:40 Oracle's New Mexico problem

13:47 Behind the meter: 40GW by 2028

18:50 How OEMs are responding

20:24 Signed deal to powered GPUs

23:44 Hyperscaler market share

27:49 How SemiAnalysis tracks data centers

31:34 What behind the meter means

33:31 The gas turbine supply chain

38:22 Peak turbine

42:54 Solar, batteries, and nuclear

46:13 Final thoughts

48:20 Favorite projects

Referenced:

Stop Saying Half of 2026 US Datacenter Capacity Is Canceled: https://newsletter.semianalysis.com/p/stop-saying-half-of-2026-us-datacenter

US Grid Constraints: Towards 40GW+ of Behind-The-Meter Datacenter by 2028?: https://newsletter.semianalysis.com/p/us-grid-constraints-towards-40gw

DeepSeek V4 claims a 100x KVcache reduction versus a standard MoE model, hitting 1M context length through compressed sparse attention and heavily compressed attention. Kimbo (@Kimbochen), Cam Quilici (@noslawextratost), Bryan Shan join Jordan Nanos (@JordanNanos) to break down what changed from V3, why the new MHC dimension tripped up NVIDIA on day zero, and how Mega MoE fuses compute and communication into a single kernel.

The vLLM versus SGLang NDA access gap and the Huawei day zero optimization guide circulating on Twitter.The crew walks through the InferenceX article on going from day zero to day 43 support and what that grind actually looks like across different hardware. Subscribe for weekly semiconductor and AI infrastructure analysis from the SemiAnalysis team.

Referenced:DeepSeekV4 1.6T Day 0 to Day 43 Performance Over Time - GB300 NVL72, Huawei, MI355X, B200: https://newsletter.semianalysis.com/p/deepseekv4-16t-day-0-to-day-43-performanceChapters:

(00:00) DeepSeek V4 vs V3 changes(01:00) Sparse attention and KV cache reduction(03:04) Day zero runtime support challenges(05:34) What Mega MoE actually is(08:38) Downsides of fusing kernels(10:25) MegaKernel benchmark claims(12:59) AMD FP4 optimization gains(15:14) Compounding step by step improvements(17:59) vLLM versus SGLang competition(19:34) Open source vs vendor libraries

Unitree is going public, boasting 67% gross margins on its humanoid robots. Jordan Nanos (@JordanNanos), Reyk Knuhtsen (@robotknower), and Niko Ciminelli discuss how the Chinese company achieves this through aggressive pricing, rapid iteration, and a focus on "good enough" hardware for the research and hobbyist markets. This strategy allows Unitree to dominate, much like DJI and BYD did in their respective fields.The discussion explores the reality of humanoid robot deployment versus market hype. While industrial applications are in their "baby days," Unitree's approach leverages economies of scale to create a significant moat, challenging US competitors to match their production volume and cost efficiency. The team analyzes if the US can truly compete with China's manufacturing might in the emerging robotics sector.Join SemiAnalysis Weekly for expert insights into the semiconductor, AI infrastructure, and robotics markets. Subscribe for deep dives into AI supply chain, chip economics, and market analysis.Article: https://newsletter.semianalysis.com/p/chinas-unitree-will-dominate-globalTimestamps:00:00 — Intro, deployment reality vs. hype02:39 — Unitree Business: why go public, margins, pricing05:48 — Parallels to DJI and BYD, economies of scale as China's moat16:52 — When is Claude Code moment for robotics18:38 — Real use cases and future demand shocks26:39 — Shenzhen and the humanoid BOM35:56 — The bear case: who actually buys them?42:17 — Can the US compete?

NVIDIA's next architecture will demand 800V DC and datacenters can't wait to build the infrastructure. Haroon Inam, CEO and Co-Founder of DG Matrix, joins Jordan Nanos (@JordanNanos), Jeremie Eliahou Ontiveros (@JeremieEO), and Nicolas Bontigui to explain why megawatt GPU racks make 800V DC a necessity, not an option. Haroon details his journey from uncool power electronics to enabling superhuman intelligence infrastructure.

