Jensen Huang, CEO of NVIDIA, discusses the company's transformation from a GPU manufacturer to the architect of AI infrastructure powering the global artificial intelligence revolution. As leader of what has become the world's most valuable company, Jensen oversees extreme co-design spanning everything from individual chips to entire data center pods.

The conversation explores NVIDIA's unique organizational structure with Jensen's 60-person direct staff enabling real-time collaboration across all engineering disciplines. Jensen explains how the company manifests future technology through systematic belief-shaping across employees, partners, and the broader industry ecosystem.

Key topics include the four scaling laws driving AI advancement, NVIDIA's evolution toward building AI factories rather than individual computers, and Jensen's perspective on achieving AGI through agentic systems. The discussion also covers supply chain orchestration, power infrastructure challenges, and Jensen's philosophy on leadership through extreme co-design principles.

Extreme Co-Design: From Chips to Data Centers

NVIDIA has moved beyond GPU optimization to extreme co-design of entire systems including 'GPU, CPU, memory, networking, storage, power, cooling, software, the rack itself, the pod, and even the data center.'

The shift became necessary because 'the problem no longer fits inside one computer to be accelerated by one GPU' - distributed computing at massive scale requires solving networking, switching, and workload distribution simultaneously.

Jensen's 60-person direct staff includes experts across all disciplines who collaborate in real-time: 'We present a problem and all of us attack it... because we're doing extreme code design.'

The Blackwell rack contains 1.3 million components and 1,300 chips weighing 4,000 pounds, while the Rubin pod includes 1.2 quadrillion transistors across 40 racks.

The Four Scaling Laws Driving AI Growth

Pre-training scaling initially faced data limitations until synthetic data generation solved the bottleneck: 'most of the data that we are training... is synthetic because it didn't come out of nature.'

Test-time scaling proves 'inference is thinking' and 'thinking is way harder than reading' - contradicting early predictions that inference would be simple and cheap.

Agentic scaling enables AI multiplication: 'It's so much easier to scale NVIDIA by hiring more employees than it is to scale myself... we could spin off agents as fast as you want.'

The scaling cycle creates a feedback loop where 'agents and agentic systems... create a lot more data' that feeds back into pre-training and post-training improvements.

CUDA Install Base as NVIDIA's Primary Moat

CUDA's 20-year install base across millions of developers represents NVIDIA's strongest competitive advantage: 'Install base is everything. Install base defines an architecture.'

The 2013 decision to put CUDA on GeForce GPUs 'completely consumed all of the company's gross profit dollars' and dropped market cap from $8 billion to $1.5 billion.

Developer trust in NVIDIA's long-term commitment creates a virtuous cycle: 'if I develop it on CUDA, I reach a few hundred million computers... I trust 100% that NVIDIA is going to keep CUDA around.'

CUDA's ecosystem spans 'every cloud... every single industry... every single country' from Google Cloud to radio base stations to satellites in space.

Power Infrastructure and Grid Optimization

Current power grids run at only '60% of peak' most of the time, designed for worst-case scenarios that occur just 'a few days in winter, a few days in summer.'

Data centers could utilize excess grid capacity through graceful degradation: 'when they need the maximum power for infrastructure... the data centers would get less.'

The solution requires three-way coordination between end customers accepting occasional service degradation, data centers built for dynamic power allocation, and utilities offering tiered power guarantees.

NVIDIA pushes 'tokens per second, per watt' efficiency improvements 'orders of magnitude, every single year' to address power constraints through extreme co-design.

AGI Achievement and Job Market Transformation

Jensen believes 'we've achieved AGI' based on AI's ability to potentially create billion-dollar companies through viral applications, similar to internet-era successes.

The radiologist example proves job displacement fears wrong: despite superhuman computer vision since 2019, 'the number of radiologists grew' due to increased diagnostic capacity.

AI will expand coding from '30 million to probably 1 billion' people by enabling specification through natural language: 'every carpenter in the future will be a coder.'

Job transformation focuses on purpose over tasks: 'the purpose of a software engineer and the task of a software engineer for coding are related, not the same.'

Token Factories and NVIDIA's $10 Trillion Path

Computing has transformed from 'a retrieval-based file retrieval' system to 'contextually aware' generation requiring 'a lot more processing in this new world.'

AI factories generate revenue-producing tokens with premium pricing: 'somebody's willing to pay a thousand dollars per million tokens is just around the corner.'

NVIDIA's growth potential stems from token factory proliferation: 'how many of these factories does the world need? How many tokens does the world need?'

The path to $3 trillion revenue appears feasible because 'it's not limited by any physical limits' and leverages a 200-company supply chain ecosystem.