Malaysia CyberjayaLive in 2027 Q1
128 production nodes1,024 GPU
288GB per GPUabout 294.9TB total
800G InfiniBand
Direct-to-chip liquid cooling
Monthly$6.00from5 year term Contact sales
SMX01 architectural renderingFLAGSHIP PROJECTJakarta CBD · SMX01
Jakarta CBD · SMX01
High-density liquid-cooled data center colocation
KONST Group leases 15 MW of liquid cooling capacity at SMX01, a joint venture between SM+ (part of Sinar Mas Group) and LG Sinar Mas, and provides the colocation services that high-density GPU deployments need.
01 / 04
Global infrastructureGlobal AIDC sites
Global AIDC sites
and deliverable models
Choose a site based on training scale, data residency requirements, and latency.
Asia-Pacific
Asia-Pacific5 data centersTaiwan · Japan · Australia · Malaysia
North America
North America5 data centersUnited States · Canada
Europe
Europe6 data centersSweden · Poland · Spain · France · Czechia
Select a region or site to filter plans. Locations are indicative, at country or city level.
Explore deliverable compute
Taiwan New TaipeiIn stock, immediate delivery7 to 14 days after the 3rd unit is signed
3 nodes24 GPU
80GB per GPU1.92TB total
Optional 400G
Air Cooling
Monthly$3.80from1/3 year term Contact sales
Japan ChibaDelivery within 6 weeks of signing
4 nodes32 GPU
80GB per GPU640GB per node
400G Ethernet
Air Cooling
Monthly$3.80from1/2/3 year term Contact sales
CzechiaLive March to June 2027
64 nodes512 GPU
288GB per GPUabout 147.5TB total
800Gbps
DLC liquid cooling
Monthly$6.50from3 year term Contact sales
SwedenPOC at end of November 2026Go-live to be determined
37 nodes296 GPU
141GB per GPUabout 41.7TB total
3x ConnectX-7
Air Cooling
Monthly$4.20from1/2/3 year term Contact sales
Australia SydneyLive in November 2026
64 nodes512 GPU
288GB per GPUabout 147.5TB total
CX8 ×8 embedded
To be confirmed
Monthly$6.80from5 year term Contact sales
PolandPhased go-live from January 2027All units in place in February
128 nodes1,024 GPU
288GB per GPUabout 294.9TB total
800G InfiniBand
Air Cooling
Monthly$6.00from3/5 year term Contact sales
Spain16 to 18 weeks after signing
128 nodes1,024 GPU
288GB per GPUabout 295TB total
800G InfiniBand
Air Cooling
Monthly$6.10from5 year term Contact sales
United States HoustonDelivery in February 2027
64 nodes512 GPU
288GB per GPU147.5TB total
800G RoCE
Air Cooling
Monthly$6.30from3/5 year term Contact sales
United States CaliforniaLive by mid-December 2026
32 nodes256 GPU
288GB per GPUabout 73.7TB total
800G RoCE
Air Cooling
Monthly$6.10from3 year term Contact sales
United States CaliforniaLive by mid-December 2026
64 nodes512 GPU
288GB per GPUabout 147.5TB total
800G RoCE
Air Cooling
Monthly$6.10from3 year term Contact sales
Canada TorontoDelivery 8 to 12 weeks after order
64 nodes512 GPU
288GB per GPUabout 147.5TB total
800G InfiniBand
Liquid cooling
Monthly$6.20from3/5 year term Contact sales
Canada TorontoDelivery 8 to 12 weeks after order
140 nodes1,120 GPU
288GB per GPUabout 322.6TB total
800G InfiniBand
Liquid cooling
Monthly$6.20from5 year term Contact sales
France Paris14 to 18 weeks after signing
50 nodes400 GPU
288GB per GPUabout 115TB total
800G InfiniBand
Air Cooling
Monthly$6.30from3/5 year term Contact sales
Europe16 to 18 weeks after signing
144 nodes1,152 GPU
288GB per GPUabout 332TB total
800G InfiniBand
Air Cooling
Monthly$6.10from5 year term Contact sales
Taiwan New TaipeiPre-saleSchedule to be determined
4 nodes32 GPU
288GB per GPUabout 9.2TB total
800G InfiniBand
Air Cooling
Monthly$6.00from3/5 year term Contact sales
Prices are reference starting prices per GPU per hour and vary with contract term, memory option, and payment structure. Actual terms follow the contract signed by both parties. Site and delivery information is disclosed per plan and does not reflect live inventory.
