Dedicated bare-metal GPUs with predictable performance, racked in Dhaka.
Hosted in Bangladesh
Data stays in-country
Bangladesh's own cloud for compute, inference, and agents.
GPU clusters, serverless model APIs and an agent runtime on one cloud. Hosted in Bangladesh, billed in taka, supported from Dhaka in Bangla and English.
Llama 4Qwen 3DeepSeek V3MistralGemma 3WhisperBanglaBERTStable DiffusionPhi-4BGE-M3
Start with an API.
Scale to a cluster.
Call open models over one API from day one, then move the same workload onto dedicated GPUs in Dhaka when traffic grows.
When an API isn't enough, take the whole machine.
Bare-metal NVIDIA GPUs in data centres inside Bangladesh, built on validated reference designs.
Our cluster engine orchestrates multi-node jobs at the infrastructure layer.
Root access and custom stacks when the infrastructure is the product.
GPU pricing, in taka
From ৳299 per GPU-hour, billed monthly in BDT. Pay by bank transfer, LC, bKash or Nagad. No dollar card needed.
NVIDIA RTX PRO 6000 Blackwell Server Edition · 96 GB per GPU
| Plan | GPUs | Total VRAM | vCPU | RAM | NVMe | Per hour | 24 hours | 30 days, full use | Status |
|---|---|---|---|---|---|---|---|---|---|
| StarterInference, prototyping, small fine-tunes.৳ 299 /hour | 1 | 96 GB | 12 | 24 GB | 200 GB | ৳ 299 | ৳ 7,176 | ৳ 215,280 | Available now |
| ProfessionalFine-tuning mid-size models, multi-GPU experiments.৳ 598 /hour | 2 | 192 GB | 24 | 48 GB | 400 GB | ৳ 598 | ৳ 14,352 | ৳ 430,560 | Available now |
| BusinessTraining and large-context inference.৳ 1,196 /hour | 4 | 384 GB | 48 | 96 GB | 800 GB | ৳ 1,196 | ৳ 28,704 | ৳ 861,120 | Available now |
| EnterpriseFull-node workloads needing maximum headroom.৳ 2,392 /hour | 8 | 768 GB | 96 | 192 GB | 1,600 GB | ৳ 2,392 | ৳ 57,408 | ৳ 1,722,240 | Available now |
| Custom1 to 8 GPUs, sized to your workload.৳ 299 × N /hour | 1–8 | 96 GB × N | 12 × N | 24 GB × N | 200 GB × N | ৳ 299 × N | ৳ 7,176 × N | ৳ 215,280 × N | Quote |
- VRAM is the total across GPUs, not unified memory.
- Plans are instance sizes drawn from one shared pool of 8 GPUs; availability is confirmed in your quote.
- Ubuntu 24.04 LTS with SSH access. CPU, RAM and SSD are proposed allocations, not SLA guarantees.
Indicative launch prices. Final rates are confirmed in your quote.
Offer
Book your tensor slot
Reserve dedicated GPU time in Dhaka. Pick a plan (Starter, Professional, Business, Enterprise or Custom), and tell us how many hours and when you want to start. We confirm the slot and the final rate in a taka quote. Pricing starts at ৳299 per GPU-hour.
- Starter 1 GPUAvailable now
- Professional 2 GPUsAvailable now
- Business 4 GPUsAvailable now
- Enterprise 8 GPUsAvailable now
- Custom 1–8 GPUsRequest quote
Production AI runs better at home.
Requests from Dhaka, Chattogram and Sylhet stay on Bangladeshi networks end to end.
- <5 ms
- Round trip inside Dhaka, against ~60 ms to Singapore
- 100%
- Of your data, logs and backups stay in Bangladesh
- ৳ only
- One taka invoice. No dollar card, no FX surprises
- 24/7
- Engineers in Dhaka who answer in Bangla and English
Latency measured from a Dhaka ISP to our edge. Indicative.
Inference-first by design
Inference is serverless by default. Scaling, traffic and cost control happen on their own, all the way down to zero.
Serverless by defaultAutomatic scaling, request batching and cost-aware scheduling on every endpoint.
Performance at scaleDedicated clusters with RDMA networking hold throughput steady under load.
Flexible by designMove from a shared API to your own cluster without rewriting your stack.
One API for open models, and for Bangla.
OpenAI-compatible endpoints for text, speech, vision and embeddings, all hosted in Bangladesh.
LL
Llama 4 Maverick
LLM
QWBangla-ready
Qwen 3 235B
LLM
DS
DeepSeek V3
LLM
WHBangla-ready
Whisper Large v3
Audio
কথাBangla-ready
Bangla TTS
Audio
অক্ষরBangla-ready
Bangla OCR
Vision
SD
Stable Diffusion 3.5
Image
BGBangla-ready
BGE-M3 embeddings
Embed
Built for how Bangladesh works
Banks and MFS
Keep customer data on Bangladeshi soil while you add AI to onboarding, fraud checks and support.
- Private endpoints inside your VPC
- Audit logs you can hand to your regulator
- Bangla speech for call-centre automation
- Committed capacity with on-demand burst
Telcos and ISPs
Serve millions of Bangla conversations at a cost that works per subscriber.
- Batching tuned for chat and IVR traffic
- Peering with local ISPs for low latency
- Usage reporting per product line
- Capacity that grows with campaigns
Factories and exporters
Put vision models on the line and agents on the paperwork, without a dollar card.
- Defect detection on dedicated GPUs
- Document agents for LCs and shipping
- Taka billing with VAT invoices
- Support during Dhaka working hours and after
Startup credits
Building in Bangladesh? Claim GPU credits. Teams that receive credits also get a weekly office-hours slot with our engineers.
StartupsUp to 150 GPU-hours (≈ ৳ 45,000)
Universities and labsUp to 100 GPU-hours per project
HackathonsEvent GPU pools, e.g. 8 GPUs for 24–48 hours
Credits apply to Starter and Professional plans (RTX PRO 6000) at ৳299 per GPU-hour, valid 90 days, non-transferable and subject to GPU availability. One pool per hackathon. The final amount is confirmed in your quote.
FAQ
Quick answers. For anything else, our Dhaka team replies within one working day.
In data centres inside Bangladesh. Prompts, outputs, logs and backups stay in the country unless you move them yourself. You choose the site for dedicated clusters.
Yes. Every invoice is in BDT with VAT. Pay by bank transfer or LC for committed capacity, or top up serverless credit with bKash or Nagad.
Qwen 3, Llama 4 and the BGE-M3 embedding model handle Bangla well, and Whisper, Bangla TTS and Bangla OCR cover speech and documents. Every Bangla-ready model is marked in the library.
Start on serverless endpoints with a free trial key. When traffic is steady, reserve dedicated GPUs and point the same API calls at your private endpoint. Our Dhaka engineers help size the move.
Yes. Change the base URL and key, and most SDKs, agents and frameworks work as they are, including streaming and tool calls.
We run private endpoints, single-tenant clusters and full audit logs, and we work with your compliance team on the controls your regulator asks for.