The AI POV — All Articles
Why Grace Blackwell and Rubin Multiply Revenue Capacity Across Every Token Tier
Grace Blackwell increased tier performance by 35× and added a new tier; the highest-value tier gained 10×. Allocating power 25/25/25/25…
Tech DeskHow Nvidia and Groq LP300 Plus Dynamo Unlock 35× on the Highest-Value Inference Tier
Nvidia combined Vera Rubin GPUs with Groq LP300 processors and a new Dynamo layer: prefill runs on Rubin, decode on…
Tech DeskInside Vera Rubin Ultra: Liquid-Cooled Racks for the Next Generation of AI Factories
Vera Rubin Ultra turns AI racks into fully liquid-cooled, hot-water systems where 144 GPUs share a single NVLink domain. Kyber…
Tech DeskHow Token Pricing Tiers Will Reshape the AI Economy
Jensen Huang’s GTC keynote sketches a future where AI tokens are sold like a tiered commodity, from free and $3…
Tech DeskInside the AI Token Factory: Why Tokens Became the New Commodity of Computing
Token production has exploded, turning modern AI data centres into power-constrained factories where tokens-per-second and cost-per-token define success. Nvidia’s GTC…
Tech DeskFrom DGX-1 to Rubin: How Nvidia Turned Data Centres into AI Factories
In a decade, Nvidia’s AI systems have evolved from the DGX‑1 deep learning box to Rubin-era AI factories that treat…
Tech Desk“This Is the Beginning of Something Very, Very Big”: Nvidia’s Jensen Huang on AI-Native Companies
Nvidia CEO Jensen Huang used his GTC keynote to describe a rapidly growing wave of AI-native companies, arguing that they…
Tech DeskFrom Retrieval to Generation: How ChatGPT Marked the Start of Nvidia’s Generative AI Era
Jensen Huang used his GTC keynote to argue that ChatGPT marked the start of a generative AI era that is…
Tech DeskFrom Perception to Agentic AI: How Reasoning and Coding Agents Changed the Game
OpenAI o1 brought step-by-step reasoning; Claude Code and other agentic systems let AI read files, run tests and iterate. That…
Tech DeskThe Inference Inflection Point: Why AI Computing Demand Grew a Million Times in Two Years
Over two years, compute per task grew ~10,000x and usage ~100x — total demand effectively grew about a million times.…
Tech Desk