NVIDIA Corporation
The compute layer of the AI boom: GPUs plus the CUDA software moat make it the default platform every frontier model trains and runs on, at ~75% gross margins.
Business Overview
NVIDIA designs GPUs, networking, and rack-scale systems for AI. Q1-FY27 revenue was a record $81.6B (+85% YoY): Data Center $75.2B (92% of revenue) plus Edge $6.4B. Data Center now splits into Hyperscale (~$38B) and fast-growing ACIE — AI clouds, industrial, enterprise (~$37B, +31% QoQ).
Revenue Model
NVIDIA sells GPUs, networking, and rack-scale systems to hyperscalers, AI clouds, and enterprises at ~75% gross margin. The free CUDA software stack is the real lock-in — every major frontier model runs on it, making migration to rival silicon costly in both code and performance.
Key Metrics
- Gross Margin
- 75%
- Data Center Mix
- 92%
- Operating Margin
- 66%
- DC Networking Growth
- +199% YoY
Breakdowns
Q1 FY2027 Revenue Split ($B)
Data Center Sub-Markets ($B)
Competitive Moat
CUDA plus full-stack systems create switching costs rivals can't easily match: NVIDIA runs in every cloud and powers essentially every frontier model. AMD and custom ASICs chip at specific workloads but not the ~75%-margin platform.
Competitive Landscape
AMD
The closest GPU rival with rack-scale ambitions, but trails badly in software ecosystem, inference share, and installed base.
Custom ASICs (Google TPU, Amazon Trainium, Meta MTIA)
Cut some GPU purchases on narrow internal workloads, yet hyperscalers still depend on NVIDIA for general-purpose frontier training.
Intel
Competes in CPUs with Gaudi accelerators, but lacks comparable GPU throughput and CUDA-class software for large-scale AI.
Growth Drivers
+92% YoY
AI-factory buildout
Data Center revenue hit a record $75.2B as hyperscalers race to build AI factories; Q2 guided to $91B.
+31% QoQ
ACIE diversification
AI clouds, industrial and enterprise reached ~$37B, with AI-cloud revenue more than tripling YoY — demand broadening beyond hyperscalers.
Vera Rubin roadmap
Next-gen Vera Rubin, the agentic-AI Vera CPU, and Dynamo 1.0 software (7x inference) extend the platform and CUDA lock-in.
Risk Factors
Customer in-house silicon
Hyperscalers building custom ASICs (TPU, Trainium, MTIA) could curb GPU purchases on their largest internal workloads.
China & export controls
Q2 guidance assumes zero China data-center compute; even with US H200 licenses approved, no revenue has materialized and local rivals are gaining.
Concentration & cyclicality
A handful of hyperscalers drive most revenue, so any pause in AI capex or inventory digestion would hit hard.
Key Developments
May 2026
Q1 FY2027: record revenue $81.6B (+85% YoY), Data Center $75.2B (+92%), non-GAAP EPS $1.9; Q2 guided to $91B, excluding China.
Authorized an additional $80B buyback and hiked the quarterly dividend from $0 to $0.3; ~$20B returned in Q1.
Investor Takeaway
NVIDIA shows how a software moat (CUDA) turns a chip vendor into a platform: because every frontier model is built on its stack, hardware sales compound with switching costs. The lesson — the durable moat is the ecosystem, not the silicon.