GMI Cloud

GPU cloud selling dedicated NVIDIA GPU clusters, bare metal and serverless inference to AI companies. Headquartered in Mountain View with roots in Taiwan; building a Taiwan AI factory and a planned 1 GW site in Japan.

HQ Mountain View, United States Founded 2021 Verified

H100 list price
$2.60per GPU-hour source, 9 Nov 2026
B200 list price
$5.00per GPU-hour source, 9 Nov 2026
ClusterMAX
Silver source, 9 Nov 2026 Edition 3.0
Ownership
Private source, 9 Nov 2026

Visit GMI Cloud Compare with alternatives Request a quote

Facts

Source per value
Headquarters Headquarters city and country as stated by the company or its filings.
Mountain View, United States 7
Ownership Publicly listed, privately held, or a subsidiary of another company.
private 1
Ticker Exchange and ticker symbol if listed, e.g. NASDAQ: CRWV. Null if private or not stated on a fetched source.
Not published
Parent company Owner or group, when the provider is a subsidiary or has merged (e.g. Voltage Park under Lightning AI, Taiga Cloud under Quake AI / Rumble).
Not published
GPUs offered Data-centre accelerators (Blackwell, Hopper, Ada, Ampere and AMD Instinct) listed on the provider's own site, newest first as the provider lists them. The full list, including workstation and older cards, is under "All GPU models".
H100, H200, B200, GB200, GB300 1
All GPU models Every accelerator model the provider lists on its own site as of the verification date.
H100, H200, B200, GB200, GB300 1
H100 Lists NVIDIA H100 on its own site.
Yes 1
Blackwell Lists at least one NVIDIA Blackwell data-centre GPU (B100, B200, B300, GB200 or GB300).
Yes 1
AMD Instinct Lists at least one AMD Instinct GPU (MI300X, MI325X, MI350X, MI355X or newer).
No 1
H100 list price $/GPU-hr Published on-demand list price for one NVIDIA H100 (SXM unless pricing.note says otherwise), taken from the provider's own pricing page. Null if not published, quote-only, host-set (marketplaces) or not in USD.
$2.60 1
B200 list price $/GPU-hr Published on-demand list price for one NVIDIA B200. Same rules as the H100 price.
$5.00 1
Price checked Date the list prices were read from the provider's pricing page.
2026-11-09 1
Regions Short form of where the provider states it runs capacity. Full detail under "Regions (detail)".
N. America, Europe, APAC 1
Regions (detail) Countries or regions where the provider states it runs capacity.
North America, Europe, Asia-Pacific (company site; specific sites not named); Taiwan AI factory (announced Nov 2025); Kagoshima, Japan site planned (announced Mar 2026) 1
Stated capacity GPU count, MW/GW of power or backlog exactly as the company or its filings state it, with the as-of date. Never estimated.
30,000+ GPUs deployed (company site, Nov 2026); Taiwan AI factory: 7,000 Blackwell GPUs, 16 MW, $500M (announced Nov 2025); Japan: up to 1 GW, $12B, Vera Rubin NVL72, development from late 2026 (announced Mar 2026) 2
Public customers Customers named publicly by the provider or in reputable press.
Higgsfield, Mirelo AI, WiAdvance, HeyGen, Utopai, OXMIQ, Eigen AI, LegalSign, Trend Micro, Wistron, Reflection AI 1
Funding / debt (USD) $ Total funding if a source states a total; otherwise the latest sourced equity round or debt facility. See funding_note for what the number covers.
$668,000,000 7
Funding note What the funding/debt number is, when it was announced and by whom.
$668M package announced 30 Sep 2026: $223M Series B equity led by ARCHIV with NVIDIA participating, plus a $445M credit facility led by CTBC (company release via PR Newswire) 7
Interconnect Cluster network fabrics the provider lists (InfiniBand, RoCE, Ethernet, NVLink).
InfiniBand 1
Bare metal Offers bare-metal GPU servers (no hypervisor).
Yes 1
Kubernetes Offers a managed Kubernetes service on GPU nodes.
Yes 1
Slurm Offers managed Slurm (including Slurm-on-Kubernetes) for training clusters.
Yes 1
Reserved contracts Sells committed or reserved capacity (monthly to multi-year).
Yes 1
Self-serve Customers can sign up and launch GPU instances on demand without a sales contract.
Yes 1
ClusterMAX SemiAnalysis ClusterMAX tier for the provider. 'Participation' is the 3.0 tier for providers that took part without earning a medal. Null if not rated, listed as unavailable, or not confirmed.
Silver 6
ClusterMAX edition Edition of ClusterMAX the rating comes from (3.0 published September 2026).
3.0 6

Pricing detail

Pricing model
on-demand + reserved
From
$2.60 per GPU-hour
What the prices cover
H100 from $2.60/GPU-hr (SXM/PCIe not stated). H200 from $3.20, B200 from $5.00 (limited availability), GB200 from $8.00; GB300 pre-order. Reserved pricing via sales. Page shows '$' without a currency code.

Sources

7
  1. GMI Cloud pricing accessed 2026-11-09
  2. GMI Cloud homepage accessed 2026-11-09
  3. GMI Cloud about accessed 2026-11-09
  4. GMI Cloud brings Taiwan's first AI factory accessed 2026-11-09
  5. GMI Cloud announces 1GW sovereign AI infrastructure in Japan accessed 2026-11-09
  6. SemiAnalysis ClusterMAX 3.0 accessed 2026-11-09
  7. GMI Cloud raises $668 million (Pulse 2.0, Sep 2026) accessed 2026-11-09

GMI Cloud alternatives

H100 list price

Get GPU quotes from GMI Cloud and similar clouds

Tell us what you need once. We send it to matching GPU clouds and they reply to you directly.

Quote requests open shortly. To get in first, email hello@neocloudcompanies.com.