AI-Stack and ixCSP turn GPU clusters into governed, monetizable AI platforms. HeTone distributes INFINITIX across Europe and China, and pairs the software with the liquid-cooled facilities it builds.
INFINITIX, an NVIDIA partner headquartered in Taipei and Kaohsiung with subsidiaries across Japan, Korea, China, Hong Kong, and Singapore, holds the top AI-platform market share in Taiwan and built the software behind the island's national AI cloud. Its AI-Stack platform manages GPU resources end to end: pooling, partitioning, and aggregating NVIDIA, AMD, and NPU accelerators across nodes, with multi-tenant isolation, scheduling, and full-stack monitoring from bare metal to MLOps.
HeTone is INFINITIX's distribution partner for Europe and China, and the two companies build together: in Bangkok, HeTone leads construction of a 70 MW liquid-cooled AI edge data center while AI-Stack and ixCSP orchestrate and commercialize its GPU capacity. From compute-hub projects to enterprise clusters, HeTone delivers the facility and INFINITIX the software that runs it.
GPU resource and AI infrastructure management: pool, partition, and aggregate GPUs and NPUs across nodes (NVIDIA, AMD, Phison aiDAPTIV+), with multi-tenant isolation, storage management, and a visual monitoring dashboard.
GPU / NPU poolingElastic Distributed Training, Machine Learning Service, and inference-serving deployment with task management and scheduling, one platform from model development to production serving.
Train → ServeTurns GPU capacity into a business: GPU-, Token-, and Model-as-a-Service with billing management and a model marketplace, so operators can monetize compute without custom development.
GaaS · TaaS · MaaS
AI-Stack splits cleanly into a compute plane, the worker clusters that pool GPUs on a 100 to 400 Gb fabric, and a control plane that schedules and governs them. ixCSP adds a gateway plane on top, exposing the same pool to users and applications as cloud services secured by tokens and API keys, with tiered high-speed, object, and archiving storage behind it.
INFINITIX's stated purpose is GPU computing power on demand, as easy as turning on a faucet. One platform runs hybrid AI and HPC, mixes GPU specifications to match the job, for example H200 for model training and L40S for inference serving, and even crosses vendors between NVIDIA and AMD, so capacity follows the workload instead of the other way around.
ixCSP gives operators an Operation Center, catalog, users and tenants, resource configuration, dashboards, billing, and audit logs, while AI developers get a Dev Console with on-demand resources, one-click deployment, a GenAI studio, and team controls. Underneath sit three ready business models: GPU-, Token-, and Model-as-a-Service.
In April 2026, INFINITIX and HeTone announced a 70 MW AI edge data center in Bangkok: HeTone leads the high-density, liquid-cooled facility build, while AI-Stack schedules GPU resources across nodes into one virtual compute pool at over 90 percent utilization, commercialized through ixCSP for the fast-growing Southeast Asian market.
Platform layers and modules we deploy and support, contact us for scoping and licensing.
| Model | Type | Key Spec | Application |
|---|---|---|---|
| AI-Stack Enterprise | PaaS OS | MLOps, auto-aware scheduling, cluster engine | Enterprise AI platforms |
| AI-Stack ixInfra | IaaS OS | Cluster provisioning & monitoring, Ubuntu / RHEL | Bare-metal AI infrastructure |
| ixCSP | SaaS | AI marketplace, AI gateway, business operations | AI cloud service providers |
| ixAPI | API layer | Programmatic access to the full stack | Integration & automation |
| Security module | Platform-wide | Tenant isolation, LDAP / OAuth, ISO 27001 / 27701 / 42001 | Compliance & governance |
Send us your thermal requirements, we will recommend and validate the right material combination for your design.