Dylan Patel’s Information Machine

On a mid-March Monday morning, thousands of hardware geeks – computer science students, hyperscaler execs, accelerator architects – assembled under a formidable Jensen Huang. Welcome to the accelerator industry’s favorite conference, Nvidia GTC. A WWE-style championship belt was projected onto a 100-foot wall behind Huang. Mid-keynote, he pulled up Nvidia’s performance metrics. “This is from SemiA

On a mid-March Monday morning, thousands of hardware geeks – computer science students, hyperscaler execs, accelerator architects – assembled under a formidable Jensen Huang. Welcome to the accelerator industry’s favorite conference, Nvidia GTC. A WWE-style championship belt was projected onto a 100-foot wall behind Huang. Mid-keynote, he pulled up Nvidia’s performance metrics. “This is from SemiAnalysis. This is the largest, most comprehensive sweep of AI inference that has ever been done.” The chart, advertising 50x increased performance on Nvidia’s GB300 NVL72, depicted a decisive Nvidia…

related reading

• Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysisyoutube.com
• Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI computedwarkesh.com
• The Supply and Demand of AI Tokens | Dylan Patel Interviewyoutube.com
• The Short Case for Nvidia Stock | YouTube Transcript Optimizeryoutubetranscriptoptimizer.com
• SemiAnalysissemianalysis.com
• Who Owns the Generative AI Platform? | Andreessen Horowitza16z.com
• Generative AI’s Act Two | Sequoia Capitalsequoiacap.com
• The Economics of Generative AI: Two Years Laterapoorv03.com
• Nvidia Envy: understanding the GPU gold rushblog.johnluttig.com
• Who Owns the Generative AI Platform? | Andreessen Horowitza16z.com
• Power in the Age of Intelligencenotboring.co
• Inside the Inference Hardware Revolution Of 2026spectrum.ieee.org