Made on
Tilda
Bare-metal H800 clusters, OpenAI-compatible inference, and managed fine-tuning — operated from Hong Kong.
Dedicated NVIDIA H800 servers and multi-node clusters with InfiniBand interconnect. Full root access, persistent storage, custom OS images.
OpenAI-compatible endpoint for open-weight language models. Drop-in replacement for existing applications. Token-based billing, no minimum commitment.
Pre-configured training infrastructure for language and diffusion models up to 70B parameters. Bring your dataset, receive a trained model.
Aggregated access to leading generative video, voice cloning, and dubbing services through a unified API. Single contract, consolidated billing in USD.
Tier-3 datacenter operations at HKSTP. Robust power and cooling, multi-carrier connectivity, physical access controls.
80GB HBM3 memory per GPU, 8-GPU nodes with NVLink, multi-node clusters connected via 400 Gb/s NDR InfiniBand fabric.
Inference endpoints speak the OpenAI protocol. Switch existing applications by changing one environment variable. Streaming, function calling, embeddings supported.
Customer workloads run in isolated tenant environments. We do not train on customer data, log prompts, or retain model outputs. Operated under HK PDPO.
Volume discounts and dedicated reservations available — contact us for committed-use agreements.
> Sustained pricing applies after 720 GPU-hours/month of committed usage.
For inquiries about pricing, technical specifications, capacity, or onboarding — please reach out by email. We typically respond within one business day.