1. NVIDIA GTC Day 1: Jensen Huang’s "Agentic" Vision

NVIDIA CEO Jensen Huang took the stage this morning to redefine the AI stack, shifting the focus from model training to Inference and Agency.

• The "LPU" Reveal: In a major move to own the "production" side of AI, NVIDIA unveiled a new chip architecture specifically optimized for inference—the process of running AI in real-time. This includes a strategic partnership with Groq to license their "Language Processing Unit" (LPU) technology, designed to power near-instantaneous agentic responses.

• Photonics Investment: NVIDIA announced a combined $4 billion investment in U.S. photonics leaders Coherent and Lumentum. The goal is to replace traditional electrical wires with light-based communication, slashing the power consumption of AI data centers by half.

• Agentic AI Framework: NVIDIA introduced its Agent Platform Architecture, a software "five-layer cake" that allows companies to build agents that don't just chat, but plan, act, and use tools autonomously across entire enterprise workflows.

2. NC AI Challenges the Giants with "World Foundation Model" (WFM)

While NVIDIA dominates the hardware, South Korean firm NC AI threw down a major challenge today by demonstrating its World Foundation Model (WFM).

• The "Sim-to-Real" Fix: The biggest problem in robotics is "hallucination" in the real world. NC AI’s WFM perfectly predicts physical laws (like friction and gravity), achieving an 80% task success rate while using only 25% of the GPU resources required by models like Google’s Geni3.

• K-Physical AI Alliance: NC AI plans to use this efficiency to build specialized robot ecosystems for semiconductor cleanrooms and shipyards, moving robotics out of the lab and into heavy industry.

3. OpenAI Korea Pushes Back on "Claude Surge"

A rare public spat broke out today in Seoul as the head of OpenAI Korea disputed reports that Anthropic’s Claude is overtaking ChatGPT in the enterprise market.

• The Data Dispute: Recent card spending data suggested Claude now holds 30% of the Korean market. OpenAI’s Kim Kyoung-hoon called the data "misleading," arguing that large businesses pay via invoice rather than corporate cards, masking ChatGPT’s continued dominance.

• Anthropic’s Momentum: Regardless of the payment method, South Korea now ranks 7th globally in Claude usage, and the government is currently in talks with Anthropic CEO Dario Amodei to sign a formal Memorandum of Understanding (MoU) for public service AI.

4. Tech Spotlight: The "Future of Work" Forum

While CEOs talk chips, global leaders met in Riga today for the "Future of Work in the Age of AI" forum.

• The Productivity Paradox: The World Economic Forum’s Saadia Zahidi warned that while AI is driving massive productivity gains, there is no guarantee these benefits will be shared by the workforce.

• Job Layoffs: This coincides with Australian giant Atlassian announcing a 10% staff reduction today, fueling a growing debate over whether AI should lead to a permanent reduction in standard working hours.

Prompt Tip of the Day: The "Agentic Architect" — Inference Optimizer

With NVIDIA’s focus today on Inference (running AI efficiently), today’s prompt helps you act as a "Developer Lead" to audit and optimize your own AI interactions for speed and accuracy.

The Prompt:

"act as a professional chief ai architect specializing in inference optimization. i want to audit my current use of [insert ai tool, e.g., chatgpt or claude]. please structure a framework for this agent that includes:

• token efficiency module: instructions for the agent to rewrite my last 3 prompts to be 30% shorter without losing any critical context or instruction.

• reasoning path audit: a rule that requires the ai to provide a 'one-sentence logic summary' before every long response, so i can catch logic errors early (the 'inference check').

• cost-performance scorecard: a template for the agent to compare the speed and quality of a 'large' model (like gpt-4) vs. a 'small' model (like gpt-4o-mini) for my specific repetitive tasks.

• latency guardrail: a step-by-step process for the agent to identify 'redundant' steps in my workflow that could be combined into a single, high-speed 'agentic' command.

for each point, provide clear, step-by-step rules that would allow an ai agent to operate as a professional and high-efficiency technical consultant."