Working exhibit / NVIDIA server and platform types
DGX, HGX, MGX, EGX: who buys which NVIDIA server type?
These names are easy to blur together. The simple distinction is buyer intent: do they want a full Ferrari, a cloud-scale building block, a modular server architecture, or AI deployed outside the hyperscale data center?
DGX buyers
“Give me the full Ferrari.”
Typical buyers
- AI labs
- Governments
- Universities
- Large enterprises
- Biotech and pharma
- Sovereign AI initiatives
Examples
- OpenAI-style labs
- Tesla-style AI teams
- National labs
- Large banks
- Healthcare research centers
Why they buy
- Turnkey deployment
- Validated systems
- Fastest performance
- NVIDIA support
Strategic read
DGX buyers care less about custom hardware optimization. They pay for confidence, support, and speed.
turnkey NVIDIA system
HGX buyers
“Huge cloud-scale AI operators.”
Typical buyers
- Microsoft Azure
- Amazon AWS
- Google Cloud
- Meta
- Oracle Cloud
- CoreWeave, Lambda, neoclouds
Why they buy
- Customization
- Scale
- Cost optimization
- Own networking/control plane
What it is
The standard AI building block for hyperscalers and AI cloud operators.
Strategic read
This is probably the most important category for the Taiwan wedge because it maps to big rack-scale deployments.
cloud-scale building block
MGX buyers
“Flexible modular AI infrastructure.”
Typical buyers
- OEM server vendors
- Enterprises
- Regional cloud providers
- Telecom companies
- Industrial AI deployers
Examples
- Dell Technologies
- Hewlett Packard Enterprise
- Lenovo
- Telecom operators
- Medium-sized cloud companies
Why they buy
- Deploy faster
- Standardize designs
- Reduce engineering complexity
- Support multiple configurations
Strategic read
Think mass-customizable AI servers rather than one monolithic NVIDIA appliance.
modular server design
EGX buyers
“AI outside hyperscale data centers.”
Typical buyers
- Factories
- Hospitals
- Retailers
- Smart cities
- Telecom operators
- Airports and logistics hubs
Why they buy
- Low-latency AI
- Local inference
- Edge processing
- Computer vision
Example
A factory camera system detecting defects in real time. That is EGX territory.
Strategic read
Important, but usually not the 5-20MW Taiwan data center wedge. This is more distributed edge deployment.
edge inference site
For the Taiwan wedge
The most relevant buyer category is HGX-style infrastructure: hyperscalers, neoclouds, and sovereign AI buyers deploying rack-scale capacity.
Quick comparison
| Type | Buyer intent | Wedge relevance |
|---|---|---|
| DGX | Premium turnkey NVIDIA system | Useful for labs and sovereign programs; less about custom data center supply chains. |
| HGX | Cloud-scale AI building block | Highest relevance: needs land, power, cooling, ODM integration, and deployment path. |
| MGX | Flexible modular server infrastructure | Relevant for regional clouds, telcos, and enterprise infrastructure buyers. |
| EGX | Edge AI outside hyperscale data centers | Less relevant for 5-20MW AI factories; more relevant to factories and logistics hubs. |