Full Article Link: https://newsletter.semianalysis.com/p/inside-the-800vdc-revolution-part

Chapters:

00:00 Haroon Inam's Background and Power Electronics Journey

00:29 Introduction to 800V DC Architecture and Its Significance

02:19 Why 800V? Economics, Semiconductor Ratings, and EV Influence

04:47 Physics and Economics Driving the 800V Revolution

08:27 DG Matrix's Multi-Port SST and Its Value Proposition

11:33 Design Challenges and Innovations in Multi-Port SSTs

13:53 Current State and Adoption of 800V DC in Data Centers

17:51 Future Data Center Architectures and Power Density Trends

22:08 Adoption Curve and Market Penetration of 800V DC

26:34 Risks, Challenges, and Future Proofing of Data Centers

30:36 Cybersecurity and Power System Resilience

38:04 The Broader Impact of Power Innovations on Society

Justin Lebar (jlebar.com) recently spent $10,000 in an afternoon, uncovering critical miscompiles across NVIDIA's PTXAS, LLVM's AMD GPU, and X86 backends. He joins Jordan Nanos (@JordanNanos) to detail his methodology, which combined traditional fuzzing techniques with novel LLM-assisted bug finding. Their discussion highlights the unique challenges of detecting flaws in less-tested ML compilers compared to mature CPU environments.Lebar shares specific high-severity X86 findings, including an atomic operation bug that splits into two non-atomic operations. They explore the comparative efficacy of fuzzing versus LLM agents in identifying these elusive errors. This episode offers critical insights into compiler security and the burgeoning role of AI in automating rigorous code verification for AI infrastructure.FULL ARTICLE

00:00 Introduction and Content Overview00:25 Justin Lebar's Background and Recent Project00:59 Fuzzing Techniques for Compiler Bugs01:56 Motivation Behind the Project02:48 Challenges in Bug Detection in GPU and ML Compilers04:13 Bug Severity and Findings in AMD and x8605:38 Using LLMs to Read and Find Bugs in Code07:56 Impact of New Models and UltraCode Mode12:18 Estimating Time and Effort Without AI Assistance14:22 Limitations of Manual Code Review for Bugs15:03 Optimism About AI in Software Development16:17 Next Steps and Future Projects18:11 Key Takeaways for Developers and Researchers21:48 Call for Community Engagement and Scientific Approach

AWS operating margins jumped 10 percentage points while Microsoft Azure and Google Cloud stayed flat. The driver: Anthropic's Claude usage routing through Bedrock, Amazon's token-as-a-service platform.

Jordan Nanos (@JordanNanos), Jeremie Eliahou Ontiveros (@JeremieEO), Joey Brookhart (@SaasquatchC), and Crystal Huang (@Egg1459) break down why stabilized token margins are fundamentally richer than GPU-as-a-service for hyperscalers. The crew analyzes Anthropic's recent $65B Series H raise, Claude Opus 4.8 release, and SpaceX partnership against the backdrop of 300+ neo clouds fragmenting the traditional cloud moat.The team forecasts how AWS's workload mix advantage creates sustainable returns while competitors struggle with asset-heavy GPU service models. They examine the $22.7T TAM question, earnings-before-training dynamics, and whether the 2026 AI infrastructure beat belongs to silicon vendors or platform integrators. Subscribe for weekly deep dives into semiconductor and AI infrastructure economics.00:00 Intro: Episode 13 and the AWS margins article00:56 What is Bedrock? The three hyperscaler buckets02:33 AWS margins rising while peers lag03:33 Cloud moats collapsing and the neo cloud explosion06:32 Why stabilized token-as-a-service margins are so rich09:54 Amazon's workload mix advantage12:41 Forecasting Anthropic and the 4.8 release16:33 The SpaceX deal and the $65B Series H raise19:30 Bullish or bearish? Demand becoming supply28:55 The $22.7T TAM and does the race even matter31:59 Earnings before training and open-ended TAM36:27 The 2026 beat is basically one company40:22 Who wins long term: silicon, partnerships, integration

Jordan Nanos (@JordanNanos), Howie, and Myron Xie break down the economics of Cerebras's IPO on the eve of their public debut, examining their OpenAI and Amazon deals that have shifted the company away from Middle Eastern investor concentration toward frontier AI labs willing to pay exponentially more for speed.The discussion covers Cerebras's radical stitching innovation across a full wafer, creating compute density equivalent to an entire NVL72 rack without off-chip data movement. The hosts analyze whether businesses will accept these premium economics as fast tokens become the new standard for interactive AI applications.Subscribe for weekly deep dives into semiconductor economics and AI infrastructure developments.

