GPU Infrastructure
From a 1,000 GPU B300 cluster to H100 capacity available now, KONST draws on its own data center deployment experience across Taiwan, Asia, and Europe. We help you choose the right location, configuration, and commercial terms.
3,856
GPUs delivered
363
Compute nodes
2.20
Starting rate in USD/hr/GPU
Choose a site based on training scale, data residency, and latency requirements
GPU model
Location / code
Scale
GPU memory
GPU Fabric
Cooling
USD / hr / GPU
NVIDIA B300
AI training · Inference · HPC
Czech Republic
CZ-Brno-B300
128 nodes · 1,024 GPUs
Standard B300 configuration
800Gbps
Direct liquid cooling
From $4.25
5 year term
NVIDIA H100 80G
AI training · Inference · LLM
New Taipei
TW-Linkou-H100
3 nodes · 24 GPUs
80GB per GPU · 1.92TB total
Optional 400G
Air cooling
From $2.30
1 or 3 year term
NVIDIA H100
AI training · Inference · HPC
Chiba, Japan
JP-Chiba-H100
3 nodes · 24 GPUs
80GB per GPU · 640GB per node
400G Ethernet
Air cooling
From $2.20
36 month term
NVIDIA H200
Training · Inference · Rendering · CAD
Sweden
SE-Stockholm-H200
37 nodes · 296 GPUs
141GB per GPU · About 41.7TB total
3x ConnectX-7
Air cooling
From $2.88
12, 24, or 36 month term
NVIDIA B300
AI training · HPC
Sydney, Australia
AU-Sydney-B300
64 nodes · 512 GPUs
Standard B300 configuration
CX8 x8 embedded
To be confirmed
From $5.20
5 year term
NVIDIA B300 SXM
LLM pretraining · Inference · HPC
Portugal
PT-B300
128 nodes · 1,024 GPUs
288GB per GPU · 288TB total
800Gbps RoCE v2
To be confirmed
From $4.55
3 or 5 year term
NVIDIA B300
AI training · Inference · HPC
From $4.25
USD/hr/GPU
NVIDIA H100 80G
AI training · Inference · LLM
From $2.30
USD/hr/GPU
NVIDIA H100
AI training · Inference · HPC
From $2.20
USD/hr/GPU
NVIDIA H200
Training · Inference · Rendering · CAD
From $2.88
USD/hr/GPU
NVIDIA B300
AI training · HPC
From $5.20
USD/hr/GPU
NVIDIA B300 SXM
LLM pretraining · Inference · HPC
From $4.55
USD/hr/GPU
Rates are reference starting prices per GPU hour and vary by contract term, memory configuration, and payment structure. The city, data center name, tier rating, or contracting entity for some sites is still being confirmed and will be disclosed in the contract.
Side by side
Select two sites to compare specifications and commercial terms
NVIDIA B300
From $4.25USD/hr/GPU
NVIDIA H100 80G
From $2.30USD/hr/GPU
Location
Czech Republic
New Taipei
Location / code
CZ-Brno-B300
TW-Linkou-H100
Scale
128 nodes · 1,024 GPUs
3 nodes · 24 GPUs
GPU memory
Standard B300 configuration
80GB per GPU · 1.92TB total
CPU
Dual Intel 6776P
2x AMD EPYC 9654
System memory
DDR5 3TB / 4TB
1,024GB DDR5
GPU Fabric
800Gbps
Optional 400G
Cooling
Direct liquid cooling
Air cooling
Contract term
5 years
1 or 3 years
Availability
About 7 months after signing
Available now
Primary workloads
AI training · Inference · HPC
AI training · Inference · LLM
Capability
01
We compare GPU Fabric, cooling, and expansion limits across six data centers against your training scale, data residency, and latency needs, then recommend a practical deployment plan.
02
We arrange hardware trials and site visits. Some proof of concept fees can be credited after contract signing, with options ranging from delivery in 7 days to a 1,000 GPU cluster in T+7 months.
03
We provide 24/7 monitoring and tiered support, with a 4 hour response and 24 hour replacement for P0 incidents, an SLA of up to 99.5%, and warranties of up to 3 years.
Process
Each stage has clear deliverables and timing, so budget and engineering schedules can be planned together.
Discovery
Confirm model scale, data residency requirements, and budget, then shortlist 1 to 2 data centers.
Proof of concept
Test an 8 GPU node or visit the site to validate throughput, networking, and operating procedures.
Contract
Plan the contract term and payment structure around cash flow. Payment terms are assessed for each project.
Launch
After deployment and acceptance, the environment moves into 24/7 monitoring and its warranty and SLA take effect.
News
Recent coverage and company updates from KONST.
Aug 2026
Commercial Times
ADATA invests another NT$386 million in KONST and raises its stake to 12.37%
Aug 2026
Economic Daily News
KONST identifies fragmentation as the real AI compute bottleneck at Ai4
Aug 2026
Data Center Dynamics
KONST signs an Indonesian data center lease with LG Sinar Mas
Aug 2026
Commercial Times
KONST signs an SMX01 data center agreement with SM+ Data Centers and LG Sinar Mas
Jul 2026
Commercial Times
KONST joins Taipei City at the 2026 Taiwan Expo in Japan
Whether you need a custom hybrid colocation setup or large scale Bare Metal scheduling,
KONST specialists can help you clarify the requirements.
