AI-ready infrastructure • 2026

Data center infrastructure built for GPU servers

Host your GPU cluster in FPT's Tier III data centers in Vietnam — high-density power, multi-layer redundancy, carrier-neutral connectivity and a 24/7 operations team. You focus on training models; we take care of the infrastructure.

Tier III / TIA-942 Rated 3 4–15 kW/rack, scalable on demand Carrier-neutral
4Data centers
Hanoi & Ho Chi Minh City
3.600+Racks at
FPT Fornix HCM02
4–15 kWFlexible power per rack
(expandable further)
99,982%Minimum uptime
by design
24/7Monitoring &
Remote Hands
Why a dedicated DC

GPU servers outgrow traditional server rooms

A single GPU server can draw from a few kW to more than 10 kW — many times a conventional server. Office server rooms and legacy facilities struggle with power, cooling and networking.

⚠️ In-house server room

  • Limited power, hard to upgrade as you add GPUs
  • Cooling can't keep up, GPUs throttle
  • Single power feed, single carrier — high outage risk
  • Heavy capex for UPS, generators and fire suppression
  • No 24/7 staff when incidents happen at midnight

✅ GPU colocation at FPT Data Center

  • Flexible 4–15 kW/rack, ready to scale as your cluster grows
  • N+1, N+2 and 2N redundancy — concurrent maintainability
  • Carrier-neutral: choose your carriers, optimize bandwidth cost
  • Turn capex into opex, go live within weeks
  • FPT engineers on site 24/7/365
Infrastructure

Everything your GPU cluster needs to run at full power

⚡

High-density power

Dual A/B feeds, high-capacity PDUs, flexible 4–15 kW/rack with room to go higher for dense GPU configurations.

❄️

Purpose-built cooling

Hot/cold aisle containment, with rack layout and cooling designed around each cluster's real power density.

🔋

Multi-layer redundancy

N+1, N+2 and 2N architecture for power, UPS and generators keeps systems running even during maintenance.

🌐

Neutral connectivity

Carrier-neutral model with high-speed cross-connects to link your GPU cluster with clouds and offices.

🛡️

Layered security

Multi-layer access control, 24/7 CCTV, and private cages/suites for customers who need physical isolation.

🧑‍🔧

Remote Hands 24/7

On-site engineers handle installation, cabling, part swaps and reboots — no need for you to be there.

Plans by power density

Choose the right power level for your GPUs

From a few inference servers to large training clusters — per rack, cage or private suite, ready to scale as demand grows.

Standard
4–6 kW/rack
Inference, rendering, AI pilots
  • PCIe GPU servers (1–4 GPUs) plus standard servers
  • Dual A/B power
  • Flexible bandwidth
  • Basic Remote Hands
Get a quote
Popular for AI companies Performance
6–10 kW/rack
Mid-size training, fine-tuning
  • 1–2 GPU servers (4–8 GPUs) per rack
  • High-capacity PDUs, real-time power monitoring
  • High-speed inter-rack cabling
  • Placement in suitable cooling zones
Talk to us
High-Density
> 10 kW/rack
8×GPU servers, large training clusters
  • Custom design, expandable beyond 15 kW/rack
  • Private cage / suite for isolation
  • Dedicated engineer
  • Custom enterprise SLA
Request a design

* Plans are indicative. Configuration, location and pricing are quoted based on your equipment and contract term.

FPT Data Center network

4 data centers in Vietnam's two largest cities

Deploy in Ho Chi Minh City or Hanoi, or combine both for disaster recovery (DR) of your data and AI models.

FPT FORNIX HCM02 • SAIGON HI-TECH PARK

Flagship data center for GPUs

Opened in August 2025: 8 floors, designed for AI and cloud infrastructure with high power density.

3.600racks
10.000 m²total floor area
4–15+ kWflexible power/rack
HCM01

Ho Chi Minh City

District 7 — operated to Tier III standards, well connected across southern Vietnam.

HN01 • HN02

Hanoi

Two data centers serving northern Vietnam, ideal for active-active or DR deployments.

Standards & certifications

Built to demanding international standards

Uptime Tier IIIDesign & operations
ANSI/TIA-942 Rated 3DC infrastructure
ISO 9001Quality management
ISO 27001 / 27017 / 27018Information & cloud security
ISO 22301Business continuity
PCI DSSPayment card data security
SOC 2Service controls
TVRARisk assessment
LEEDGreen design (HCM02)
Who it's for

Typical customers

🤖

AI startups & labs

Own GPUs for model training and need a stable home with predictable costs instead of long-term cloud rental.

🏦

Banking & finance

Run AI on-premise with data stored in Vietnam, meeting compliance and audit requirements.

🏭

Enterprises & GPU cloud providers

Scale compute capacity quickly and deploy large GPU clusters in Vietnam.

Onboarding process

From first call to production in weeks

Requirements

Node count, GPU type, power draw, network and storage needs.

Design & quote

Rack placement, power and cooling plan, and a detailed quote.

Contract & SLA

Clear service commitments with flexible terms.

Install & connect

Equipment receiving, rack & stack, cabling, power and network testing.

24/7 operations

Continuous monitoring, Remote Hands and always-on technical support.

FAQ

Questions you may have

How is GPU colocation different from GPU cloud?

With colocation you own the GPU servers and host them at FPT Data Center; FPT provides space, power, cooling, network and operations. It suits continuous, long-term GPU use where you want full control of the hardware. For short-term flexibility, FPT also offers GPU Cloud services you can combine.

How many 8×GPU servers fit in one rack?

It depends on each server's configuration and actual power draw — an 8×GPU server typically needs more than 10 kW. Our engineers will calculate from your equipment specs and recommend the optimal nodes per rack with sufficient power and cooling.

How is pricing calculated?

Typically by number of racks (or cage area), committed power (kW), bandwidth, cross-connects and add-on services. Contact us for an exact quote for your configuration.

Can my team access the data center?

Yes. You register authorized personnel under our security procedures. Our 24/7 Remote Hands service can also handle physical tasks so you don't have to be on site.

How long does deployment take?

For standard configurations with available space, installation can be completed within a few weeks of signing. Very high-density cage/suite projects follow a dedicated timeline.

Knowledge blog

Insights on GPU & data center infrastructure

Practical notes on colocation, power, cooling, AI chips and industry trends.

View all articles →
Contact

Ready to bring your GPU cluster online?

QD
Đỗ Quốc DuySolutions Consultant, FPT Data Center

Request a quote

This opens your email app with a pre-filled message to DuyDQ12@fpt.com.

Zalo number +84 917 755 725 copied. Open Zalo → Add friend → paste the number.
📞 Call: +84 917 755 725