1 / 6
All-In Podcast × GTC 2026

NVIDIA's Playbook
for the AI Era

Jensen Huang's vision for inference, agents, physical AI, and the $1T opportunity ahead

Jensen Huang on the All-In Podcast at GTC 2026
The 4 Big Bets
1M×
Inference Explosion
Compute demand grew 1,000,000x in 2 years. Inference has overtaken training as the primary revenue driver. Jensen predicts another billion-fold increase ahead.
100:1
Rise of the Agent
100 AI agents per engineer at NVIDIA (7.5M agents for 75K humans). Every $500K engineer should consume $50K in tokens annually or Jensen will "go ape."
$50T
Physical AI
The largest technology market in history, "largely void of technology until now." Three computers: training, simulation (Omniverse), and edge robotics.
$50B
AI Factories
Datacenters reimagined as industrial token producers. NVIDIA's 5-layer stack from chips to applications. $194B data center revenue in FY2026.
$194B
Data center revenue FY2026
86%
AI chip market share
$20B
Groq acquisition
35×
Faster token delivery
"Most analysts just don't understand the scale and breadth of AI." — Jensen Huang
2 / 6
THE BIG SHIFT
The Inference Explosion
1
Generative AI
Baseline compute
100x
2
Reasoning AI
Chain-of-thought, multi-step logic
100x
3
Agentic AI
Tools, memory, collaboration
= 1,000,000x in 2 years
WHY IT MATTERS
DATA CENTER REVENUE
$194B
NVIDIA data center revenue FY2026 (91% of total)
INFLECTION POINT
Late 2025
Inference surpassed training in total data center revenue
Jensen predicts inference demand will grow another 1 billion times as every application, agent, and device generates tokens continuously. The era of "compute and generate" is replacing "search and retrieve."
3 / 6
The Agentic Era
Rise of the AI Agent
The $50K Token Budget
"If that $500,000 engineer did not consume at least $50,000 worth of tokens, I'm going to be deeply alarmed."
— Jensen Huang
NVIDIA is targeting $2B total spend on tokens for its engineers
Agents by the Numbers
100:1
Agent-to-human ratio at NVIDIA (7.5M agents, 75K employees)
62%
Of organizations experimenting with AI agents (McKinsey, Nov 2025)
$1T
Projected agent & model company revenue by 2030
What Changes for Enterprise Software
Before Agents After Agents
Revenue Model Per-seat licensing Per-token consumption
Usage Pattern 8 hours × human users 24/7 × hundreds of agents
Constraint Butts in seats Token budget
Software Role Tool for humans API for agents
Market Size Limited by headcount 100× more concurrent users

"Every enterprise software company will become a value-added reseller of AI tokens." — Jensen Huang

4 / 6
Beyond the Screen
Physical AI & Robotics
The Three Computers
1
Training Computer
Develops AI models. Massive GPU clusters processing data to learn patterns and capabilities.
2
Omniverse (Simulation)
Evaluates AI in virtual worlds. Robots, cars, and machines tested against physics before real deployment.
3
Edge / Robotics Computer
Runs AI in the real world. Embedded in robots, vehicles, and factory equipment for real-time decisions.
The AI Development Loop
$50T
Physical AI represents the largest technology market in history
  • Manufacturing, logistics, healthcare, agriculture -- industries largely void of AI technology until now
  • The "GPT moment for the bots" -- AI moves from digital screens into physical machines
  • NVIDIA building simulation and training platforms to secure its edge in the next decade
Real-World Applications
250+
Autonomous Vehicles
Car companies using NVIDIA's full stack: Alpamayo reasoning for decision-making, Omniverse for simulation testing, and DRIVE Orin for in-vehicle compute. NVIDIA provides everything except the car itself.
$1.7T
Healthcare AI
Global pharma market being reshaped by AI-driven drug discovery and diagnostic agents. Partnerships with Hippocratic (clinical AI) and Open Evidence (medical reasoning) show the path.
$18B
Industrial Robotics
Warehouse robotics market by 2028. Factory automation with AI agents that perceive, reason, plan, and act in physical environments. The "GPT moment for the bots" starts here.
$30B+
Sovereign AI
Revenue tripled YoY from nations building their own AI infrastructure. UK, France, Netherlands, Canada, and Singapore leading. Every country wants domestic token production capability.
5 / 6
The New Infrastructure
AI Factories & Token Economics
The 5-Layer AI Factory Stack
5. Applications Revenue-generating end-user tools
4. Models Proprietary + open-source integration
3. Platforms CUDA X, AI Factory environments
2. Chips Blackwell, Vera Rubin hardware
1. Infrastructure Physical facilities and power
Disaggregated Inference
The Old Way
Monolithic GPU handles entire inference pipeline. One-size-fits-all approach. Limited optimization per task stage.
The NVIDIA Way
Pipeline split across specialized chips. GPUs for reasoning (prefill). Groq LPUs for delivery (decode). 35x faster token generation.
$20B Groq Acquisition • 35x Speed • 33-50% More Market
The Economics of Scale
$50B
Cost of a world-class AI factory datacenter
35x
Cost reduction with Vera technology vs previous gen
$95.2B
NVIDIA supply commitments (doubled in one quarter)
"A $50 billion datacenter is the most cost-effective solution for producing tokens due to extreme efficiency. These are not data centers. They are AI factories -- they produce tokens consumed worldwide."
-- Jensen Huang
6 / 6
YOUR PLAYBOOK
What This Means for Companies
1
Think in Tokens, Not Seats
Your pricing model is about to change. Per-seat licensing gives way to token-based consumption. Start modeling what your product costs per agent interaction, not per human user. A SaaS tool generating 10,000 agent API calls per day at $0.002/call costs $20/day per customer. Model this against your current $50/seat/month. The math will surprise you.
2
Design for Agent Users
Your next 100x growth in usage won't come from humans. Build APIs, not just UIs. Every feature needs to work for both human clicks and agent API calls. Start with your most-used features. Can an agent trigger a report generation via API? Can it read a dashboard without rendering a browser? Every workflow needs a headless path.
3
Budget for AI Like You Budget for Cloud
Jensen expects $50K in token spend per $500K engineer. If your team isn't consuming significant AI resources, you're underinvesting in productivity. Track your team's token consumption monthly. If engineers are spending less than $5K/month, they're doing manual work that AI could handle. Set token budgets like you set AWS budgets.
4
Physical AI Opens New Markets
The $50T physical AI market is emerging. If your product touches manufacturing, logistics, healthcare, or automotive, the competitive landscape is about to shift dramatically. Warehouse robotics alone is projected at $18B by 2028. If your product has a physical-world touchpoint, explore what an AI-powered version looks like.
5
Inference Cost Drives Roadmap
With 35x cost improvements per generation, features that were cost-prohibitive last year are viable now. Reassess what's possible quarterly. A feature that cost $1/query last year might cost $0.03 now. Build a quarterly 'cost feasibility review' into your roadmap process to catch these opportunities.
The Bottom Line
The Age of Agents Is Not Coming. It's Here.
NVIDIA is betting its entire strategy on a world where AI agents outnumber humans 100:1 in the enterprise. Companies that adapt their products, pricing, and platforms for this reality will capture the next wave. Those that don't will find their "butts-in-seats" model obsolete.

Source: All-In Podcast x GTC 2026 -- Jensen Huang interview with Chamath, Jason, Sacks & Friedberg

Rizvi Haider by Rizvi Haider