(00:00) Cerebras IPO Preview

(18:58) Need for Speed Fast Tokens

(21:19) Wafer Scale Engine Architecture

(25:31) Radical Stitching Innovation

(31:55) Power Delivery and Cooling

(34:31) Bandwidth and IO Limitations

(37:12) Scaling Beyond Wafer Size

(40:05) Manufacturing and Assembly Bottlenecks

(42:54) Data Center Service Model

OpenAI was in serious trouble at the beginning of this year. Anthropic's Claude Opus 4.5 release had triggered a wave of developers to start using Claude Code, pushing Anthropic's revenue past OpenAI's on a like-for-like basis by April. OpenAI's GPT 5.4 response was such an embarrassment they didn't even compare it to Claude in their model release card. Then came GPT 5.5 - finally back on the frontier, but is it enough to reclaim the crown?

Jordan Nanos (@JordanNanos), Dylan Patel (@Dylan522p), Doug O'Laughlin (@FabricatedKnowledge), and Max Kan (@maxkan_) break down the latest AI model wars, from Claude 4.7's coding dominance to DeepSeek's long-delayed v4 release and what it reveals about China's AI capabilities. They analyze token efficiency, benchmark gaming, and why fast mode might be fake news. Subscribe for weekly deep dives into the semiconductor and AI infrastructure powering the future.

The Coding Assistant Breakdown

AI Value Capture

Timestamps:

00:00 OpenAI's Comeback and the Latest AI Model Wars

04:05 The High Cost of AI Models and Fast Mode Effectiveness

08:16 When AI Tokens Become Too Expensive for Tasks

13:11 Why AI Model Quality Degrades and Benchmarks Fail

18:42 Deep Dive into Claude 4.7 Features and Tokenizer Changes

25:29 DeepSeek's Release and China's AI Compute Constraints

28:20 The Future of Context Windows and Agent Orchestration

30:47 The Great Debate: CLI vs. App for AI Interaction

36:33 Debunking AI Fake News and Context Window Limitations

40:51 The AI Race: China, Meta, and the Neo Cloud Vision

43:46 Final Thoughts and Listener Feedback Request

This episode features Jordan Nanos (@JordanNanos) and Daniel Nishball (@dnishball) breaking down the economics of GPU clusters through real-world data and experience. Joined with Kang Wen Cheang and Zane Fong, the team discussed moving beyond theoretical TCO models as they examine how reliability differences between top-tier and lower-tier providers create significant cost disparities that aren't captured in simple per-GPU pricing. The discussion introduces practical frameworks for measuring goodput and understanding how system failures cascade through entire training jobs.Nanos walks through the mechanics of fault-tolerant frameworks including AWS's Checkpointless Training and explains why a single GPU failure can halt progress across hundreds of nodes. The conversation reveals how hyperscalers and NeoClouds price their services and why paying premium rates for reliable infrastructure often delivers better value than chasing the lowest per-hour costs. Subscribe to SemiAnalysis for in-depth analysis of AI hardware economics and infrastructure trends that impact the entire semiconductor ecosystem.

This week the team from ChipBook (formerly Chips & Wafers) joins. Jordan Nanos talks with Chaim Eisenberg and Simi Sherman as they explore how they build the ChipBook with open source data, and how that drives investment decisions. This episode dives deep into the ChipBook itself, revealing how granular, historical data collection provides insights into supply chain dynamics, memory markets, wafer fabrication equipment trends and more. The guests also share compelling examples of how their data-driven approach has generated some viral social media recently.for more: SemiAnalysis.com/chipbook00:00 The Chipbook: Understanding Open Source Data07:56 Granularity in Data: The Key to Investment Insights09:11 Understanding the Semiconductor Supply Chain10:52 Memory Market Insights and Trends14:23 Tracking WFE and Its Impact on Production15:20 The Importance of Early Signals in Investment16:51 Geopolitical Implications on Semiconductor Supply20:24 The Impact of Tariffs and Regulations22:48 Granular Tracking for Investment Decisions27:07 Data-Driven Insights and Investment Strategies29:13 The Structure of the Chipbook33:02 Collaboration and Integration at Semi Analysis37:02 Geopolitical Analysis and Its Impact on TSMC43:48 Helium Supply Chain and Its Importance