On demand
Pay-as-you-go GPU instances and inference endpoints for variable or exploratory workloads.
- Hourly billing
- No minimum term
- Self-service
Full-stack compute, one partner.
From the power and cooling plant to the scheduler and the API, we own the stack end-to-end — so capacity arrives faster, runs cooler and costs less.
AI Data Centres
Purpose-built facilities for the power and heat of modern accelerators. We develop, own and operate halls engineered for liquid-cooled racks — and deliver them as dedicated suites, build-to-suit facilities or colocation.
High-density halls
White space planned for liquid-cooled racks of up to 130 kW, with structural loading and power distribution to match.
Build-to-suit
Dedicated halls or whole facilities delivered against your specification and your timeline.
Colocation for AI
Cages and private suites for customer-owned clusters, with liquid-cooling connections at the rack.
Carrier-neutral interconnection
Meet-me rooms and diverse fibre paths to carriers, clouds and internet exchanges.
Best forHyperscalers, AI labs and enterprises deploying their own hardware.
See how we engineer facilitiesGPU Compute Platform
GPUs as a service rather than a construction project. Reserve a dedicated cluster for years or rent capacity by the hour — on bare metal, managed Kubernetes or Slurm, with production-grade inference on top.
Reserved clusters
Dedicated capacity on multi-year terms, sized from a handful of nodes to thousands of GPUs.
On-demand capacity
Hourly access for experiments, bursts and evaluation — with no long-term commitment.
Managed orchestration
Kubernetes and Slurm operated for you, with queues, quotas and node health checks built in.
Inference endpoints
Autoscaling model serving behind low-latency APIs, billed on what you use.
Best forAI-native companies, enterprise AI teams and research groups.
Explore the platformGridPort
Compute distribution platformgridport.ai
GridPort is our compute distribution platform. It turns GPU capacity into something any developer can use — open-model inference through a single OpenAI-compatible API, billed by the token.
Your applicationOpenAI-compatible SDK
GridPortRoute · Meter · Fail over
Model capacityOpen models, served at scale
One API, drop-in
OpenAI-compatible. Migrate by changing three settings: base URL, API key and model ID.
Every request traceable
Latency, tokens, cost and each routing attempt are recorded request by request.
Built-in resilience
Failing upstream routes are switched automatically, before the first byte reaches you.
Budgets & controls
Model and IP allowlists, budgets per key and per project, and balance alerts.
Best forDevelopers and enterprise teams who want open-model inference without running GPUs.
Visit gridport.aiManaged Operations
A GPU cluster is only as valuable as its utilisation. Our operations team runs customer and partner clusters to a production standard — in our facilities or yours.
24/7 monitoring
A network operations centre watching power, cooling, fabric and node health around the clock.
Lifecycle management
Firmware, drivers, break-fix and hardware refresh handled end to end.
Performance engineering
Burn-in, benchmarking and tuning, so clusters perform as specified from day one.
SLA-backed support
Named engineers, defined response times and transparent reporting.
Best forOwners of GPU clusters who need them run to a production standard.
Talk to operationsAdvisory & Partnerships
AI infrastructure is capital-intensive and long-dated. We work with enterprises, institutions, landowners and investors to plan, structure and deliver capacity that performs over its whole life.
Capacity strategy
Demand modelling, technology roadmaps and total-cost analysis.
Site & power
Site selection, grid connection and permitting support across the region.
Structuring & finance
Project finance, co-investment and offtake structures for long-horizon assets.
Joint development
Delivery partnerships with landowners, utilities and technology providers.
Best forInvestors, landowners and institutions entering AI infrastructure.
Discuss a partnership(06)Engagement models
Three ways to
work with us.
Start with a single node or commit to a campus. Models can be combined, and capacity can move between them as your needs change.
Reserved capacity
Dedicated clusters or halls on multi-year terms — predictable cost and guaranteed availability.
- Dedicated hardware
- Committed pricing
- Named support team
Build-to-suit
Facilities and clusters designed, built and operated around a single customer’s requirements.
- Custom specification
- Phased delivery
- Long-term partnership
(07)How we deliver
From megawatts
to models.
One accountable team across the full lifecycle — fewer hand-offs, faster time-to-compute.
01
Plan
Site, power and demand modelling; capacity roadmaps aligned to your AI strategy.
02
Build
Design and construction of high-density halls, power and liquid-cooling systems.
03
Deploy
Cluster integration, network fabric, burn-in testing and performance validation.
04
Operate
24/7 operations, continuous optimisation and seamless scaling as demand grows.
(08)Who we serve
Built for the
builders of AI.
From frontier labs to regulated enterprises, our customers share one need: reliable compute, delivered at scale.
Foundation model training
Large, tightly coupled GPU clusters for pre-training and fine-tuning at scale.
Enterprise AI & inference
Low-latency, cost-efficient inference for copilots, agents and AI-native products.
Financial services
Secure, low-latency compute for quantitative research, risk and fraud analytics.
Life sciences
Accelerated workloads for genomics, protein modelling and drug discovery.
Smart city & robotics
Simulation, digital twins and embodied-AI training.
Research & academia
Flexible capacity for universities and research institutes across the region.
(09)Questions
Good to
know.
Can’t find what you’re looking for? Our team is happy to go into detail.
Contact usCan we bring our own hardware?
Yes. Our colocation and build-to-suit offerings are designed for customer-owned clusters, including liquid-cooled systems. We can also procure and integrate hardware on your behalf.
Is there a minimum commitment?
On-demand capacity has no minimum term. Reserved clusters and dedicated halls are contracted on multi-year terms, sized to your roadmap.
Can our data stay in Hong Kong?
Yes. Data-residency options keep workloads and data in Hong Kong, with private tenancy and dedicated networks available for regulated industries.
Which accelerators do you support?
Our facilities are engineered for current and next-generation liquid-cooled accelerator platforms. Specific hardware is agreed deployment by deployment.
Do you only serve customers in Hong Kong?
No. We are based in Hong Kong and serve customers across the Greater Bay Area, Asia-Pacific and beyond.
How do we get started?
Tell us about your workload, timeline and capacity needs. We will come back with options, indicative pricing and next steps.