18 listings on machines physically in Taiwan, across 5 GPUs. Converted at $31.86/$ (2026-09-18). The cheapest H100 hosted in Taiwan is $99.92/hr at Vast.ai.
Prices collected 2026-09-18 · every figure is a provider's own published rate
Listings whose host reports a location inside Taiwan — marketplace machines and hyperscaler regions. This is the section that matters for latency and data residency; the provider may be headquartered anywhere.
| GPU | Cheapest at | Where | $ / GPU / hr | Local median | Listings | Global |
|---|---|---|---|---|---|---|
| Taiwan, TW | $99.92 | $99.97 | 2 | $111.36 | ||
| Taiwan, TW | $48.88 | $48.88 | 1 | $65.16 | ||
| Taiwan, TW | $25.51 | $25.51 | 3 | $46.84 | ||
| Taiwan, TW | $17.03 | $17.06 | 3 | $21.99 | ||
| Taiwan, TW | $14.46 | $14.94 | 9 | $19.12 |
The whole index — 32 GPUs from every provider we track, each of which will rent to a customer in Taiwan — with the global median and the single cheapest listing converted to TWD at the snapshot's rate. Tax is not included.
On a machine physically located in Taiwan: $99.92 per GPU-hour at Vast.ai (Taiwan, TW), 2 such listings today. Across every provider we track, the global median is $3.50 ($111.36) and the cheapest listing anywhere is $1.68 at Latitude.sh — any of them will sell to a customer in Taiwan.
None of the 48 providers we track is headquartered in Taiwan. But Vast.ai lists machines physically located there — 18 listings today. The nearest by headquarters are on the same continent: GPU.ai, E2E Networks, Jarvislabs, Cyfuture AI, HyperAI. Every provider in the index sells worldwide; the question is latency, data residency and invoicing, not availability.
Providers publish in dollars (a few in euros). This page converts at the rate captured with the snapshot — $31.86 per dollar on 2026-09-18 — so every GPU sits on one axis in TWD. Your card will be charged in the provider's billing currency at your bank's rate, and local tax is added on top, so treat the TWD figure as a comparison, not an invoice.
For holding a model, memory per TWD: the NVIDIA RTX 3090 gives 96 GB per $1 (at $7.97/hr). The cheapest GPU-hour of any kind is the NVIDIA RTX A4000 at $5.42. The full table above prices every GPU we track in TWD; the cheapest-listing column links to who sells it.
Nearly. Most providers publish one price list regardless of where you are, so the market table on this page is the same market a buyer in any country sees — the differences are tax, currency and, for the hyperscalers, region. What does change by country is who has hardware nearby: the hosted-hardware section lists machines actually located in Taiwan, where any exist, which is what matters for latency and data residency.