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 Premium turnkey NVIDIA system. The buyer wants validated performance and support.
HGX Cloud-scale AI building block. The buyer wants customization and scale.
MGX Modular server reference architecture. The buyer wants flexibility and faster deployment.
EGX AI at the edge. The buyer wants low-latency inference outside big data centers.
DGX

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.

NVIDIA AI rack system

turnkey NVIDIA system

HGX

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.

Technicians integrating server racks

cloud-scale building block

MGX

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

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.