NVIDIA & The State of AI
Q4 FY26 — Reading the AI Supercycle Through Every Layer
The most granular real-time readout on the AI economy
The Numbers That Redefined the Quarter
Margins EXPANDING during the largest product ramp in computing history
Compute = Revenue
PRE-AI ERA
Compute = Cost Center
Minimize → $300-400B/yr IT CapEx
→
AI ERA
Compute = Revenue Engine
Maximize → $660-690B CapEx in 2026
Amazon $200B
Google $185B
Meta $135B
Microsoft $120B+
Oracle $50B
Cutting CapEx = Cutting Future Revenue
The Five-Layer Constraint Stack
1
ENERGY
19GW gap — the ultimate binding variable
2
SEMICONDUCTORS
TSMC at max capacity — 40-50 EUV machines/year
3
NETWORKING
$11B/quarter — now critical-path scarce
4
COOLING
Air cooling hitting physical limits
5
HUMAN EXPERTISE
~10,000 EUV technicians globally
The constraint isn't demand — it's supply. Every solution creates new bottlenecks.
The Chip Wars
NVIDIA
66% compute · 79% revenue
Google TPU
13% compute · 3x price-perf
Next Generation
Vera Rubin Platform — 6 Co-Designed Chips, Annual Architecture Cadence
No single competitor matches NVIDIA on any dimension
The Networking Moat
InfiniBand
(training)
Spectrum-X
(scale-out)
NVLink
(scale-up)
$11B/quarter. +267% YoY. The hidden lock-in nobody saw coming.
Each ring deepens switching costs for the next.
The Agentic Inflection
CHAT ERA
Tokens~100-500 / interaction
DurationSeconds
TriggerHuman-initiated
ScopeSingle turn
AGENTIC ERA
Tokens100,000+ / session
DurationMinutes to hours
TriggerSpawning sub-agents
ScopeMulti-step autonomous
Token demand goes exponential. Inference surpassed training in 2025.
Jensen: "The ChatGPT moment of agentic AI has arrived."
Three Demand Pillars
~50% of DC Revenue
Hyperscalers
$660-690B CapEx committed for 2026
Tripled YoY
Sovereign AI
$30B+ — 100+ nations as GPU customers
Fastest Growing
Enterprise + Model Makers
Anthropic 10x — capacity-constrained
Demand diversified. Even if one pillar weakens, the others sustain.
The $3-4 Trillion Question
The question isn't whether the world WANTS $3-4T in AI infrastructure.
It's whether the world can physically BUILD, POWER, and COOL it.
The largest infrastructure buildout in history.
The fastest technology adoption ever.
The widest revenue gap yet.
All at once. This is the state of AI in February 2026.
Source: The Business Engineer by FourWeekMBA