0 / 2 selected
FLAGSHIP PROJECTS
Flagship projects
High-density liquid-cooled data center colocation
GPU cluster compute services
OUR PERSPECTIVEEnterprises want
Enterprises want compute itself,
not the engineering work needed to get it
not the engineering work
AI now sits at the core of enterprise operations, making compute as fundamental as power and connectivity. Yet most companies still have to manage site selection, MEP systems, cooling, clusters, and operations themselves.
Specs do not match,
so high-density equipment
Each site must be checked against the equipment specs for per-rack power, air intake conditions, rack weight capacity, and power distribution type.
KONSTRA AIGPU Colocation
The workload is clear,
but data center and staffing
Building an entire data center for a defined workload does not justify the investment, and high-density machines also need a high-power data center and on-duty staff.
KONSTRA AICompute Rental
Equipment delivery does not mean
someone takes over daily operations
Someone must watch the alerts, someone must open an RMA and chase it until the parts arrive, and someone must isolate faulty nodes in the cluster network.
KONSTRA AIData Center Operations
Fragmented models:
switching models means changing code
Token usage and costs for each project are spread across each provider's dashboard, keys are scattered, and permission boundaries are unclear.
HORIZON AIATP Token
ONE CONNECTED ECOSYSTEM | Three brands
Three layers of capability, supported as one
Physical compute, cloud platforms, and enterprise AI adoption come together as one integrated ecosystem.
LAYER / 01Konstra AIInfrastructure
- AI data center build
- Mechanical, electrical, and cooling
- Monitoring systems
- GPU Cluster Deployment
LAYER / 02Glows.aiCompute and model deployment
- GPU Cloud Services
- Global compute sites
- AI Training Calculator
LAYER / 03Horizon AIPlatform governance and application rollout
- Enterprise AI Integration
- Horizon AI adoption service
- ATP Token
BUILT AROUND YOUR NEEDSFrom AIDC site selection to 24/7 operations
From AIDC site selection to 24/7 operations
Professional specifications, efficient delivery
Site selection and spec alignment
Based on training scale, data residency, and latency requirements, we compare each data center's GPU fabric, cooling method, and expansion limit, and recommend a deployment plan you can act on.
Explore AIDC sitesPOC to delivery
We arrange POCs on real hardware and site visits. At some data centers, the POC fee can be credited after signing. Plans range from delivery in as little as 7 days to thousand-GPU clusters at T+7 months.
Explore GPU clustersOperations and SLA
We provide 24×7 monitoring and tiered support, and set responsibility boundaries, service levels, and operations support for each data center and GPU cluster plan.
Learn about operations servicesPRODUCTS & SERVICES
Five service lines, one point of contact
Engage us for individual services or full-project delivery, from data center build to enterprise AI adoption.
FROM VISION TO OPERATIONFour-step adoption,
Requirements → Candidate data centers Real-hardware workload → Test validation Term and payment → Contract alignment Deployment acceptance → Monitoring handover
Four-step adoption,
from requirements to go-live
Every stage has defined deliverables and milestones,
so budgets and engineering schedules can be planned together.
Workload and spec alignment
We confirm model scale, data residency requirements, and budget range, then narrow the field to 1 to 2 candidate data centers.
Requirements reviewReal-hardware testing and site visit
We arrange a single-node, 8-GPU test or a data center site visit to validate throughput, networking, and operating procedures.
POCCommercial terms and payment design
We plan the term and payment structure around cash flow. Payment terms are assessed project by project.
ContractBuild, acceptance, and operations handover
Once deployment and acceptance are complete, 24×7 monitoring begins, and the warranty and SLA take effect at the same time.
Delivery and go-liveNEWSROOM
News & Events
DIGITIMES2026-09-10
KONST CEO Avery Tsai takes the stage at Boba Tech Summit with top speakers from YouTube, Twitch, and OpenAI
Read the article
Commercial Times2026-08-13
ADATA invests another NT$386 million in KONST, raising its stake to 12.37%
Read the article
Economic Daily News2026-08-10
KONST at Ai4: the real bottleneck in the AI compute crisis is fragmentation, not shortage
Read the articleOne KONST.
One AI Stack.
From Cloud to Application.