🌏 中文版
Deal Terms
| Item | Value |
|---|---|
| Company | GMI Cloud (Mountain View, California, US) |
| Round | Series B (plus a separate credit facility) |
| Amount | $223M equity (Series B) + $445M credit facility, $668M total |
| Lead | ARCHIV (newly founded investment firm focused on AI and robotics) |
| Participants | NVIDIA, DSC Investment, Trend Micro, KB Investment, Kyobo Life, KT Corporation, and other Asia-Pacific investors; credit facility arranged by Taiwan's CTBC Bank |
| Valuation | Undisclosed |
| Total raised | Undisclosed overall (this round alone totals $668M) |
| Founded | 2021 |
| Headcount | Undisclosed |
What the Company Does
GMI Cloud is an AI-native cloud provider — it delivers high-performance GPU infrastructure and inference services so AI teams needing heavy compute can rent capacity instead of building their own data centers.
Its platform spans the US and Asia-Pacific, offering on-demand access to GPUs ranging from Nvidia's H100 and H200 up through the newer Vera Rubin generation, layered with bare-metal cluster management and inference optimization on top of raw compute. GMI Cloud positions itself as one of the few GPU clouds built to serve both sides of that demand at once: US AI companies and hyperscalers need capacity in both the US and Asia-Pacific, while Asia-Pacific enterprises want data and compute kept local to satisfy regulatory requirements — and most competitors have only built out one side. GMI Cloud leans on deep supply-chain ties in Taiwan, where most of the world's AI servers are manufactured, to deliver more predictable timelines than rivals. It has already launched its "Taiwan AI Factory" and is backing a sovereign AI initiative in Japan.
Contracted ARR has now topped $600M, up more than 9x from year-end 2025, while live ARR in production has grown more than 4.5x over the same period. Its inference platform processes roughly 4 trillion tokens a week. Notable customers include Fireworks, Higgsfield, Nous Research, OpenRouter, Reflection, Cartesia, Trend Micro, and Utopai Studios.
What This Round Signals
What It Means for the Agent Ecosystem
What stands out most isn't the dollar amount — it's how the money is split: more capacity, more inference services, more hiring all at once, which signals the company sees the bottleneck as the entire compute-to-inference-to-delivery chain, not any single link. As agent applications turn one-shot chat requests into multi-step tasks that run for longer stretches, the shape of inference load itself is changing — GPU cloud providers now have to handle sustained background computation, not just traffic spikes.
What Investors Are Betting On
Lead investor ARCHIV is a newly founded firm focused on AI and robotics, and Nvidia itself participated directly in the round — a chipmaker taking equity in a downstream cloud reseller is a strategic move to secure a steady outlet for its newest silicon (GB200, GB300 NVL72), not just a financial bet. Participation from several Asia-Pacific strategic investors (KT, Kyobo, Trend Micro, among others) also effectively vouches for GMI Cloud's customer base and supply-chain relationships in the region.
Numbers Worth Watching
- Contracted ARR up more than 9x versus live ARR up more than 4.5x — that gap is itself worth tracking: whether contract conversion is keeping pace with sales velocity
- A $223M equity-to-$445M debt ratio shows a deliberate choice to fund hardware expansion with debt rather than equity dilution, a financing pattern that tends to show up once capital-intensive GPU cloud businesses mature
- Roughly 4 trillion tokens processed per week is one of the few cases where an AI infrastructure startup leads its external messaging with usage scale rather than the size of its raise
Watchlist Status
GMI Cloud is not currently on the watchlist. Recommend adding it under section A3 (Inference Infrastructure), alongside CoreWeave, Lambda Labs, Nebius, and RunPod — tracking whether its "US plus Asia-Pacific, both sides at once" positioning translates into a real pricing or delivery-time edge over US-centric rivals like CoreWeave.
Today's Takeaway
GPU cloud has mostly been a story about who gets the newest chips first. GMI Cloud's round reframes the differentiator as supply-chain geography — deep ties in Taiwan let it treat a delivery date as a promise rather than a guess, on both sides of the Pacific. As raw compute itself gets harder to out-capitalize, delivery certainty turns into a moat investors are willing to price.
References
Loading...