# GMI Cloud reports $500 million in signed ARR as capacity trails demand > Source: > Published: 2026-07-22 03:00:18+00:00 [Alex Yeh (@alex_yehya)](https://x.com/alex_yehya)-led [GMI Cloud](https://www.gmicloud.ai/en) said in a [July 21st announcement](https://www.prnewswire.com/news-releases/gmi-cloud-secures-8x-arr-growth-as-global-ai-infrastructure-footprint-expands-302830473.html) that signed customer commitments pushed contracted annual recurring revenue above $500 million. GMI Cloud summarized the figures [on X](https://x.com/gmi_cloud/status/2079759577139269946), adding that live ARR grew 2.4x during the first half of 2026 from its year-end 2025 level. Yeh built GMI Cloud out of an earlier data center operation that served cryptocurrency miners. He studied computational materials science at Johns Hopkins University, where limited access forced him to borrow GPUs from another department, according to [an account published by investor Headline Asia](https://headline.com/asia/en-us/post/founder-journey-navigating-pivots-with-gmi-cloud). When demand for mining infrastructure shifted, Yeh moved the operation toward AI compute, putting his background in data centers and GPU procurement behind a cloud provider focused on model training and inference. ### Contracts ahead of revenue GMI Cloud defined live ARR as annualized revenue from GPU capacity already operating in production. Contracted ARR covers signed commitments, including multiple multi-year agreements, that are expected to convert as GMI Cloud brings additional power and computing capacity online. That distinction is central to the update. Contracted ARR is a bookings-style measure tied to future service delivery; it does not mean GMI Cloud is producing $500 million in annualized revenue today. GMI Cloud did not disclose the dollar value of live ARR, the length of the contracted backlog, customer concentration, or how much of the $500 million is scheduled to enter production during 2026. The disclosure makes capacity delivery the immediate test. GMI Cloud says contracted ARR exceeds eight times its year-end 2025 ARR, while live ARR increased 2.4x over the following six months. Turning the remaining commitments into revenue requires data center space, GPUs, networking and power to arrive on schedule. GMI Cloud has been financing that buildout with a mix of venture equity, debt and infrastructure partnerships. In October 2024, GMI Cloud [raised $15 million in equity and $67 million in debt](https://www.prnewswire.com/news-releases/gmi-cloud-raises-82-million-in-total-series-a-funding-to-drive-global-access-to-advanced-gpus-and-cloud-infrastructure-302289529.html), bringing total capital raised at the time to more than $93 million. Headline Asia led the round, with strategic participation from energy provider [Banpu](https://www.banpu.com/) and electronics manufacturer [Wistron](https://www.wistron.com/en). The financing structure reflected the economics Yeh chose: GPU clouds need large amounts of capital before customers can consume the capacity. GMI Cloud's latest figures show signed demand running ahead of installed infrastructure, leaving execution around power and hardware as the bridge between bookings and revenue. ### Inference volume jumps GMI Cloud also said its inference operation grew 34x over four weeks and is processing about 2.5 trillion tokens per week. The announcement did not provide a starting volume, revenue contribution, model mix or definition of the four-week measurement period, making the growth multiple difficult to compare with other inference providers. GMI Cloud attributed part of that expansion to OpenRouter, the model marketplace and routing service. GMI Cloud called itself a top-three token provider on the platform, serving customers through U.S. and Asia-Pacific endpoints. OpenRouter's public data gives a more qualified picture. As of July 22nd, its [provider directory](https://openrouter.ai/providers/) listed GMICloud at 172.3 billion daily tokens and 7.6 trillion monthly tokens, behind more than 10 providers by daily volume. The public table does not corroborate a general top-three provider rank. GMI Cloud's claim may cover a narrower model, region, endpoint category or measurement window, none of which the announcement identified. OpenRouter separately lists [13 models available through GMICloud](https://openrouter.ai/provider/gmicloud). The inference push expands GMI Cloud beyond renting dedicated GPU clusters. GMI Cloud's current platform pairs serverless model APIs with dedicated H100, H200 and future Blackwell capacity, allowing a customer to begin with metered inference and move into reserved infrastructure as usage grows. That product path gives GMI Cloud a way to capture smaller workloads before they justify full clusters. ### Capacity moves across the Pacific GMI Cloud's largest announced infrastructure commitments sit in Asia. In November 2025, GMI Cloud [announced a $500 million Taiwan AI factory](https://www.prnewswire.com/news-releases/gmi-cloud-to-launch-next-gen-ai-factory-in-taiwan-with-nvidia-to-power-the-future-of-ai-infrastructure-in-asia-302616532.html) designed around 7,000 Nvidia Blackwell Ultra GPUs across 96 high-density racks. GMI Cloud said the 16-megawatt facility could process close to 2 million tokens per second, with Wistron, Trend Micro, VAST Data and TECO attached to initial deployments and infrastructure work. In March 2026, GMI Cloud also [announced a proposed $12 billion sovereign AI infrastructure initiative in Japan](https://www.prnewswire.com/news-releases/gmi-cloud-unveils-12-billion-1gw-sovereign-ai-infrastructure-initiative-in-japan-302715607.html). The first phase is planned as a 200-megawatt deployment in 2028, within a development that GMI Cloud says could eventually reach one gigawatt. Those projects explain why GMI Cloud is emphasizing contracted ARR. Yeh has signed demand against infrastructure that is still being expanded. The $500 million figure establishes the size of that order book. GMI Cloud's live ARR, cash generation and delivery pace will determine how much of it becomes an operating cloud business.