Prices in Asia-Pacific
| GPU | On demand, per GPU-hour | 24–36 month term | 8-GPU node per month (on demand rate) |
|---|---|---|---|
| H100 80GB | $3.23 | $2.78 | $18,863 |
| H200 141GB | $4.31 | $3.71 | $25,170 |
| RTX PRO 6000 96GB | $2.03 | $1.75 | — |
| B200 192GB | Reserved 8-GPU nodes, priced on request | ||
Strata list prices as of 2026-11-17, the same in every region. Node per month = price × 8 GPUs × 730 hours.
When a local region matters
- Latency for users. Chat and voice inference served from Singapore, Tokyo or Mumbai is faster for users in the region than from Europe or the US.
- Data residency. If data or model weights must stay in India, Japan or Singapore, the datacenter should be named in the reservation, not just "Asia".
- When it does not matter. Training and batch inference are not latency-sensitive; if no local capacity is free, a European or Middle East region at the same list price works just as well.
Which city to pick
- Singapore: a central hub for Southeast Asia and Australia.
- Tokyo: for users and data in Japan and North Asia.
- Mumbai: for users in India and data that must stay in the country.
Availability differs by city and GPU model. Name the country, GPU, quantity and term in the capacity request form; we confirm the datacenter and its tier for every reservation. Dedicated clusters are usually ready in a few days to two weeks.
Other options
Hyperscalers and GPU marketplaces also have Asian capacity. Marketplaces can be cheaper per hour but usually do not let you pick the exact city or guarantee that a host is not shared. Strata servers are dedicated, with the datacenter named in your reservation, reserved terms 3–14% below on demand and payment in stablecoins or by bank transfer. See also GPUs in the Middle East and GPUs in Europe.


